xref: /openbmc/linux/arch/mips/kvm/vz.c (revision f4474d50)
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 
33 #include "trace.h"
34 
35 /* Pointers to last VCPU loaded on each physical CPU */
36 static struct kvm_vcpu *last_vcpu[NR_CPUS];
37 /* Pointers to last VCPU executed on each physical CPU */
38 static struct kvm_vcpu *last_exec_vcpu[NR_CPUS];
39 
40 /*
41  * Number of guest VTLB entries to use, so we can catch inconsistency between
42  * CPUs.
43  */
44 static unsigned int kvm_vz_guest_vtlb_size;
45 
46 static inline long kvm_vz_read_gc0_ebase(void)
47 {
48 	if (sizeof(long) == 8 && cpu_has_ebase_wg)
49 		return read_gc0_ebase_64();
50 	else
51 		return read_gc0_ebase();
52 }
53 
54 static inline void kvm_vz_write_gc0_ebase(long v)
55 {
56 	/*
57 	 * First write with WG=1 to write upper bits, then write again in case
58 	 * WG should be left at 0.
59 	 * write_gc0_ebase_64() is no longer UNDEFINED since R6.
60 	 */
61 	if (sizeof(long) == 8 &&
62 	    (cpu_has_mips64r6 || cpu_has_ebase_wg)) {
63 		write_gc0_ebase_64(v | MIPS_EBASE_WG);
64 		write_gc0_ebase_64(v);
65 	} else {
66 		write_gc0_ebase(v | MIPS_EBASE_WG);
67 		write_gc0_ebase(v);
68 	}
69 }
70 
71 /*
72  * These Config bits may be writable by the guest:
73  * Config:	[K23, KU] (!TLB), K0
74  * Config1:	(none)
75  * Config2:	[TU, SU] (impl)
76  * Config3:	ISAOnExc
77  * Config4:	FTLBPageSize
78  * Config5:	K, CV, MSAEn, UFE, FRE, SBRI, UFR
79  */
80 
81 static inline unsigned int kvm_vz_config_guest_wrmask(struct kvm_vcpu *vcpu)
82 {
83 	return CONF_CM_CMASK;
84 }
85 
86 static inline unsigned int kvm_vz_config1_guest_wrmask(struct kvm_vcpu *vcpu)
87 {
88 	return 0;
89 }
90 
91 static inline unsigned int kvm_vz_config2_guest_wrmask(struct kvm_vcpu *vcpu)
92 {
93 	return 0;
94 }
95 
96 static inline unsigned int kvm_vz_config3_guest_wrmask(struct kvm_vcpu *vcpu)
97 {
98 	return MIPS_CONF3_ISA_OE;
99 }
100 
101 static inline unsigned int kvm_vz_config4_guest_wrmask(struct kvm_vcpu *vcpu)
102 {
103 	/* no need to be exact */
104 	return MIPS_CONF4_VFTLBPAGESIZE;
105 }
106 
107 static inline unsigned int kvm_vz_config5_guest_wrmask(struct kvm_vcpu *vcpu)
108 {
109 	unsigned int mask = MIPS_CONF5_K | MIPS_CONF5_CV | MIPS_CONF5_SBRI;
110 
111 	/* Permit MSAEn changes if MSA supported and enabled */
112 	if (kvm_mips_guest_has_msa(&vcpu->arch))
113 		mask |= MIPS_CONF5_MSAEN;
114 
115 	/*
116 	 * Permit guest FPU mode changes if FPU is enabled and the relevant
117 	 * feature exists according to FIR register.
118 	 */
119 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
120 		if (cpu_has_ufr)
121 			mask |= MIPS_CONF5_UFR;
122 		if (cpu_has_fre)
123 			mask |= MIPS_CONF5_FRE | MIPS_CONF5_UFE;
124 	}
125 
126 	return mask;
127 }
128 
129 /*
130  * VZ optionally allows these additional Config bits to be written by root:
131  * Config:	M, [MT]
132  * Config1:	M, [MMUSize-1, C2, MD, PC, WR, CA], FP
133  * Config2:	M
134  * Config3:	M, MSAP, [BPG], ULRI, [DSP2P, DSPP], CTXTC, [ITL, LPA, VEIC,
135  *		VInt, SP, CDMM, MT, SM, TL]
136  * Config4:	M, [VTLBSizeExt, MMUSizeExt]
137  * Config5:	MRP
138  */
139 
140 static inline unsigned int kvm_vz_config_user_wrmask(struct kvm_vcpu *vcpu)
141 {
142 	return kvm_vz_config_guest_wrmask(vcpu) | MIPS_CONF_M;
143 }
144 
145 static inline unsigned int kvm_vz_config1_user_wrmask(struct kvm_vcpu *vcpu)
146 {
147 	unsigned int mask = kvm_vz_config1_guest_wrmask(vcpu) | MIPS_CONF_M;
148 
149 	/* Permit FPU to be present if FPU is supported */
150 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch))
151 		mask |= MIPS_CONF1_FP;
152 
153 	return mask;
154 }
155 
156 static inline unsigned int kvm_vz_config2_user_wrmask(struct kvm_vcpu *vcpu)
157 {
158 	return kvm_vz_config2_guest_wrmask(vcpu) | MIPS_CONF_M;
159 }
160 
161 static inline unsigned int kvm_vz_config3_user_wrmask(struct kvm_vcpu *vcpu)
162 {
163 	unsigned int mask = kvm_vz_config3_guest_wrmask(vcpu) | MIPS_CONF_M |
164 		MIPS_CONF3_ULRI | MIPS_CONF3_CTXTC;
165 
166 	/* Permit MSA to be present if MSA is supported */
167 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
168 		mask |= MIPS_CONF3_MSA;
169 
170 	return mask;
171 }
172 
173 static inline unsigned int kvm_vz_config4_user_wrmask(struct kvm_vcpu *vcpu)
174 {
175 	return kvm_vz_config4_guest_wrmask(vcpu) | MIPS_CONF_M;
176 }
177 
178 static inline unsigned int kvm_vz_config5_user_wrmask(struct kvm_vcpu *vcpu)
179 {
180 	return kvm_vz_config5_guest_wrmask(vcpu) | MIPS_CONF5_MRP;
181 }
182 
183 static gpa_t kvm_vz_gva_to_gpa_cb(gva_t gva)
184 {
185 	/* VZ guest has already converted gva to gpa */
186 	return gva;
187 }
188 
189 static void kvm_vz_queue_irq(struct kvm_vcpu *vcpu, unsigned int priority)
190 {
191 	set_bit(priority, &vcpu->arch.pending_exceptions);
192 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
193 }
194 
195 static void kvm_vz_dequeue_irq(struct kvm_vcpu *vcpu, unsigned int priority)
196 {
197 	clear_bit(priority, &vcpu->arch.pending_exceptions);
198 	set_bit(priority, &vcpu->arch.pending_exceptions_clr);
199 }
200 
201 static void kvm_vz_queue_timer_int_cb(struct kvm_vcpu *vcpu)
202 {
203 	/*
204 	 * timer expiry is asynchronous to vcpu execution therefore defer guest
205 	 * cp0 accesses
206 	 */
207 	kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
208 }
209 
210 static void kvm_vz_dequeue_timer_int_cb(struct kvm_vcpu *vcpu)
211 {
212 	/*
213 	 * timer expiry is asynchronous to vcpu execution therefore defer guest
214 	 * cp0 accesses
215 	 */
216 	kvm_vz_dequeue_irq(vcpu, MIPS_EXC_INT_TIMER);
217 }
218 
219 static void kvm_vz_queue_io_int_cb(struct kvm_vcpu *vcpu,
220 				   struct kvm_mips_interrupt *irq)
221 {
222 	int intr = (int)irq->irq;
223 
224 	/*
225 	 * interrupts are asynchronous to vcpu execution therefore defer guest
226 	 * cp0 accesses
227 	 */
228 	switch (intr) {
229 	case 2:
230 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_IO);
231 		break;
232 
233 	case 3:
234 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_IPI_1);
235 		break;
236 
237 	case 4:
238 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_IPI_2);
239 		break;
240 
241 	default:
242 		break;
243 	}
244 
245 }
246 
247 static void kvm_vz_dequeue_io_int_cb(struct kvm_vcpu *vcpu,
248 				     struct kvm_mips_interrupt *irq)
249 {
250 	int intr = (int)irq->irq;
251 
252 	/*
253 	 * interrupts are asynchronous to vcpu execution therefore defer guest
254 	 * cp0 accesses
255 	 */
256 	switch (intr) {
257 	case -2:
258 		kvm_vz_dequeue_irq(vcpu, MIPS_EXC_INT_IO);
259 		break;
260 
261 	case -3:
262 		kvm_vz_dequeue_irq(vcpu, MIPS_EXC_INT_IPI_1);
263 		break;
264 
265 	case -4:
266 		kvm_vz_dequeue_irq(vcpu, MIPS_EXC_INT_IPI_2);
267 		break;
268 
269 	default:
270 		break;
271 	}
272 
273 }
274 
275 static u32 kvm_vz_priority_to_irq[MIPS_EXC_MAX] = {
276 	[MIPS_EXC_INT_TIMER] = C_IRQ5,
277 	[MIPS_EXC_INT_IO]    = C_IRQ0,
278 	[MIPS_EXC_INT_IPI_1] = C_IRQ1,
279 	[MIPS_EXC_INT_IPI_2] = C_IRQ2,
280 };
281 
282 static int kvm_vz_irq_deliver_cb(struct kvm_vcpu *vcpu, unsigned int priority,
283 				 u32 cause)
284 {
285 	u32 irq = (priority < MIPS_EXC_MAX) ?
286 		kvm_vz_priority_to_irq[priority] : 0;
287 
288 	switch (priority) {
289 	case MIPS_EXC_INT_TIMER:
290 		set_gc0_cause(C_TI);
291 		break;
292 
293 	case MIPS_EXC_INT_IO:
294 	case MIPS_EXC_INT_IPI_1:
295 	case MIPS_EXC_INT_IPI_2:
296 		if (cpu_has_guestctl2)
297 			set_c0_guestctl2(irq);
298 		else
299 			set_gc0_cause(irq);
300 		break;
301 
302 	default:
303 		break;
304 	}
305 
306 	clear_bit(priority, &vcpu->arch.pending_exceptions);
307 	return 1;
308 }
309 
310 static int kvm_vz_irq_clear_cb(struct kvm_vcpu *vcpu, unsigned int priority,
311 			       u32 cause)
312 {
313 	u32 irq = (priority < MIPS_EXC_MAX) ?
314 		kvm_vz_priority_to_irq[priority] : 0;
315 
316 	switch (priority) {
317 	case MIPS_EXC_INT_TIMER:
318 		/*
319 		 * Call to kvm_write_c0_guest_compare() clears Cause.TI in
320 		 * kvm_mips_emulate_CP0(). Explicitly clear irq associated with
321 		 * Cause.IP[IPTI] if GuestCtl2 virtual interrupt register not
322 		 * supported or if not using GuestCtl2 Hardware Clear.
323 		 */
324 		if (cpu_has_guestctl2) {
325 			if (!(read_c0_guestctl2() & (irq << 14)))
326 				clear_c0_guestctl2(irq);
327 		} else {
328 			clear_gc0_cause(irq);
329 		}
330 		break;
331 
332 	case MIPS_EXC_INT_IO:
333 	case MIPS_EXC_INT_IPI_1:
334 	case MIPS_EXC_INT_IPI_2:
335 		/* Clear GuestCtl2.VIP irq if not using Hardware Clear */
336 		if (cpu_has_guestctl2) {
337 			if (!(read_c0_guestctl2() & (irq << 14)))
338 				clear_c0_guestctl2(irq);
339 		} else {
340 			clear_gc0_cause(irq);
341 		}
342 		break;
343 
344 	default:
345 		break;
346 	}
347 
348 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
349 	return 1;
350 }
351 
352 /*
353  * VZ guest timer handling.
354  */
355 
356 /**
357  * kvm_vz_should_use_htimer() - Find whether to use the VZ hard guest timer.
358  * @vcpu:	Virtual CPU.
359  *
360  * Returns:	true if the VZ GTOffset & real guest CP0_Count should be used
361  *		instead of software emulation of guest timer.
362  *		false otherwise.
363  */
364 static bool kvm_vz_should_use_htimer(struct kvm_vcpu *vcpu)
365 {
366 	if (kvm_mips_count_disabled(vcpu))
367 		return false;
368 
369 	/* Chosen frequency must match real frequency */
370 	if (mips_hpt_frequency != vcpu->arch.count_hz)
371 		return false;
372 
373 	/* We don't support a CP0_GTOffset with fewer bits than CP0_Count */
374 	if (current_cpu_data.gtoffset_mask != 0xffffffff)
375 		return false;
376 
377 	return true;
378 }
379 
380 /**
381  * _kvm_vz_restore_stimer() - Restore soft timer state.
382  * @vcpu:	Virtual CPU.
383  * @compare:	CP0_Compare register value, restored by caller.
384  * @cause:	CP0_Cause register to restore.
385  *
386  * Restore VZ state relating to the soft timer. The hard timer can be enabled
387  * later.
388  */
389 static void _kvm_vz_restore_stimer(struct kvm_vcpu *vcpu, u32 compare,
390 				   u32 cause)
391 {
392 	/*
393 	 * Avoid spurious counter interrupts by setting Guest CP0_Count to just
394 	 * after Guest CP0_Compare.
395 	 */
396 	write_c0_gtoffset(compare - read_c0_count());
397 
398 	back_to_back_c0_hazard();
399 	write_gc0_cause(cause);
400 }
401 
402 /**
403  * _kvm_vz_restore_htimer() - Restore hard timer state.
404  * @vcpu:	Virtual CPU.
405  * @compare:	CP0_Compare register value, restored by caller.
406  * @cause:	CP0_Cause register to restore.
407  *
408  * Restore hard timer Guest.Count & Guest.Cause taking care to preserve the
409  * value of Guest.CP0_Cause.TI while restoring Guest.CP0_Cause.
410  */
411 static void _kvm_vz_restore_htimer(struct kvm_vcpu *vcpu,
412 				   u32 compare, u32 cause)
413 {
414 	u32 start_count, after_count;
415 	ktime_t freeze_time;
416 	unsigned long flags;
417 
418 	/*
419 	 * Freeze the soft-timer and sync the guest CP0_Count with it. We do
420 	 * this with interrupts disabled to avoid latency.
421 	 */
422 	local_irq_save(flags);
423 	freeze_time = kvm_mips_freeze_hrtimer(vcpu, &start_count);
424 	write_c0_gtoffset(start_count - read_c0_count());
425 	local_irq_restore(flags);
426 
427 	/* restore guest CP0_Cause, as TI may already be set */
428 	back_to_back_c0_hazard();
429 	write_gc0_cause(cause);
430 
431 	/*
432 	 * The above sequence isn't atomic and would result in lost timer
433 	 * interrupts if we're not careful. Detect if a timer interrupt is due
434 	 * and assert it.
435 	 */
436 	back_to_back_c0_hazard();
437 	after_count = read_gc0_count();
438 	if (after_count - start_count > compare - start_count - 1)
439 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
440 }
441 
442 /**
443  * kvm_vz_restore_timer() - Restore timer state.
444  * @vcpu:	Virtual CPU.
445  *
446  * Restore soft timer state from saved context.
447  */
448 static void kvm_vz_restore_timer(struct kvm_vcpu *vcpu)
449 {
450 	struct mips_coproc *cop0 = vcpu->arch.cop0;
451 	u32 cause, compare;
452 
453 	compare = kvm_read_sw_gc0_compare(cop0);
454 	cause = kvm_read_sw_gc0_cause(cop0);
455 
456 	write_gc0_compare(compare);
457 	_kvm_vz_restore_stimer(vcpu, compare, cause);
458 }
459 
460 /**
461  * kvm_vz_acquire_htimer() - Switch to hard timer state.
462  * @vcpu:	Virtual CPU.
463  *
464  * Restore hard timer state on top of existing soft timer state if possible.
465  *
466  * Since hard timer won't remain active over preemption, preemption should be
467  * disabled by the caller.
468  */
469 void kvm_vz_acquire_htimer(struct kvm_vcpu *vcpu)
470 {
471 	u32 gctl0;
472 
473 	gctl0 = read_c0_guestctl0();
474 	if (!(gctl0 & MIPS_GCTL0_GT) && kvm_vz_should_use_htimer(vcpu)) {
475 		/* enable guest access to hard timer */
476 		write_c0_guestctl0(gctl0 | MIPS_GCTL0_GT);
477 
478 		_kvm_vz_restore_htimer(vcpu, read_gc0_compare(),
479 				       read_gc0_cause());
480 	}
481 }
482 
483 /**
484  * _kvm_vz_save_htimer() - Switch to software emulation of guest timer.
485  * @vcpu:	Virtual CPU.
486  * @compare:	Pointer to write compare value to.
487  * @cause:	Pointer to write cause value to.
488  *
489  * Save VZ guest timer state and switch to software emulation of guest CP0
490  * timer. The hard timer must already be in use, so preemption should be
491  * disabled.
492  */
493 static void _kvm_vz_save_htimer(struct kvm_vcpu *vcpu,
494 				u32 *out_compare, u32 *out_cause)
495 {
496 	u32 cause, compare, before_count, end_count;
497 	ktime_t before_time;
498 
499 	compare = read_gc0_compare();
500 	*out_compare = compare;
501 
502 	before_time = ktime_get();
503 
504 	/*
505 	 * Record the CP0_Count *prior* to saving CP0_Cause, so we have a time
506 	 * at which no pending timer interrupt is missing.
507 	 */
508 	before_count = read_gc0_count();
509 	back_to_back_c0_hazard();
510 	cause = read_gc0_cause();
511 	*out_cause = cause;
512 
513 	/*
514 	 * Record a final CP0_Count which we will transfer to the soft-timer.
515 	 * This is recorded *after* saving CP0_Cause, so we don't get any timer
516 	 * interrupts from just after the final CP0_Count point.
517 	 */
518 	back_to_back_c0_hazard();
519 	end_count = read_gc0_count();
520 
521 	/*
522 	 * The above sequence isn't atomic, so we could miss a timer interrupt
523 	 * between reading CP0_Cause and end_count. Detect and record any timer
524 	 * interrupt due between before_count and end_count.
525 	 */
526 	if (end_count - before_count > compare - before_count - 1)
527 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
528 
529 	/*
530 	 * Restore soft-timer, ignoring a small amount of negative drift due to
531 	 * delay between freeze_hrtimer and setting CP0_GTOffset.
532 	 */
533 	kvm_mips_restore_hrtimer(vcpu, before_time, end_count, -0x10000);
534 }
535 
536 /**
537  * kvm_vz_save_timer() - Save guest timer state.
538  * @vcpu:	Virtual CPU.
539  *
540  * Save VZ guest timer state and switch to soft guest timer if hard timer was in
541  * use.
542  */
543 static void kvm_vz_save_timer(struct kvm_vcpu *vcpu)
544 {
545 	struct mips_coproc *cop0 = vcpu->arch.cop0;
546 	u32 gctl0, compare, cause;
547 
548 	gctl0 = read_c0_guestctl0();
549 	if (gctl0 & MIPS_GCTL0_GT) {
550 		/* disable guest use of hard timer */
551 		write_c0_guestctl0(gctl0 & ~MIPS_GCTL0_GT);
552 
553 		/* save hard timer state */
554 		_kvm_vz_save_htimer(vcpu, &compare, &cause);
555 	} else {
556 		compare = read_gc0_compare();
557 		cause = read_gc0_cause();
558 	}
559 
560 	/* save timer-related state to VCPU context */
561 	kvm_write_sw_gc0_cause(cop0, cause);
562 	kvm_write_sw_gc0_compare(cop0, compare);
563 }
564 
565 /**
566  * kvm_vz_lose_htimer() - Ensure hard guest timer is not in use.
567  * @vcpu:	Virtual CPU.
568  *
569  * Transfers the state of the hard guest timer to the soft guest timer, leaving
570  * guest state intact so it can continue to be used with the soft timer.
571  */
572 void kvm_vz_lose_htimer(struct kvm_vcpu *vcpu)
573 {
574 	u32 gctl0, compare, cause;
575 
576 	preempt_disable();
577 	gctl0 = read_c0_guestctl0();
578 	if (gctl0 & MIPS_GCTL0_GT) {
579 		/* disable guest use of timer */
580 		write_c0_guestctl0(gctl0 & ~MIPS_GCTL0_GT);
581 
582 		/* switch to soft timer */
583 		_kvm_vz_save_htimer(vcpu, &compare, &cause);
584 
585 		/* leave soft timer in usable state */
586 		_kvm_vz_restore_stimer(vcpu, compare, cause);
587 	}
588 	preempt_enable();
589 }
590 
591 /**
592  * is_eva_access() - Find whether an instruction is an EVA memory accessor.
593  * @inst:	32-bit instruction encoding.
594  *
595  * Finds whether @inst encodes an EVA memory access instruction, which would
596  * indicate that emulation of it should access the user mode address space
597  * instead of the kernel mode address space. This matters for MUSUK segments
598  * which are TLB mapped for user mode but unmapped for kernel mode.
599  *
600  * Returns:	Whether @inst encodes an EVA accessor instruction.
601  */
602 static bool is_eva_access(union mips_instruction inst)
603 {
604 	if (inst.spec3_format.opcode != spec3_op)
605 		return false;
606 
607 	switch (inst.spec3_format.func) {
608 	case lwle_op:
609 	case lwre_op:
610 	case cachee_op:
611 	case sbe_op:
612 	case she_op:
613 	case sce_op:
614 	case swe_op:
615 	case swle_op:
616 	case swre_op:
617 	case prefe_op:
618 	case lbue_op:
619 	case lhue_op:
620 	case lbe_op:
621 	case lhe_op:
622 	case lle_op:
623 	case lwe_op:
624 		return true;
625 	default:
626 		return false;
627 	}
628 }
629 
630 /**
631  * is_eva_am_mapped() - Find whether an access mode is mapped.
632  * @vcpu:	KVM VCPU state.
633  * @am:		3-bit encoded access mode.
634  * @eu:		Segment becomes unmapped and uncached when Status.ERL=1.
635  *
636  * Decode @am to find whether it encodes a mapped segment for the current VCPU
637  * state. Where necessary @eu and the actual instruction causing the fault are
638  * taken into account to make the decision.
639  *
640  * Returns:	Whether the VCPU faulted on a TLB mapped address.
641  */
642 static bool is_eva_am_mapped(struct kvm_vcpu *vcpu, unsigned int am, bool eu)
643 {
644 	u32 am_lookup;
645 	int err;
646 
647 	/*
648 	 * Interpret access control mode. We assume address errors will already
649 	 * have been caught by the guest, leaving us with:
650 	 *      AM      UM  SM  KM  31..24 23..16
651 	 * UK    0 000          Unm   0      0
652 	 * MK    1 001          TLB   1
653 	 * MSK   2 010      TLB TLB   1
654 	 * MUSK  3 011  TLB TLB TLB   1
655 	 * MUSUK 4 100  TLB TLB Unm   0      1
656 	 * USK   5 101      Unm Unm   0      0
657 	 * -     6 110                0      0
658 	 * UUSK  7 111  Unm Unm Unm   0      0
659 	 *
660 	 * We shift a magic value by AM across the sign bit to find if always
661 	 * TLB mapped, and if not shift by 8 again to find if it depends on KM.
662 	 */
663 	am_lookup = 0x70080000 << am;
664 	if ((s32)am_lookup < 0) {
665 		/*
666 		 * MK, MSK, MUSK
667 		 * Always TLB mapped, unless SegCtl.EU && ERL
668 		 */
669 		if (!eu || !(read_gc0_status() & ST0_ERL))
670 			return true;
671 	} else {
672 		am_lookup <<= 8;
673 		if ((s32)am_lookup < 0) {
674 			union mips_instruction inst;
675 			unsigned int status;
676 			u32 *opc;
677 
678 			/*
679 			 * MUSUK
680 			 * TLB mapped if not in kernel mode
681 			 */
682 			status = read_gc0_status();
683 			if (!(status & (ST0_EXL | ST0_ERL)) &&
684 			    (status & ST0_KSU))
685 				return true;
686 			/*
687 			 * EVA access instructions in kernel
688 			 * mode access user address space.
689 			 */
690 			opc = (u32 *)vcpu->arch.pc;
691 			if (vcpu->arch.host_cp0_cause & CAUSEF_BD)
692 				opc += 1;
693 			err = kvm_get_badinstr(opc, vcpu, &inst.word);
694 			if (!err && is_eva_access(inst))
695 				return true;
696 		}
697 	}
698 
699 	return false;
700 }
701 
702 /**
703  * kvm_vz_gva_to_gpa() - Convert valid GVA to GPA.
704  * @vcpu:	KVM VCPU state.
705  * @gva:	Guest virtual address to convert.
706  * @gpa:	Output guest physical address.
707  *
708  * Convert a guest virtual address (GVA) which is valid according to the guest
709  * context, to a guest physical address (GPA).
710  *
711  * Returns:	0 on success.
712  *		-errno on failure.
713  */
714 static int kvm_vz_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
715 			     unsigned long *gpa)
716 {
717 	u32 gva32 = gva;
718 	unsigned long segctl;
719 
720 	if ((long)gva == (s32)gva32) {
721 		/* Handle canonical 32-bit virtual address */
722 		if (cpu_guest_has_segments) {
723 			unsigned long mask, pa;
724 
725 			switch (gva32 >> 29) {
726 			case 0:
727 			case 1: /* CFG5 (1GB) */
728 				segctl = read_gc0_segctl2() >> 16;
729 				mask = (unsigned long)0xfc0000000ull;
730 				break;
731 			case 2:
732 			case 3: /* CFG4 (1GB) */
733 				segctl = read_gc0_segctl2();
734 				mask = (unsigned long)0xfc0000000ull;
735 				break;
736 			case 4: /* CFG3 (512MB) */
737 				segctl = read_gc0_segctl1() >> 16;
738 				mask = (unsigned long)0xfe0000000ull;
739 				break;
740 			case 5: /* CFG2 (512MB) */
741 				segctl = read_gc0_segctl1();
742 				mask = (unsigned long)0xfe0000000ull;
743 				break;
744 			case 6: /* CFG1 (512MB) */
745 				segctl = read_gc0_segctl0() >> 16;
746 				mask = (unsigned long)0xfe0000000ull;
747 				break;
748 			case 7: /* CFG0 (512MB) */
749 				segctl = read_gc0_segctl0();
750 				mask = (unsigned long)0xfe0000000ull;
751 				break;
752 			default:
753 				/*
754 				 * GCC 4.9 isn't smart enough to figure out that
755 				 * segctl and mask are always initialised.
756 				 */
757 				unreachable();
758 			}
759 
760 			if (is_eva_am_mapped(vcpu, (segctl >> 4) & 0x7,
761 					     segctl & 0x0008))
762 				goto tlb_mapped;
763 
764 			/* Unmapped, find guest physical address */
765 			pa = (segctl << 20) & mask;
766 			pa |= gva32 & ~mask;
767 			*gpa = pa;
768 			return 0;
769 		} else if ((s32)gva32 < (s32)0xc0000000) {
770 			/* legacy unmapped KSeg0 or KSeg1 */
771 			*gpa = gva32 & 0x1fffffff;
772 			return 0;
773 		}
774 #ifdef CONFIG_64BIT
775 	} else if ((gva & 0xc000000000000000) == 0x8000000000000000) {
776 		/* XKPHYS */
777 		if (cpu_guest_has_segments) {
778 			/*
779 			 * Each of the 8 regions can be overridden by SegCtl2.XR
780 			 * to use SegCtl1.XAM.
781 			 */
782 			segctl = read_gc0_segctl2();
783 			if (segctl & (1ull << (56 + ((gva >> 59) & 0x7)))) {
784 				segctl = read_gc0_segctl1();
785 				if (is_eva_am_mapped(vcpu, (segctl >> 59) & 0x7,
786 						     0))
787 					goto tlb_mapped;
788 			}
789 
790 		}
791 		/*
792 		 * Traditionally fully unmapped.
793 		 * Bits 61:59 specify the CCA, which we can just mask off here.
794 		 * Bits 58:PABITS should be zero, but we shouldn't have got here
795 		 * if it wasn't.
796 		 */
797 		*gpa = gva & 0x07ffffffffffffff;
798 		return 0;
799 #endif
800 	}
801 
802 tlb_mapped:
803 	return kvm_vz_guest_tlb_lookup(vcpu, gva, gpa);
804 }
805 
806 /**
807  * kvm_vz_badvaddr_to_gpa() - Convert GVA BadVAddr from root exception to GPA.
808  * @vcpu:	KVM VCPU state.
809  * @badvaddr:	Root BadVAddr.
810  * @gpa:	Output guest physical address.
811  *
812  * VZ implementations are permitted to report guest virtual addresses (GVA) in
813  * BadVAddr on a root exception during guest execution, instead of the more
814  * convenient guest physical addresses (GPA). When we get a GVA, this function
815  * converts it to a GPA, taking into account guest segmentation and guest TLB
816  * state.
817  *
818  * Returns:	0 on success.
819  *		-errno on failure.
820  */
821 static int kvm_vz_badvaddr_to_gpa(struct kvm_vcpu *vcpu, unsigned long badvaddr,
822 				  unsigned long *gpa)
823 {
824 	unsigned int gexccode = (vcpu->arch.host_cp0_guestctl0 &
825 				 MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
826 
827 	/* If BadVAddr is GPA, then all is well in the world */
828 	if (likely(gexccode == MIPS_GCTL0_GEXC_GPA)) {
829 		*gpa = badvaddr;
830 		return 0;
831 	}
832 
833 	/* Otherwise we'd expect it to be GVA ... */
834 	if (WARN(gexccode != MIPS_GCTL0_GEXC_GVA,
835 		 "Unexpected gexccode %#x\n", gexccode))
836 		return -EINVAL;
837 
838 	/* ... and we need to perform the GVA->GPA translation in software */
839 	return kvm_vz_gva_to_gpa(vcpu, badvaddr, gpa);
840 }
841 
842 static int kvm_trap_vz_no_handler(struct kvm_vcpu *vcpu)
843 {
844 	u32 *opc = (u32 *) vcpu->arch.pc;
845 	u32 cause = vcpu->arch.host_cp0_cause;
846 	u32 exccode = (cause & CAUSEF_EXCCODE) >> CAUSEB_EXCCODE;
847 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
848 	u32 inst = 0;
849 
850 	/*
851 	 *  Fetch the instruction.
852 	 */
853 	if (cause & CAUSEF_BD)
854 		opc += 1;
855 	kvm_get_badinstr(opc, vcpu, &inst);
856 
857 	kvm_err("Exception Code: %d not handled @ PC: %p, inst: 0x%08x BadVaddr: %#lx Status: %#x\n",
858 		exccode, opc, inst, badvaddr,
859 		read_gc0_status());
860 	kvm_arch_vcpu_dump_regs(vcpu);
861 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
862 	return RESUME_HOST;
863 }
864 
865 static unsigned long mips_process_maar(unsigned int op, unsigned long val)
866 {
867 	/* Mask off unused bits */
868 	unsigned long mask = 0xfffff000 | MIPS_MAAR_S | MIPS_MAAR_VL;
869 
870 	if (read_gc0_pagegrain() & PG_ELPA)
871 		mask |= 0x00ffffff00000000ull;
872 	if (cpu_guest_has_mvh)
873 		mask |= MIPS_MAAR_VH;
874 
875 	/* Set or clear VH */
876 	if (op == mtc_op) {
877 		/* clear VH */
878 		val &= ~MIPS_MAAR_VH;
879 	} else if (op == dmtc_op) {
880 		/* set VH to match VL */
881 		val &= ~MIPS_MAAR_VH;
882 		if (val & MIPS_MAAR_VL)
883 			val |= MIPS_MAAR_VH;
884 	}
885 
886 	return val & mask;
887 }
888 
889 static void kvm_write_maari(struct kvm_vcpu *vcpu, unsigned long val)
890 {
891 	struct mips_coproc *cop0 = vcpu->arch.cop0;
892 
893 	val &= MIPS_MAARI_INDEX;
894 	if (val == MIPS_MAARI_INDEX)
895 		kvm_write_sw_gc0_maari(cop0, ARRAY_SIZE(vcpu->arch.maar) - 1);
896 	else if (val < ARRAY_SIZE(vcpu->arch.maar))
897 		kvm_write_sw_gc0_maari(cop0, val);
898 }
899 
900 static enum emulation_result kvm_vz_gpsi_cop0(union mips_instruction inst,
901 					      u32 *opc, u32 cause,
902 					      struct kvm_run *run,
903 					      struct kvm_vcpu *vcpu)
904 {
905 	struct mips_coproc *cop0 = vcpu->arch.cop0;
906 	enum emulation_result er = EMULATE_DONE;
907 	u32 rt, rd, sel;
908 	unsigned long curr_pc;
909 	unsigned long val;
910 
911 	/*
912 	 * Update PC and hold onto current PC in case there is
913 	 * an error and we want to rollback the PC
914 	 */
915 	curr_pc = vcpu->arch.pc;
916 	er = update_pc(vcpu, cause);
917 	if (er == EMULATE_FAIL)
918 		return er;
919 
920 	if (inst.co_format.co) {
921 		switch (inst.co_format.func) {
922 		case wait_op:
923 			er = kvm_mips_emul_wait(vcpu);
924 			break;
925 		default:
926 			er = EMULATE_FAIL;
927 		}
928 	} else {
929 		rt = inst.c0r_format.rt;
930 		rd = inst.c0r_format.rd;
931 		sel = inst.c0r_format.sel;
932 
933 		switch (inst.c0r_format.rs) {
934 		case dmfc_op:
935 		case mfc_op:
936 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
937 			cop0->stat[rd][sel]++;
938 #endif
939 			if (rd == MIPS_CP0_COUNT &&
940 			    sel == 0) {			/* Count */
941 				val = kvm_mips_read_count(vcpu);
942 			} else if (rd == MIPS_CP0_COMPARE &&
943 				   sel == 0) {		/* Compare */
944 				val = read_gc0_compare();
945 			} else if (rd == MIPS_CP0_LLADDR &&
946 				   sel == 0) {		/* LLAddr */
947 				if (cpu_guest_has_rw_llb)
948 					val = read_gc0_lladdr() &
949 						MIPS_LLADDR_LLB;
950 				else
951 					val = 0;
952 			} else if (rd == MIPS_CP0_LLADDR &&
953 				   sel == 1 &&		/* MAAR */
954 				   cpu_guest_has_maar &&
955 				   !cpu_guest_has_dyn_maar) {
956 				/* MAARI must be in range */
957 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
958 						ARRAY_SIZE(vcpu->arch.maar));
959 				val = vcpu->arch.maar[
960 					kvm_read_sw_gc0_maari(cop0)];
961 			} else if ((rd == MIPS_CP0_PRID &&
962 				    (sel == 0 ||	/* PRid */
963 				     sel == 2 ||	/* CDMMBase */
964 				     sel == 3)) ||	/* CMGCRBase */
965 				   (rd == MIPS_CP0_STATUS &&
966 				    (sel == 2 ||	/* SRSCtl */
967 				     sel == 3)) ||	/* SRSMap */
968 				   (rd == MIPS_CP0_CONFIG &&
969 				    (sel == 7)) ||	/* Config7 */
970 				   (rd == MIPS_CP0_LLADDR &&
971 				    (sel == 2) &&	/* MAARI */
972 				    cpu_guest_has_maar &&
973 				    !cpu_guest_has_dyn_maar) ||
974 				   (rd == MIPS_CP0_ERRCTL &&
975 				    (sel == 0))) {	/* ErrCtl */
976 				val = cop0->reg[rd][sel];
977 			} else {
978 				val = 0;
979 				er = EMULATE_FAIL;
980 			}
981 
982 			if (er != EMULATE_FAIL) {
983 				/* Sign extend */
984 				if (inst.c0r_format.rs == mfc_op)
985 					val = (int)val;
986 				vcpu->arch.gprs[rt] = val;
987 			}
988 
989 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mfc_op) ?
990 					KVM_TRACE_MFC0 : KVM_TRACE_DMFC0,
991 				      KVM_TRACE_COP0(rd, sel), val);
992 			break;
993 
994 		case dmtc_op:
995 		case mtc_op:
996 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
997 			cop0->stat[rd][sel]++;
998 #endif
999 			val = vcpu->arch.gprs[rt];
1000 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mtc_op) ?
1001 					KVM_TRACE_MTC0 : KVM_TRACE_DMTC0,
1002 				      KVM_TRACE_COP0(rd, sel), val);
1003 
1004 			if (rd == MIPS_CP0_COUNT &&
1005 			    sel == 0) {			/* Count */
1006 				kvm_vz_lose_htimer(vcpu);
1007 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
1008 			} else if (rd == MIPS_CP0_COMPARE &&
1009 				   sel == 0) {		/* Compare */
1010 				kvm_mips_write_compare(vcpu,
1011 						       vcpu->arch.gprs[rt],
1012 						       true);
1013 			} else if (rd == MIPS_CP0_LLADDR &&
1014 				   sel == 0) {		/* LLAddr */
1015 				/*
1016 				 * P5600 generates GPSI on guest MTC0 LLAddr.
1017 				 * Only allow the guest to clear LLB.
1018 				 */
1019 				if (cpu_guest_has_rw_llb &&
1020 				    !(val & MIPS_LLADDR_LLB))
1021 					write_gc0_lladdr(0);
1022 			} else if (rd == MIPS_CP0_LLADDR &&
1023 				   sel == 1 &&		/* MAAR */
1024 				   cpu_guest_has_maar &&
1025 				   !cpu_guest_has_dyn_maar) {
1026 				val = mips_process_maar(inst.c0r_format.rs,
1027 							val);
1028 
1029 				/* MAARI must be in range */
1030 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
1031 						ARRAY_SIZE(vcpu->arch.maar));
1032 				vcpu->arch.maar[kvm_read_sw_gc0_maari(cop0)] =
1033 									val;
1034 			} else if (rd == MIPS_CP0_LLADDR &&
1035 				   (sel == 2) &&	/* MAARI */
1036 				   cpu_guest_has_maar &&
1037 				   !cpu_guest_has_dyn_maar) {
1038 				kvm_write_maari(vcpu, val);
1039 			} else if (rd == MIPS_CP0_ERRCTL &&
1040 				   (sel == 0)) {	/* ErrCtl */
1041 				/* ignore the written value */
1042 			} else {
1043 				er = EMULATE_FAIL;
1044 			}
1045 			break;
1046 
1047 		default:
1048 			er = EMULATE_FAIL;
1049 			break;
1050 		}
1051 	}
1052 	/* Rollback PC only if emulation was unsuccessful */
1053 	if (er == EMULATE_FAIL) {
1054 		kvm_err("[%#lx]%s: unsupported cop0 instruction 0x%08x\n",
1055 			curr_pc, __func__, inst.word);
1056 
1057 		vcpu->arch.pc = curr_pc;
1058 	}
1059 
1060 	return er;
1061 }
1062 
1063 static enum emulation_result kvm_vz_gpsi_cache(union mips_instruction inst,
1064 					       u32 *opc, u32 cause,
1065 					       struct kvm_run *run,
1066 					       struct kvm_vcpu *vcpu)
1067 {
1068 	enum emulation_result er = EMULATE_DONE;
1069 	u32 cache, op_inst, op, base;
1070 	s16 offset;
1071 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1072 	unsigned long va, curr_pc;
1073 
1074 	/*
1075 	 * Update PC and hold onto current PC in case there is
1076 	 * an error and we want to rollback the PC
1077 	 */
1078 	curr_pc = vcpu->arch.pc;
1079 	er = update_pc(vcpu, cause);
1080 	if (er == EMULATE_FAIL)
1081 		return er;
1082 
1083 	base = inst.i_format.rs;
1084 	op_inst = inst.i_format.rt;
1085 	if (cpu_has_mips_r6)
1086 		offset = inst.spec3_format.simmediate;
1087 	else
1088 		offset = inst.i_format.simmediate;
1089 	cache = op_inst & CacheOp_Cache;
1090 	op = op_inst & CacheOp_Op;
1091 
1092 	va = arch->gprs[base] + offset;
1093 
1094 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1095 		  cache, op, base, arch->gprs[base], offset);
1096 
1097 	/* Secondary or tirtiary cache ops ignored */
1098 	if (cache != Cache_I && cache != Cache_D)
1099 		return EMULATE_DONE;
1100 
1101 	switch (op_inst) {
1102 	case Index_Invalidate_I:
1103 		flush_icache_line_indexed(va);
1104 		return EMULATE_DONE;
1105 	case Index_Writeback_Inv_D:
1106 		flush_dcache_line_indexed(va);
1107 		return EMULATE_DONE;
1108 	default:
1109 		break;
1110 	};
1111 
1112 	kvm_err("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1113 		curr_pc, vcpu->arch.gprs[31], cache, op, base, arch->gprs[base],
1114 		offset);
1115 	/* Rollback PC */
1116 	vcpu->arch.pc = curr_pc;
1117 
1118 	return EMULATE_FAIL;
1119 }
1120 
1121 static enum emulation_result kvm_trap_vz_handle_gpsi(u32 cause, u32 *opc,
1122 						     struct kvm_vcpu *vcpu)
1123 {
1124 	enum emulation_result er = EMULATE_DONE;
1125 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1126 	struct kvm_run *run = vcpu->run;
1127 	union mips_instruction inst;
1128 	int rd, rt, sel;
1129 	int err;
1130 
1131 	/*
1132 	 *  Fetch the instruction.
1133 	 */
1134 	if (cause & CAUSEF_BD)
1135 		opc += 1;
1136 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1137 	if (err)
1138 		return EMULATE_FAIL;
1139 
1140 	switch (inst.r_format.opcode) {
1141 	case cop0_op:
1142 		er = kvm_vz_gpsi_cop0(inst, opc, cause, run, vcpu);
1143 		break;
1144 #ifndef CONFIG_CPU_MIPSR6
1145 	case cache_op:
1146 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1147 		er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
1148 		break;
1149 #endif
1150 	case spec3_op:
1151 		switch (inst.spec3_format.func) {
1152 #ifdef CONFIG_CPU_MIPSR6
1153 		case cache6_op:
1154 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1155 			er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
1156 			break;
1157 #endif
1158 		case rdhwr_op:
1159 			if (inst.r_format.rs || (inst.r_format.re >> 3))
1160 				goto unknown;
1161 
1162 			rd = inst.r_format.rd;
1163 			rt = inst.r_format.rt;
1164 			sel = inst.r_format.re & 0x7;
1165 
1166 			switch (rd) {
1167 			case MIPS_HWR_CC:	/* Read count register */
1168 				arch->gprs[rt] =
1169 					(long)(int)kvm_mips_read_count(vcpu);
1170 				break;
1171 			default:
1172 				trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1173 					      KVM_TRACE_HWR(rd, sel), 0);
1174 				goto unknown;
1175 			};
1176 
1177 			trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1178 				      KVM_TRACE_HWR(rd, sel), arch->gprs[rt]);
1179 
1180 			er = update_pc(vcpu, cause);
1181 			break;
1182 		default:
1183 			goto unknown;
1184 		};
1185 		break;
1186 unknown:
1187 
1188 	default:
1189 		kvm_err("GPSI exception not supported (%p/%#x)\n",
1190 				opc, inst.word);
1191 		kvm_arch_vcpu_dump_regs(vcpu);
1192 		er = EMULATE_FAIL;
1193 		break;
1194 	}
1195 
1196 	return er;
1197 }
1198 
1199 static enum emulation_result kvm_trap_vz_handle_gsfc(u32 cause, u32 *opc,
1200 						     struct kvm_vcpu *vcpu)
1201 {
1202 	enum emulation_result er = EMULATE_DONE;
1203 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1204 	union mips_instruction inst;
1205 	int err;
1206 
1207 	/*
1208 	 *  Fetch the instruction.
1209 	 */
1210 	if (cause & CAUSEF_BD)
1211 		opc += 1;
1212 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1213 	if (err)
1214 		return EMULATE_FAIL;
1215 
1216 	/* complete MTC0 on behalf of guest and advance EPC */
1217 	if (inst.c0r_format.opcode == cop0_op &&
1218 	    inst.c0r_format.rs == mtc_op &&
1219 	    inst.c0r_format.z == 0) {
1220 		int rt = inst.c0r_format.rt;
1221 		int rd = inst.c0r_format.rd;
1222 		int sel = inst.c0r_format.sel;
1223 		unsigned int val = arch->gprs[rt];
1224 		unsigned int old_val, change;
1225 
1226 		trace_kvm_hwr(vcpu, KVM_TRACE_MTC0, KVM_TRACE_COP0(rd, sel),
1227 			      val);
1228 
1229 		if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1230 			/* FR bit should read as zero if no FPU */
1231 			if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1232 				val &= ~(ST0_CU1 | ST0_FR);
1233 
1234 			/*
1235 			 * Also don't allow FR to be set if host doesn't support
1236 			 * it.
1237 			 */
1238 			if (!(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
1239 				val &= ~ST0_FR;
1240 
1241 			old_val = read_gc0_status();
1242 			change = val ^ old_val;
1243 
1244 			if (change & ST0_FR) {
1245 				/*
1246 				 * FPU and Vector register state is made
1247 				 * UNPREDICTABLE by a change of FR, so don't
1248 				 * even bother saving it.
1249 				 */
1250 				kvm_drop_fpu(vcpu);
1251 			}
1252 
1253 			/*
1254 			 * If MSA state is already live, it is undefined how it
1255 			 * interacts with FR=0 FPU state, and we don't want to
1256 			 * hit reserved instruction exceptions trying to save
1257 			 * the MSA state later when CU=1 && FR=1, so play it
1258 			 * safe and save it first.
1259 			 */
1260 			if (change & ST0_CU1 && !(val & ST0_FR) &&
1261 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1262 				kvm_lose_fpu(vcpu);
1263 
1264 			write_gc0_status(val);
1265 		} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1266 			u32 old_cause = read_gc0_cause();
1267 			u32 change = old_cause ^ val;
1268 
1269 			/* DC bit enabling/disabling timer? */
1270 			if (change & CAUSEF_DC) {
1271 				if (val & CAUSEF_DC) {
1272 					kvm_vz_lose_htimer(vcpu);
1273 					kvm_mips_count_disable_cause(vcpu);
1274 				} else {
1275 					kvm_mips_count_enable_cause(vcpu);
1276 				}
1277 			}
1278 
1279 			/* Only certain bits are RW to the guest */
1280 			change &= (CAUSEF_DC | CAUSEF_IV | CAUSEF_WP |
1281 				   CAUSEF_IP0 | CAUSEF_IP1);
1282 
1283 			/* WP can only be cleared */
1284 			change &= ~CAUSEF_WP | old_cause;
1285 
1286 			write_gc0_cause(old_cause ^ change);
1287 		} else if ((rd == MIPS_CP0_STATUS) && (sel == 1)) { /* IntCtl */
1288 			write_gc0_intctl(val);
1289 		} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1290 			old_val = read_gc0_config5();
1291 			change = val ^ old_val;
1292 			/* Handle changes in FPU/MSA modes */
1293 			preempt_disable();
1294 
1295 			/*
1296 			 * Propagate FRE changes immediately if the FPU
1297 			 * context is already loaded.
1298 			 */
1299 			if (change & MIPS_CONF5_FRE &&
1300 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1301 				change_c0_config5(MIPS_CONF5_FRE, val);
1302 
1303 			preempt_enable();
1304 
1305 			val = old_val ^
1306 				(change & kvm_vz_config5_guest_wrmask(vcpu));
1307 			write_gc0_config5(val);
1308 		} else {
1309 			kvm_err("Handle GSFC, unsupported field change @ %p: %#x\n",
1310 			    opc, inst.word);
1311 			er = EMULATE_FAIL;
1312 		}
1313 
1314 		if (er != EMULATE_FAIL)
1315 			er = update_pc(vcpu, cause);
1316 	} else {
1317 		kvm_err("Handle GSFC, unrecognized instruction @ %p: %#x\n",
1318 			opc, inst.word);
1319 		er = EMULATE_FAIL;
1320 	}
1321 
1322 	return er;
1323 }
1324 
1325 static enum emulation_result kvm_trap_vz_handle_hc(u32 cause, u32 *opc,
1326 						   struct kvm_vcpu *vcpu)
1327 {
1328 	enum emulation_result er;
1329 	union mips_instruction inst;
1330 	unsigned long curr_pc;
1331 	int err;
1332 
1333 	if (cause & CAUSEF_BD)
1334 		opc += 1;
1335 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1336 	if (err)
1337 		return EMULATE_FAIL;
1338 
1339 	/*
1340 	 * Update PC and hold onto current PC in case there is
1341 	 * an error and we want to rollback the PC
1342 	 */
1343 	curr_pc = vcpu->arch.pc;
1344 	er = update_pc(vcpu, cause);
1345 	if (er == EMULATE_FAIL)
1346 		return er;
1347 
1348 	er = kvm_mips_emul_hypcall(vcpu, inst);
1349 	if (er == EMULATE_FAIL)
1350 		vcpu->arch.pc = curr_pc;
1351 
1352 	return er;
1353 }
1354 
1355 static enum emulation_result kvm_trap_vz_no_handler_guest_exit(u32 gexccode,
1356 							u32 cause,
1357 							u32 *opc,
1358 							struct kvm_vcpu *vcpu)
1359 {
1360 	u32 inst;
1361 
1362 	/*
1363 	 *  Fetch the instruction.
1364 	 */
1365 	if (cause & CAUSEF_BD)
1366 		opc += 1;
1367 	kvm_get_badinstr(opc, vcpu, &inst);
1368 
1369 	kvm_err("Guest Exception Code: %d not yet handled @ PC: %p, inst: 0x%08x  Status: %#x\n",
1370 		gexccode, opc, inst, read_gc0_status());
1371 
1372 	return EMULATE_FAIL;
1373 }
1374 
1375 static int kvm_trap_vz_handle_guest_exit(struct kvm_vcpu *vcpu)
1376 {
1377 	u32 *opc = (u32 *) vcpu->arch.pc;
1378 	u32 cause = vcpu->arch.host_cp0_cause;
1379 	enum emulation_result er = EMULATE_DONE;
1380 	u32 gexccode = (vcpu->arch.host_cp0_guestctl0 &
1381 			MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
1382 	int ret = RESUME_GUEST;
1383 
1384 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_GEXCCODE_BASE + gexccode);
1385 	switch (gexccode) {
1386 	case MIPS_GCTL0_GEXC_GPSI:
1387 		++vcpu->stat.vz_gpsi_exits;
1388 		er = kvm_trap_vz_handle_gpsi(cause, opc, vcpu);
1389 		break;
1390 	case MIPS_GCTL0_GEXC_GSFC:
1391 		++vcpu->stat.vz_gsfc_exits;
1392 		er = kvm_trap_vz_handle_gsfc(cause, opc, vcpu);
1393 		break;
1394 	case MIPS_GCTL0_GEXC_HC:
1395 		++vcpu->stat.vz_hc_exits;
1396 		er = kvm_trap_vz_handle_hc(cause, opc, vcpu);
1397 		break;
1398 	case MIPS_GCTL0_GEXC_GRR:
1399 		++vcpu->stat.vz_grr_exits;
1400 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1401 						       vcpu);
1402 		break;
1403 	case MIPS_GCTL0_GEXC_GVA:
1404 		++vcpu->stat.vz_gva_exits;
1405 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1406 						       vcpu);
1407 		break;
1408 	case MIPS_GCTL0_GEXC_GHFC:
1409 		++vcpu->stat.vz_ghfc_exits;
1410 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1411 						       vcpu);
1412 		break;
1413 	case MIPS_GCTL0_GEXC_GPA:
1414 		++vcpu->stat.vz_gpa_exits;
1415 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1416 						       vcpu);
1417 		break;
1418 	default:
1419 		++vcpu->stat.vz_resvd_exits;
1420 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1421 						       vcpu);
1422 		break;
1423 
1424 	}
1425 
1426 	if (er == EMULATE_DONE) {
1427 		ret = RESUME_GUEST;
1428 	} else if (er == EMULATE_HYPERCALL) {
1429 		ret = kvm_mips_handle_hypcall(vcpu);
1430 	} else {
1431 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1432 		ret = RESUME_HOST;
1433 	}
1434 	return ret;
1435 }
1436 
1437 /**
1438  * kvm_trap_vz_handle_cop_unusuable() - Guest used unusable coprocessor.
1439  * @vcpu:	Virtual CPU context.
1440  *
1441  * Handle when the guest attempts to use a coprocessor which hasn't been allowed
1442  * by the root context.
1443  */
1444 static int kvm_trap_vz_handle_cop_unusable(struct kvm_vcpu *vcpu)
1445 {
1446 	struct kvm_run *run = vcpu->run;
1447 	u32 cause = vcpu->arch.host_cp0_cause;
1448 	enum emulation_result er = EMULATE_FAIL;
1449 	int ret = RESUME_GUEST;
1450 
1451 	if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 1) {
1452 		/*
1453 		 * If guest FPU not present, the FPU operation should have been
1454 		 * treated as a reserved instruction!
1455 		 * If FPU already in use, we shouldn't get this at all.
1456 		 */
1457 		if (WARN_ON(!kvm_mips_guest_has_fpu(&vcpu->arch) ||
1458 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1459 			preempt_enable();
1460 			return EMULATE_FAIL;
1461 		}
1462 
1463 		kvm_own_fpu(vcpu);
1464 		er = EMULATE_DONE;
1465 	}
1466 	/* other coprocessors not handled */
1467 
1468 	switch (er) {
1469 	case EMULATE_DONE:
1470 		ret = RESUME_GUEST;
1471 		break;
1472 
1473 	case EMULATE_FAIL:
1474 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1475 		ret = RESUME_HOST;
1476 		break;
1477 
1478 	default:
1479 		BUG();
1480 	}
1481 	return ret;
1482 }
1483 
1484 /**
1485  * kvm_trap_vz_handle_msa_disabled() - Guest used MSA while disabled in root.
1486  * @vcpu:	Virtual CPU context.
1487  *
1488  * Handle when the guest attempts to use MSA when it is disabled in the root
1489  * context.
1490  */
1491 static int kvm_trap_vz_handle_msa_disabled(struct kvm_vcpu *vcpu)
1492 {
1493 	struct kvm_run *run = vcpu->run;
1494 
1495 	/*
1496 	 * If MSA not present or not exposed to guest or FR=0, the MSA operation
1497 	 * should have been treated as a reserved instruction!
1498 	 * Same if CU1=1, FR=0.
1499 	 * If MSA already in use, we shouldn't get this at all.
1500 	 */
1501 	if (!kvm_mips_guest_has_msa(&vcpu->arch) ||
1502 	    (read_gc0_status() & (ST0_CU1 | ST0_FR)) == ST0_CU1 ||
1503 	    !(read_gc0_config5() & MIPS_CONF5_MSAEN) ||
1504 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1505 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1506 		return RESUME_HOST;
1507 	}
1508 
1509 	kvm_own_msa(vcpu);
1510 
1511 	return RESUME_GUEST;
1512 }
1513 
1514 static int kvm_trap_vz_handle_tlb_ld_miss(struct kvm_vcpu *vcpu)
1515 {
1516 	struct kvm_run *run = vcpu->run;
1517 	u32 *opc = (u32 *) vcpu->arch.pc;
1518 	u32 cause = vcpu->arch.host_cp0_cause;
1519 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1520 	union mips_instruction inst;
1521 	enum emulation_result er = EMULATE_DONE;
1522 	int err, ret = RESUME_GUEST;
1523 
1524 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, false)) {
1525 		/* A code fetch fault doesn't count as an MMIO */
1526 		if (kvm_is_ifetch_fault(&vcpu->arch)) {
1527 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1528 			return RESUME_HOST;
1529 		}
1530 
1531 		/* Fetch the instruction */
1532 		if (cause & CAUSEF_BD)
1533 			opc += 1;
1534 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1535 		if (err) {
1536 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1537 			return RESUME_HOST;
1538 		}
1539 
1540 		/* Treat as MMIO */
1541 		er = kvm_mips_emulate_load(inst, cause, run, vcpu);
1542 		if (er == EMULATE_FAIL) {
1543 			kvm_err("Guest Emulate Load from MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1544 				opc, badvaddr);
1545 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1546 		}
1547 	}
1548 
1549 	if (er == EMULATE_DONE) {
1550 		ret = RESUME_GUEST;
1551 	} else if (er == EMULATE_DO_MMIO) {
1552 		run->exit_reason = KVM_EXIT_MMIO;
1553 		ret = RESUME_HOST;
1554 	} else {
1555 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1556 		ret = RESUME_HOST;
1557 	}
1558 	return ret;
1559 }
1560 
1561 static int kvm_trap_vz_handle_tlb_st_miss(struct kvm_vcpu *vcpu)
1562 {
1563 	struct kvm_run *run = vcpu->run;
1564 	u32 *opc = (u32 *) vcpu->arch.pc;
1565 	u32 cause = vcpu->arch.host_cp0_cause;
1566 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1567 	union mips_instruction inst;
1568 	enum emulation_result er = EMULATE_DONE;
1569 	int err;
1570 	int ret = RESUME_GUEST;
1571 
1572 	/* Just try the access again if we couldn't do the translation */
1573 	if (kvm_vz_badvaddr_to_gpa(vcpu, badvaddr, &badvaddr))
1574 		return RESUME_GUEST;
1575 	vcpu->arch.host_cp0_badvaddr = badvaddr;
1576 
1577 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, true)) {
1578 		/* Fetch the instruction */
1579 		if (cause & CAUSEF_BD)
1580 			opc += 1;
1581 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1582 		if (err) {
1583 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1584 			return RESUME_HOST;
1585 		}
1586 
1587 		/* Treat as MMIO */
1588 		er = kvm_mips_emulate_store(inst, cause, run, vcpu);
1589 		if (er == EMULATE_FAIL) {
1590 			kvm_err("Guest Emulate Store to MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1591 				opc, badvaddr);
1592 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1593 		}
1594 	}
1595 
1596 	if (er == EMULATE_DONE) {
1597 		ret = RESUME_GUEST;
1598 	} else if (er == EMULATE_DO_MMIO) {
1599 		run->exit_reason = KVM_EXIT_MMIO;
1600 		ret = RESUME_HOST;
1601 	} else {
1602 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1603 		ret = RESUME_HOST;
1604 	}
1605 	return ret;
1606 }
1607 
1608 static u64 kvm_vz_get_one_regs[] = {
1609 	KVM_REG_MIPS_CP0_INDEX,
1610 	KVM_REG_MIPS_CP0_ENTRYLO0,
1611 	KVM_REG_MIPS_CP0_ENTRYLO1,
1612 	KVM_REG_MIPS_CP0_CONTEXT,
1613 	KVM_REG_MIPS_CP0_PAGEMASK,
1614 	KVM_REG_MIPS_CP0_PAGEGRAIN,
1615 	KVM_REG_MIPS_CP0_WIRED,
1616 	KVM_REG_MIPS_CP0_HWRENA,
1617 	KVM_REG_MIPS_CP0_BADVADDR,
1618 	KVM_REG_MIPS_CP0_COUNT,
1619 	KVM_REG_MIPS_CP0_ENTRYHI,
1620 	KVM_REG_MIPS_CP0_COMPARE,
1621 	KVM_REG_MIPS_CP0_STATUS,
1622 	KVM_REG_MIPS_CP0_INTCTL,
1623 	KVM_REG_MIPS_CP0_CAUSE,
1624 	KVM_REG_MIPS_CP0_EPC,
1625 	KVM_REG_MIPS_CP0_PRID,
1626 	KVM_REG_MIPS_CP0_EBASE,
1627 	KVM_REG_MIPS_CP0_CONFIG,
1628 	KVM_REG_MIPS_CP0_CONFIG1,
1629 	KVM_REG_MIPS_CP0_CONFIG2,
1630 	KVM_REG_MIPS_CP0_CONFIG3,
1631 	KVM_REG_MIPS_CP0_CONFIG4,
1632 	KVM_REG_MIPS_CP0_CONFIG5,
1633 #ifdef CONFIG_64BIT
1634 	KVM_REG_MIPS_CP0_XCONTEXT,
1635 #endif
1636 	KVM_REG_MIPS_CP0_ERROREPC,
1637 
1638 	KVM_REG_MIPS_COUNT_CTL,
1639 	KVM_REG_MIPS_COUNT_RESUME,
1640 	KVM_REG_MIPS_COUNT_HZ,
1641 };
1642 
1643 static u64 kvm_vz_get_one_regs_contextconfig[] = {
1644 	KVM_REG_MIPS_CP0_CONTEXTCONFIG,
1645 #ifdef CONFIG_64BIT
1646 	KVM_REG_MIPS_CP0_XCONTEXTCONFIG,
1647 #endif
1648 };
1649 
1650 static u64 kvm_vz_get_one_regs_segments[] = {
1651 	KVM_REG_MIPS_CP0_SEGCTL0,
1652 	KVM_REG_MIPS_CP0_SEGCTL1,
1653 	KVM_REG_MIPS_CP0_SEGCTL2,
1654 };
1655 
1656 static u64 kvm_vz_get_one_regs_htw[] = {
1657 	KVM_REG_MIPS_CP0_PWBASE,
1658 	KVM_REG_MIPS_CP0_PWFIELD,
1659 	KVM_REG_MIPS_CP0_PWSIZE,
1660 	KVM_REG_MIPS_CP0_PWCTL,
1661 };
1662 
1663 static u64 kvm_vz_get_one_regs_kscratch[] = {
1664 	KVM_REG_MIPS_CP0_KSCRATCH1,
1665 	KVM_REG_MIPS_CP0_KSCRATCH2,
1666 	KVM_REG_MIPS_CP0_KSCRATCH3,
1667 	KVM_REG_MIPS_CP0_KSCRATCH4,
1668 	KVM_REG_MIPS_CP0_KSCRATCH5,
1669 	KVM_REG_MIPS_CP0_KSCRATCH6,
1670 };
1671 
1672 static unsigned long kvm_vz_num_regs(struct kvm_vcpu *vcpu)
1673 {
1674 	unsigned long ret;
1675 
1676 	ret = ARRAY_SIZE(kvm_vz_get_one_regs);
1677 	if (cpu_guest_has_userlocal)
1678 		++ret;
1679 	if (cpu_guest_has_badinstr)
1680 		++ret;
1681 	if (cpu_guest_has_badinstrp)
1682 		++ret;
1683 	if (cpu_guest_has_contextconfig)
1684 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1685 	if (cpu_guest_has_segments)
1686 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1687 	if (cpu_guest_has_htw)
1688 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1689 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar)
1690 		ret += 1 + ARRAY_SIZE(vcpu->arch.maar);
1691 	ret += __arch_hweight8(cpu_data[0].guest.kscratch_mask);
1692 
1693 	return ret;
1694 }
1695 
1696 static int kvm_vz_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
1697 {
1698 	u64 index;
1699 	unsigned int i;
1700 
1701 	if (copy_to_user(indices, kvm_vz_get_one_regs,
1702 			 sizeof(kvm_vz_get_one_regs)))
1703 		return -EFAULT;
1704 	indices += ARRAY_SIZE(kvm_vz_get_one_regs);
1705 
1706 	if (cpu_guest_has_userlocal) {
1707 		index = KVM_REG_MIPS_CP0_USERLOCAL;
1708 		if (copy_to_user(indices, &index, sizeof(index)))
1709 			return -EFAULT;
1710 		++indices;
1711 	}
1712 	if (cpu_guest_has_badinstr) {
1713 		index = KVM_REG_MIPS_CP0_BADINSTR;
1714 		if (copy_to_user(indices, &index, sizeof(index)))
1715 			return -EFAULT;
1716 		++indices;
1717 	}
1718 	if (cpu_guest_has_badinstrp) {
1719 		index = KVM_REG_MIPS_CP0_BADINSTRP;
1720 		if (copy_to_user(indices, &index, sizeof(index)))
1721 			return -EFAULT;
1722 		++indices;
1723 	}
1724 	if (cpu_guest_has_contextconfig) {
1725 		if (copy_to_user(indices, kvm_vz_get_one_regs_contextconfig,
1726 				 sizeof(kvm_vz_get_one_regs_contextconfig)))
1727 			return -EFAULT;
1728 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1729 	}
1730 	if (cpu_guest_has_segments) {
1731 		if (copy_to_user(indices, kvm_vz_get_one_regs_segments,
1732 				 sizeof(kvm_vz_get_one_regs_segments)))
1733 			return -EFAULT;
1734 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1735 	}
1736 	if (cpu_guest_has_htw) {
1737 		if (copy_to_user(indices, kvm_vz_get_one_regs_htw,
1738 				 sizeof(kvm_vz_get_one_regs_htw)))
1739 			return -EFAULT;
1740 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1741 	}
1742 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar) {
1743 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.maar); ++i) {
1744 			index = KVM_REG_MIPS_CP0_MAAR(i);
1745 			if (copy_to_user(indices, &index, sizeof(index)))
1746 				return -EFAULT;
1747 			++indices;
1748 		}
1749 
1750 		index = KVM_REG_MIPS_CP0_MAARI;
1751 		if (copy_to_user(indices, &index, sizeof(index)))
1752 			return -EFAULT;
1753 		++indices;
1754 	}
1755 	for (i = 0; i < 6; ++i) {
1756 		if (!cpu_guest_has_kscr(i + 2))
1757 			continue;
1758 
1759 		if (copy_to_user(indices, &kvm_vz_get_one_regs_kscratch[i],
1760 				 sizeof(kvm_vz_get_one_regs_kscratch[i])))
1761 			return -EFAULT;
1762 		++indices;
1763 	}
1764 
1765 	return 0;
1766 }
1767 
1768 static inline s64 entrylo_kvm_to_user(unsigned long v)
1769 {
1770 	s64 mask, ret = v;
1771 
1772 	if (BITS_PER_LONG == 32) {
1773 		/*
1774 		 * KVM API exposes 64-bit version of the register, so move the
1775 		 * RI/XI bits up into place.
1776 		 */
1777 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1778 		ret &= ~mask;
1779 		ret |= ((s64)v & mask) << 32;
1780 	}
1781 	return ret;
1782 }
1783 
1784 static inline unsigned long entrylo_user_to_kvm(s64 v)
1785 {
1786 	unsigned long mask, ret = v;
1787 
1788 	if (BITS_PER_LONG == 32) {
1789 		/*
1790 		 * KVM API exposes 64-bit versiono of the register, so move the
1791 		 * RI/XI bits down into place.
1792 		 */
1793 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1794 		ret &= ~mask;
1795 		ret |= (v >> 32) & mask;
1796 	}
1797 	return ret;
1798 }
1799 
1800 static int kvm_vz_get_one_reg(struct kvm_vcpu *vcpu,
1801 			      const struct kvm_one_reg *reg,
1802 			      s64 *v)
1803 {
1804 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1805 	unsigned int idx;
1806 
1807 	switch (reg->id) {
1808 	case KVM_REG_MIPS_CP0_INDEX:
1809 		*v = (long)read_gc0_index();
1810 		break;
1811 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1812 		*v = entrylo_kvm_to_user(read_gc0_entrylo0());
1813 		break;
1814 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1815 		*v = entrylo_kvm_to_user(read_gc0_entrylo1());
1816 		break;
1817 	case KVM_REG_MIPS_CP0_CONTEXT:
1818 		*v = (long)read_gc0_context();
1819 		break;
1820 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1821 		if (!cpu_guest_has_contextconfig)
1822 			return -EINVAL;
1823 		*v = read_gc0_contextconfig();
1824 		break;
1825 	case KVM_REG_MIPS_CP0_USERLOCAL:
1826 		if (!cpu_guest_has_userlocal)
1827 			return -EINVAL;
1828 		*v = read_gc0_userlocal();
1829 		break;
1830 #ifdef CONFIG_64BIT
1831 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1832 		if (!cpu_guest_has_contextconfig)
1833 			return -EINVAL;
1834 		*v = read_gc0_xcontextconfig();
1835 		break;
1836 #endif
1837 	case KVM_REG_MIPS_CP0_PAGEMASK:
1838 		*v = (long)read_gc0_pagemask();
1839 		break;
1840 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1841 		*v = (long)read_gc0_pagegrain();
1842 		break;
1843 	case KVM_REG_MIPS_CP0_SEGCTL0:
1844 		if (!cpu_guest_has_segments)
1845 			return -EINVAL;
1846 		*v = read_gc0_segctl0();
1847 		break;
1848 	case KVM_REG_MIPS_CP0_SEGCTL1:
1849 		if (!cpu_guest_has_segments)
1850 			return -EINVAL;
1851 		*v = read_gc0_segctl1();
1852 		break;
1853 	case KVM_REG_MIPS_CP0_SEGCTL2:
1854 		if (!cpu_guest_has_segments)
1855 			return -EINVAL;
1856 		*v = read_gc0_segctl2();
1857 		break;
1858 	case KVM_REG_MIPS_CP0_PWBASE:
1859 		if (!cpu_guest_has_htw)
1860 			return -EINVAL;
1861 		*v = read_gc0_pwbase();
1862 		break;
1863 	case KVM_REG_MIPS_CP0_PWFIELD:
1864 		if (!cpu_guest_has_htw)
1865 			return -EINVAL;
1866 		*v = read_gc0_pwfield();
1867 		break;
1868 	case KVM_REG_MIPS_CP0_PWSIZE:
1869 		if (!cpu_guest_has_htw)
1870 			return -EINVAL;
1871 		*v = read_gc0_pwsize();
1872 		break;
1873 	case KVM_REG_MIPS_CP0_WIRED:
1874 		*v = (long)read_gc0_wired();
1875 		break;
1876 	case KVM_REG_MIPS_CP0_PWCTL:
1877 		if (!cpu_guest_has_htw)
1878 			return -EINVAL;
1879 		*v = read_gc0_pwctl();
1880 		break;
1881 	case KVM_REG_MIPS_CP0_HWRENA:
1882 		*v = (long)read_gc0_hwrena();
1883 		break;
1884 	case KVM_REG_MIPS_CP0_BADVADDR:
1885 		*v = (long)read_gc0_badvaddr();
1886 		break;
1887 	case KVM_REG_MIPS_CP0_BADINSTR:
1888 		if (!cpu_guest_has_badinstr)
1889 			return -EINVAL;
1890 		*v = read_gc0_badinstr();
1891 		break;
1892 	case KVM_REG_MIPS_CP0_BADINSTRP:
1893 		if (!cpu_guest_has_badinstrp)
1894 			return -EINVAL;
1895 		*v = read_gc0_badinstrp();
1896 		break;
1897 	case KVM_REG_MIPS_CP0_COUNT:
1898 		*v = kvm_mips_read_count(vcpu);
1899 		break;
1900 	case KVM_REG_MIPS_CP0_ENTRYHI:
1901 		*v = (long)read_gc0_entryhi();
1902 		break;
1903 	case KVM_REG_MIPS_CP0_COMPARE:
1904 		*v = (long)read_gc0_compare();
1905 		break;
1906 	case KVM_REG_MIPS_CP0_STATUS:
1907 		*v = (long)read_gc0_status();
1908 		break;
1909 	case KVM_REG_MIPS_CP0_INTCTL:
1910 		*v = read_gc0_intctl();
1911 		break;
1912 	case KVM_REG_MIPS_CP0_CAUSE:
1913 		*v = (long)read_gc0_cause();
1914 		break;
1915 	case KVM_REG_MIPS_CP0_EPC:
1916 		*v = (long)read_gc0_epc();
1917 		break;
1918 	case KVM_REG_MIPS_CP0_PRID:
1919 		*v = (long)kvm_read_c0_guest_prid(cop0);
1920 		break;
1921 	case KVM_REG_MIPS_CP0_EBASE:
1922 		*v = kvm_vz_read_gc0_ebase();
1923 		break;
1924 	case KVM_REG_MIPS_CP0_CONFIG:
1925 		*v = read_gc0_config();
1926 		break;
1927 	case KVM_REG_MIPS_CP0_CONFIG1:
1928 		if (!cpu_guest_has_conf1)
1929 			return -EINVAL;
1930 		*v = read_gc0_config1();
1931 		break;
1932 	case KVM_REG_MIPS_CP0_CONFIG2:
1933 		if (!cpu_guest_has_conf2)
1934 			return -EINVAL;
1935 		*v = read_gc0_config2();
1936 		break;
1937 	case KVM_REG_MIPS_CP0_CONFIG3:
1938 		if (!cpu_guest_has_conf3)
1939 			return -EINVAL;
1940 		*v = read_gc0_config3();
1941 		break;
1942 	case KVM_REG_MIPS_CP0_CONFIG4:
1943 		if (!cpu_guest_has_conf4)
1944 			return -EINVAL;
1945 		*v = read_gc0_config4();
1946 		break;
1947 	case KVM_REG_MIPS_CP0_CONFIG5:
1948 		if (!cpu_guest_has_conf5)
1949 			return -EINVAL;
1950 		*v = read_gc0_config5();
1951 		break;
1952 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
1953 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
1954 			return -EINVAL;
1955 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
1956 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
1957 			return -EINVAL;
1958 		*v = vcpu->arch.maar[idx];
1959 		break;
1960 	case KVM_REG_MIPS_CP0_MAARI:
1961 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
1962 			return -EINVAL;
1963 		*v = kvm_read_sw_gc0_maari(vcpu->arch.cop0);
1964 		break;
1965 #ifdef CONFIG_64BIT
1966 	case KVM_REG_MIPS_CP0_XCONTEXT:
1967 		*v = read_gc0_xcontext();
1968 		break;
1969 #endif
1970 	case KVM_REG_MIPS_CP0_ERROREPC:
1971 		*v = (long)read_gc0_errorepc();
1972 		break;
1973 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
1974 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
1975 		if (!cpu_guest_has_kscr(idx))
1976 			return -EINVAL;
1977 		switch (idx) {
1978 		case 2:
1979 			*v = (long)read_gc0_kscratch1();
1980 			break;
1981 		case 3:
1982 			*v = (long)read_gc0_kscratch2();
1983 			break;
1984 		case 4:
1985 			*v = (long)read_gc0_kscratch3();
1986 			break;
1987 		case 5:
1988 			*v = (long)read_gc0_kscratch4();
1989 			break;
1990 		case 6:
1991 			*v = (long)read_gc0_kscratch5();
1992 			break;
1993 		case 7:
1994 			*v = (long)read_gc0_kscratch6();
1995 			break;
1996 		}
1997 		break;
1998 	case KVM_REG_MIPS_COUNT_CTL:
1999 		*v = vcpu->arch.count_ctl;
2000 		break;
2001 	case KVM_REG_MIPS_COUNT_RESUME:
2002 		*v = ktime_to_ns(vcpu->arch.count_resume);
2003 		break;
2004 	case KVM_REG_MIPS_COUNT_HZ:
2005 		*v = vcpu->arch.count_hz;
2006 		break;
2007 	default:
2008 		return -EINVAL;
2009 	}
2010 	return 0;
2011 }
2012 
2013 static int kvm_vz_set_one_reg(struct kvm_vcpu *vcpu,
2014 			      const struct kvm_one_reg *reg,
2015 			      s64 v)
2016 {
2017 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2018 	unsigned int idx;
2019 	int ret = 0;
2020 	unsigned int cur, change;
2021 
2022 	switch (reg->id) {
2023 	case KVM_REG_MIPS_CP0_INDEX:
2024 		write_gc0_index(v);
2025 		break;
2026 	case KVM_REG_MIPS_CP0_ENTRYLO0:
2027 		write_gc0_entrylo0(entrylo_user_to_kvm(v));
2028 		break;
2029 	case KVM_REG_MIPS_CP0_ENTRYLO1:
2030 		write_gc0_entrylo1(entrylo_user_to_kvm(v));
2031 		break;
2032 	case KVM_REG_MIPS_CP0_CONTEXT:
2033 		write_gc0_context(v);
2034 		break;
2035 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
2036 		if (!cpu_guest_has_contextconfig)
2037 			return -EINVAL;
2038 		write_gc0_contextconfig(v);
2039 		break;
2040 	case KVM_REG_MIPS_CP0_USERLOCAL:
2041 		if (!cpu_guest_has_userlocal)
2042 			return -EINVAL;
2043 		write_gc0_userlocal(v);
2044 		break;
2045 #ifdef CONFIG_64BIT
2046 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
2047 		if (!cpu_guest_has_contextconfig)
2048 			return -EINVAL;
2049 		write_gc0_xcontextconfig(v);
2050 		break;
2051 #endif
2052 	case KVM_REG_MIPS_CP0_PAGEMASK:
2053 		write_gc0_pagemask(v);
2054 		break;
2055 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
2056 		write_gc0_pagegrain(v);
2057 		break;
2058 	case KVM_REG_MIPS_CP0_SEGCTL0:
2059 		if (!cpu_guest_has_segments)
2060 			return -EINVAL;
2061 		write_gc0_segctl0(v);
2062 		break;
2063 	case KVM_REG_MIPS_CP0_SEGCTL1:
2064 		if (!cpu_guest_has_segments)
2065 			return -EINVAL;
2066 		write_gc0_segctl1(v);
2067 		break;
2068 	case KVM_REG_MIPS_CP0_SEGCTL2:
2069 		if (!cpu_guest_has_segments)
2070 			return -EINVAL;
2071 		write_gc0_segctl2(v);
2072 		break;
2073 	case KVM_REG_MIPS_CP0_PWBASE:
2074 		if (!cpu_guest_has_htw)
2075 			return -EINVAL;
2076 		write_gc0_pwbase(v);
2077 		break;
2078 	case KVM_REG_MIPS_CP0_PWFIELD:
2079 		if (!cpu_guest_has_htw)
2080 			return -EINVAL;
2081 		write_gc0_pwfield(v);
2082 		break;
2083 	case KVM_REG_MIPS_CP0_PWSIZE:
2084 		if (!cpu_guest_has_htw)
2085 			return -EINVAL;
2086 		write_gc0_pwsize(v);
2087 		break;
2088 	case KVM_REG_MIPS_CP0_WIRED:
2089 		change_gc0_wired(MIPSR6_WIRED_WIRED, v);
2090 		break;
2091 	case KVM_REG_MIPS_CP0_PWCTL:
2092 		if (!cpu_guest_has_htw)
2093 			return -EINVAL;
2094 		write_gc0_pwctl(v);
2095 		break;
2096 	case KVM_REG_MIPS_CP0_HWRENA:
2097 		write_gc0_hwrena(v);
2098 		break;
2099 	case KVM_REG_MIPS_CP0_BADVADDR:
2100 		write_gc0_badvaddr(v);
2101 		break;
2102 	case KVM_REG_MIPS_CP0_BADINSTR:
2103 		if (!cpu_guest_has_badinstr)
2104 			return -EINVAL;
2105 		write_gc0_badinstr(v);
2106 		break;
2107 	case KVM_REG_MIPS_CP0_BADINSTRP:
2108 		if (!cpu_guest_has_badinstrp)
2109 			return -EINVAL;
2110 		write_gc0_badinstrp(v);
2111 		break;
2112 	case KVM_REG_MIPS_CP0_COUNT:
2113 		kvm_mips_write_count(vcpu, v);
2114 		break;
2115 	case KVM_REG_MIPS_CP0_ENTRYHI:
2116 		write_gc0_entryhi(v);
2117 		break;
2118 	case KVM_REG_MIPS_CP0_COMPARE:
2119 		kvm_mips_write_compare(vcpu, v, false);
2120 		break;
2121 	case KVM_REG_MIPS_CP0_STATUS:
2122 		write_gc0_status(v);
2123 		break;
2124 	case KVM_REG_MIPS_CP0_INTCTL:
2125 		write_gc0_intctl(v);
2126 		break;
2127 	case KVM_REG_MIPS_CP0_CAUSE:
2128 		/*
2129 		 * If the timer is stopped or started (DC bit) it must look
2130 		 * atomic with changes to the timer interrupt pending bit (TI).
2131 		 * A timer interrupt should not happen in between.
2132 		 */
2133 		if ((read_gc0_cause() ^ v) & CAUSEF_DC) {
2134 			if (v & CAUSEF_DC) {
2135 				/* disable timer first */
2136 				kvm_mips_count_disable_cause(vcpu);
2137 				change_gc0_cause((u32)~CAUSEF_DC, v);
2138 			} else {
2139 				/* enable timer last */
2140 				change_gc0_cause((u32)~CAUSEF_DC, v);
2141 				kvm_mips_count_enable_cause(vcpu);
2142 			}
2143 		} else {
2144 			write_gc0_cause(v);
2145 		}
2146 		break;
2147 	case KVM_REG_MIPS_CP0_EPC:
2148 		write_gc0_epc(v);
2149 		break;
2150 	case KVM_REG_MIPS_CP0_PRID:
2151 		kvm_write_c0_guest_prid(cop0, v);
2152 		break;
2153 	case KVM_REG_MIPS_CP0_EBASE:
2154 		kvm_vz_write_gc0_ebase(v);
2155 		break;
2156 	case KVM_REG_MIPS_CP0_CONFIG:
2157 		cur = read_gc0_config();
2158 		change = (cur ^ v) & kvm_vz_config_user_wrmask(vcpu);
2159 		if (change) {
2160 			v = cur ^ change;
2161 			write_gc0_config(v);
2162 		}
2163 		break;
2164 	case KVM_REG_MIPS_CP0_CONFIG1:
2165 		if (!cpu_guest_has_conf1)
2166 			break;
2167 		cur = read_gc0_config1();
2168 		change = (cur ^ v) & kvm_vz_config1_user_wrmask(vcpu);
2169 		if (change) {
2170 			v = cur ^ change;
2171 			write_gc0_config1(v);
2172 		}
2173 		break;
2174 	case KVM_REG_MIPS_CP0_CONFIG2:
2175 		if (!cpu_guest_has_conf2)
2176 			break;
2177 		cur = read_gc0_config2();
2178 		change = (cur ^ v) & kvm_vz_config2_user_wrmask(vcpu);
2179 		if (change) {
2180 			v = cur ^ change;
2181 			write_gc0_config2(v);
2182 		}
2183 		break;
2184 	case KVM_REG_MIPS_CP0_CONFIG3:
2185 		if (!cpu_guest_has_conf3)
2186 			break;
2187 		cur = read_gc0_config3();
2188 		change = (cur ^ v) & kvm_vz_config3_user_wrmask(vcpu);
2189 		if (change) {
2190 			v = cur ^ change;
2191 			write_gc0_config3(v);
2192 		}
2193 		break;
2194 	case KVM_REG_MIPS_CP0_CONFIG4:
2195 		if (!cpu_guest_has_conf4)
2196 			break;
2197 		cur = read_gc0_config4();
2198 		change = (cur ^ v) & kvm_vz_config4_user_wrmask(vcpu);
2199 		if (change) {
2200 			v = cur ^ change;
2201 			write_gc0_config4(v);
2202 		}
2203 		break;
2204 	case KVM_REG_MIPS_CP0_CONFIG5:
2205 		if (!cpu_guest_has_conf5)
2206 			break;
2207 		cur = read_gc0_config5();
2208 		change = (cur ^ v) & kvm_vz_config5_user_wrmask(vcpu);
2209 		if (change) {
2210 			v = cur ^ change;
2211 			write_gc0_config5(v);
2212 		}
2213 		break;
2214 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2215 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2216 			return -EINVAL;
2217 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2218 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2219 			return -EINVAL;
2220 		vcpu->arch.maar[idx] = mips_process_maar(dmtc_op, v);
2221 		break;
2222 	case KVM_REG_MIPS_CP0_MAARI:
2223 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2224 			return -EINVAL;
2225 		kvm_write_maari(vcpu, v);
2226 		break;
2227 #ifdef CONFIG_64BIT
2228 	case KVM_REG_MIPS_CP0_XCONTEXT:
2229 		write_gc0_xcontext(v);
2230 		break;
2231 #endif
2232 	case KVM_REG_MIPS_CP0_ERROREPC:
2233 		write_gc0_errorepc(v);
2234 		break;
2235 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2236 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2237 		if (!cpu_guest_has_kscr(idx))
2238 			return -EINVAL;
2239 		switch (idx) {
2240 		case 2:
2241 			write_gc0_kscratch1(v);
2242 			break;
2243 		case 3:
2244 			write_gc0_kscratch2(v);
2245 			break;
2246 		case 4:
2247 			write_gc0_kscratch3(v);
2248 			break;
2249 		case 5:
2250 			write_gc0_kscratch4(v);
2251 			break;
2252 		case 6:
2253 			write_gc0_kscratch5(v);
2254 			break;
2255 		case 7:
2256 			write_gc0_kscratch6(v);
2257 			break;
2258 		}
2259 		break;
2260 	case KVM_REG_MIPS_COUNT_CTL:
2261 		ret = kvm_mips_set_count_ctl(vcpu, v);
2262 		break;
2263 	case KVM_REG_MIPS_COUNT_RESUME:
2264 		ret = kvm_mips_set_count_resume(vcpu, v);
2265 		break;
2266 	case KVM_REG_MIPS_COUNT_HZ:
2267 		ret = kvm_mips_set_count_hz(vcpu, v);
2268 		break;
2269 	default:
2270 		return -EINVAL;
2271 	}
2272 	return ret;
2273 }
2274 
2275 #define guestid_cache(cpu)	(cpu_data[cpu].guestid_cache)
2276 static void kvm_vz_get_new_guestid(unsigned long cpu, struct kvm_vcpu *vcpu)
2277 {
2278 	unsigned long guestid = guestid_cache(cpu);
2279 
2280 	if (!(++guestid & GUESTID_MASK)) {
2281 		if (cpu_has_vtag_icache)
2282 			flush_icache_all();
2283 
2284 		if (!guestid)		/* fix version if needed */
2285 			guestid = GUESTID_FIRST_VERSION;
2286 
2287 		++guestid;		/* guestid 0 reserved for root */
2288 
2289 		/* start new guestid cycle */
2290 		kvm_vz_local_flush_roottlb_all_guests();
2291 		kvm_vz_local_flush_guesttlb_all();
2292 	}
2293 
2294 	guestid_cache(cpu) = guestid;
2295 }
2296 
2297 /* Returns 1 if the guest TLB may be clobbered */
2298 static int kvm_vz_check_requests(struct kvm_vcpu *vcpu, int cpu)
2299 {
2300 	int ret = 0;
2301 	int i;
2302 
2303 	if (!vcpu->requests)
2304 		return 0;
2305 
2306 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2307 		if (cpu_has_guestid) {
2308 			/* Drop all GuestIDs for this VCPU */
2309 			for_each_possible_cpu(i)
2310 				vcpu->arch.vzguestid[i] = 0;
2311 			/* This will clobber guest TLB contents too */
2312 			ret = 1;
2313 		}
2314 		/*
2315 		 * For Root ASID Dealias (RAD) we don't do anything here, but we
2316 		 * still need the request to ensure we recheck asid_flush_mask.
2317 		 * We can still return 0 as only the root TLB will be affected
2318 		 * by a root ASID flush.
2319 		 */
2320 	}
2321 
2322 	return ret;
2323 }
2324 
2325 static void kvm_vz_vcpu_save_wired(struct kvm_vcpu *vcpu)
2326 {
2327 	unsigned int wired = read_gc0_wired();
2328 	struct kvm_mips_tlb *tlbs;
2329 	int i;
2330 
2331 	/* Expand the wired TLB array if necessary */
2332 	wired &= MIPSR6_WIRED_WIRED;
2333 	if (wired > vcpu->arch.wired_tlb_limit) {
2334 		tlbs = krealloc(vcpu->arch.wired_tlb, wired *
2335 				sizeof(*vcpu->arch.wired_tlb), GFP_ATOMIC);
2336 		if (WARN_ON(!tlbs)) {
2337 			/* Save whatever we can */
2338 			wired = vcpu->arch.wired_tlb_limit;
2339 		} else {
2340 			vcpu->arch.wired_tlb = tlbs;
2341 			vcpu->arch.wired_tlb_limit = wired;
2342 		}
2343 	}
2344 
2345 	if (wired)
2346 		/* Save wired entries from the guest TLB */
2347 		kvm_vz_save_guesttlb(vcpu->arch.wired_tlb, 0, wired);
2348 	/* Invalidate any dropped entries since last time */
2349 	for (i = wired; i < vcpu->arch.wired_tlb_used; ++i) {
2350 		vcpu->arch.wired_tlb[i].tlb_hi = UNIQUE_GUEST_ENTRYHI(i);
2351 		vcpu->arch.wired_tlb[i].tlb_lo[0] = 0;
2352 		vcpu->arch.wired_tlb[i].tlb_lo[1] = 0;
2353 		vcpu->arch.wired_tlb[i].tlb_mask = 0;
2354 	}
2355 	vcpu->arch.wired_tlb_used = wired;
2356 }
2357 
2358 static void kvm_vz_vcpu_load_wired(struct kvm_vcpu *vcpu)
2359 {
2360 	/* Load wired entries into the guest TLB */
2361 	if (vcpu->arch.wired_tlb)
2362 		kvm_vz_load_guesttlb(vcpu->arch.wired_tlb, 0,
2363 				     vcpu->arch.wired_tlb_used);
2364 }
2365 
2366 static void kvm_vz_vcpu_load_tlb(struct kvm_vcpu *vcpu, int cpu)
2367 {
2368 	struct kvm *kvm = vcpu->kvm;
2369 	struct mm_struct *gpa_mm = &kvm->arch.gpa_mm;
2370 	bool migrated;
2371 
2372 	/*
2373 	 * Are we entering guest context on a different CPU to last time?
2374 	 * If so, the VCPU's guest TLB state on this CPU may be stale.
2375 	 */
2376 	migrated = (vcpu->arch.last_exec_cpu != cpu);
2377 	vcpu->arch.last_exec_cpu = cpu;
2378 
2379 	/*
2380 	 * A vcpu's GuestID is set in GuestCtl1.ID when the vcpu is loaded and
2381 	 * remains set until another vcpu is loaded in.  As a rule GuestRID
2382 	 * remains zeroed when in root context unless the kernel is busy
2383 	 * manipulating guest tlb entries.
2384 	 */
2385 	if (cpu_has_guestid) {
2386 		/*
2387 		 * Check if our GuestID is of an older version and thus invalid.
2388 		 *
2389 		 * We also discard the stored GuestID if we've executed on
2390 		 * another CPU, as the guest mappings may have changed without
2391 		 * hypervisor knowledge.
2392 		 */
2393 		if (migrated ||
2394 		    (vcpu->arch.vzguestid[cpu] ^ guestid_cache(cpu)) &
2395 					GUESTID_VERSION_MASK) {
2396 			kvm_vz_get_new_guestid(cpu, vcpu);
2397 			vcpu->arch.vzguestid[cpu] = guestid_cache(cpu);
2398 			trace_kvm_guestid_change(vcpu,
2399 						 vcpu->arch.vzguestid[cpu]);
2400 		}
2401 
2402 		/* Restore GuestID */
2403 		change_c0_guestctl1(GUESTID_MASK, vcpu->arch.vzguestid[cpu]);
2404 	} else {
2405 		/*
2406 		 * The Guest TLB only stores a single guest's TLB state, so
2407 		 * flush it if another VCPU has executed on this CPU.
2408 		 *
2409 		 * We also flush if we've executed on another CPU, as the guest
2410 		 * mappings may have changed without hypervisor knowledge.
2411 		 */
2412 		if (migrated || last_exec_vcpu[cpu] != vcpu)
2413 			kvm_vz_local_flush_guesttlb_all();
2414 		last_exec_vcpu[cpu] = vcpu;
2415 
2416 		/*
2417 		 * Root ASID dealiases guest GPA mappings in the root TLB.
2418 		 * Allocate new root ASID if needed.
2419 		 */
2420 		if (cpumask_test_and_clear_cpu(cpu, &kvm->arch.asid_flush_mask)
2421 		    || (cpu_context(cpu, gpa_mm) ^ asid_cache(cpu)) &
2422 						asid_version_mask(cpu))
2423 			get_new_mmu_context(gpa_mm, cpu);
2424 	}
2425 }
2426 
2427 static int kvm_vz_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2428 {
2429 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2430 	bool migrated, all;
2431 
2432 	/*
2433 	 * Have we migrated to a different CPU?
2434 	 * If so, any old guest TLB state may be stale.
2435 	 */
2436 	migrated = (vcpu->arch.last_sched_cpu != cpu);
2437 
2438 	/*
2439 	 * Was this the last VCPU to run on this CPU?
2440 	 * If not, any old guest state from this VCPU will have been clobbered.
2441 	 */
2442 	all = migrated || (last_vcpu[cpu] != vcpu);
2443 	last_vcpu[cpu] = vcpu;
2444 
2445 	/*
2446 	 * Restore CP0_Wired unconditionally as we clear it after use, and
2447 	 * restore wired guest TLB entries (while in guest context).
2448 	 */
2449 	kvm_restore_gc0_wired(cop0);
2450 	if (current->flags & PF_VCPU) {
2451 		tlbw_use_hazard();
2452 		kvm_vz_vcpu_load_tlb(vcpu, cpu);
2453 		kvm_vz_vcpu_load_wired(vcpu);
2454 	}
2455 
2456 	/*
2457 	 * Restore timer state regardless, as e.g. Cause.TI can change over time
2458 	 * if left unmaintained.
2459 	 */
2460 	kvm_vz_restore_timer(vcpu);
2461 
2462 	/* Don't bother restoring registers multiple times unless necessary */
2463 	if (!all)
2464 		return 0;
2465 
2466 	/*
2467 	 * Restore config registers first, as some implementations restrict
2468 	 * writes to other registers when the corresponding feature bits aren't
2469 	 * set. For example Status.CU1 cannot be set unless Config1.FP is set.
2470 	 */
2471 	kvm_restore_gc0_config(cop0);
2472 	if (cpu_guest_has_conf1)
2473 		kvm_restore_gc0_config1(cop0);
2474 	if (cpu_guest_has_conf2)
2475 		kvm_restore_gc0_config2(cop0);
2476 	if (cpu_guest_has_conf3)
2477 		kvm_restore_gc0_config3(cop0);
2478 	if (cpu_guest_has_conf4)
2479 		kvm_restore_gc0_config4(cop0);
2480 	if (cpu_guest_has_conf5)
2481 		kvm_restore_gc0_config5(cop0);
2482 	if (cpu_guest_has_conf6)
2483 		kvm_restore_gc0_config6(cop0);
2484 	if (cpu_guest_has_conf7)
2485 		kvm_restore_gc0_config7(cop0);
2486 
2487 	kvm_restore_gc0_index(cop0);
2488 	kvm_restore_gc0_entrylo0(cop0);
2489 	kvm_restore_gc0_entrylo1(cop0);
2490 	kvm_restore_gc0_context(cop0);
2491 	if (cpu_guest_has_contextconfig)
2492 		kvm_restore_gc0_contextconfig(cop0);
2493 #ifdef CONFIG_64BIT
2494 	kvm_restore_gc0_xcontext(cop0);
2495 	if (cpu_guest_has_contextconfig)
2496 		kvm_restore_gc0_xcontextconfig(cop0);
2497 #endif
2498 	kvm_restore_gc0_pagemask(cop0);
2499 	kvm_restore_gc0_pagegrain(cop0);
2500 	kvm_restore_gc0_hwrena(cop0);
2501 	kvm_restore_gc0_badvaddr(cop0);
2502 	kvm_restore_gc0_entryhi(cop0);
2503 	kvm_restore_gc0_status(cop0);
2504 	kvm_restore_gc0_intctl(cop0);
2505 	kvm_restore_gc0_epc(cop0);
2506 	kvm_vz_write_gc0_ebase(kvm_read_sw_gc0_ebase(cop0));
2507 	if (cpu_guest_has_userlocal)
2508 		kvm_restore_gc0_userlocal(cop0);
2509 
2510 	kvm_restore_gc0_errorepc(cop0);
2511 
2512 	/* restore KScratch registers if enabled in guest */
2513 	if (cpu_guest_has_conf4) {
2514 		if (cpu_guest_has_kscr(2))
2515 			kvm_restore_gc0_kscratch1(cop0);
2516 		if (cpu_guest_has_kscr(3))
2517 			kvm_restore_gc0_kscratch2(cop0);
2518 		if (cpu_guest_has_kscr(4))
2519 			kvm_restore_gc0_kscratch3(cop0);
2520 		if (cpu_guest_has_kscr(5))
2521 			kvm_restore_gc0_kscratch4(cop0);
2522 		if (cpu_guest_has_kscr(6))
2523 			kvm_restore_gc0_kscratch5(cop0);
2524 		if (cpu_guest_has_kscr(7))
2525 			kvm_restore_gc0_kscratch6(cop0);
2526 	}
2527 
2528 	if (cpu_guest_has_badinstr)
2529 		kvm_restore_gc0_badinstr(cop0);
2530 	if (cpu_guest_has_badinstrp)
2531 		kvm_restore_gc0_badinstrp(cop0);
2532 
2533 	if (cpu_guest_has_segments) {
2534 		kvm_restore_gc0_segctl0(cop0);
2535 		kvm_restore_gc0_segctl1(cop0);
2536 		kvm_restore_gc0_segctl2(cop0);
2537 	}
2538 
2539 	/* restore HTW registers */
2540 	if (cpu_guest_has_htw) {
2541 		kvm_restore_gc0_pwbase(cop0);
2542 		kvm_restore_gc0_pwfield(cop0);
2543 		kvm_restore_gc0_pwsize(cop0);
2544 		kvm_restore_gc0_pwctl(cop0);
2545 	}
2546 
2547 	/* restore Root.GuestCtl2 from unused Guest guestctl2 register */
2548 	if (cpu_has_guestctl2)
2549 		write_c0_guestctl2(
2550 			cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL]);
2551 
2552 	/*
2553 	 * We should clear linked load bit to break interrupted atomics. This
2554 	 * prevents a SC on the next VCPU from succeeding by matching a LL on
2555 	 * the previous VCPU.
2556 	 */
2557 	if (cpu_guest_has_rw_llb)
2558 		write_gc0_lladdr(0);
2559 
2560 	return 0;
2561 }
2562 
2563 static int kvm_vz_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
2564 {
2565 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2566 
2567 	if (current->flags & PF_VCPU)
2568 		kvm_vz_vcpu_save_wired(vcpu);
2569 
2570 	kvm_lose_fpu(vcpu);
2571 
2572 	kvm_save_gc0_index(cop0);
2573 	kvm_save_gc0_entrylo0(cop0);
2574 	kvm_save_gc0_entrylo1(cop0);
2575 	kvm_save_gc0_context(cop0);
2576 	if (cpu_guest_has_contextconfig)
2577 		kvm_save_gc0_contextconfig(cop0);
2578 #ifdef CONFIG_64BIT
2579 	kvm_save_gc0_xcontext(cop0);
2580 	if (cpu_guest_has_contextconfig)
2581 		kvm_save_gc0_xcontextconfig(cop0);
2582 #endif
2583 	kvm_save_gc0_pagemask(cop0);
2584 	kvm_save_gc0_pagegrain(cop0);
2585 	kvm_save_gc0_wired(cop0);
2586 	/* allow wired TLB entries to be overwritten */
2587 	clear_gc0_wired(MIPSR6_WIRED_WIRED);
2588 	kvm_save_gc0_hwrena(cop0);
2589 	kvm_save_gc0_badvaddr(cop0);
2590 	kvm_save_gc0_entryhi(cop0);
2591 	kvm_save_gc0_status(cop0);
2592 	kvm_save_gc0_intctl(cop0);
2593 	kvm_save_gc0_epc(cop0);
2594 	kvm_write_sw_gc0_ebase(cop0, kvm_vz_read_gc0_ebase());
2595 	if (cpu_guest_has_userlocal)
2596 		kvm_save_gc0_userlocal(cop0);
2597 
2598 	/* only save implemented config registers */
2599 	kvm_save_gc0_config(cop0);
2600 	if (cpu_guest_has_conf1)
2601 		kvm_save_gc0_config1(cop0);
2602 	if (cpu_guest_has_conf2)
2603 		kvm_save_gc0_config2(cop0);
2604 	if (cpu_guest_has_conf3)
2605 		kvm_save_gc0_config3(cop0);
2606 	if (cpu_guest_has_conf4)
2607 		kvm_save_gc0_config4(cop0);
2608 	if (cpu_guest_has_conf5)
2609 		kvm_save_gc0_config5(cop0);
2610 	if (cpu_guest_has_conf6)
2611 		kvm_save_gc0_config6(cop0);
2612 	if (cpu_guest_has_conf7)
2613 		kvm_save_gc0_config7(cop0);
2614 
2615 	kvm_save_gc0_errorepc(cop0);
2616 
2617 	/* save KScratch registers if enabled in guest */
2618 	if (cpu_guest_has_conf4) {
2619 		if (cpu_guest_has_kscr(2))
2620 			kvm_save_gc0_kscratch1(cop0);
2621 		if (cpu_guest_has_kscr(3))
2622 			kvm_save_gc0_kscratch2(cop0);
2623 		if (cpu_guest_has_kscr(4))
2624 			kvm_save_gc0_kscratch3(cop0);
2625 		if (cpu_guest_has_kscr(5))
2626 			kvm_save_gc0_kscratch4(cop0);
2627 		if (cpu_guest_has_kscr(6))
2628 			kvm_save_gc0_kscratch5(cop0);
2629 		if (cpu_guest_has_kscr(7))
2630 			kvm_save_gc0_kscratch6(cop0);
2631 	}
2632 
2633 	if (cpu_guest_has_badinstr)
2634 		kvm_save_gc0_badinstr(cop0);
2635 	if (cpu_guest_has_badinstrp)
2636 		kvm_save_gc0_badinstrp(cop0);
2637 
2638 	if (cpu_guest_has_segments) {
2639 		kvm_save_gc0_segctl0(cop0);
2640 		kvm_save_gc0_segctl1(cop0);
2641 		kvm_save_gc0_segctl2(cop0);
2642 	}
2643 
2644 	/* save HTW registers if enabled in guest */
2645 	if (cpu_guest_has_htw &&
2646 	    kvm_read_sw_gc0_config3(cop0) & MIPS_CONF3_PW) {
2647 		kvm_save_gc0_pwbase(cop0);
2648 		kvm_save_gc0_pwfield(cop0);
2649 		kvm_save_gc0_pwsize(cop0);
2650 		kvm_save_gc0_pwctl(cop0);
2651 	}
2652 
2653 	kvm_vz_save_timer(vcpu);
2654 
2655 	/* save Root.GuestCtl2 in unused Guest guestctl2 register */
2656 	if (cpu_has_guestctl2)
2657 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] =
2658 			read_c0_guestctl2();
2659 
2660 	return 0;
2661 }
2662 
2663 /**
2664  * kvm_vz_resize_guest_vtlb() - Attempt to resize guest VTLB.
2665  * @size:	Number of guest VTLB entries (0 < @size <= root VTLB entries).
2666  *
2667  * Attempt to resize the guest VTLB by writing guest Config registers. This is
2668  * necessary for cores with a shared root/guest TLB to avoid overlap with wired
2669  * entries in the root VTLB.
2670  *
2671  * Returns:	The resulting guest VTLB size.
2672  */
2673 static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
2674 {
2675 	unsigned int config4 = 0, ret = 0, limit;
2676 
2677 	/* Write MMUSize - 1 into guest Config registers */
2678 	if (cpu_guest_has_conf1)
2679 		change_gc0_config1(MIPS_CONF1_TLBS,
2680 				   (size - 1) << MIPS_CONF1_TLBS_SHIFT);
2681 	if (cpu_guest_has_conf4) {
2682 		config4 = read_gc0_config4();
2683 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2684 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT) {
2685 			config4 &= ~MIPS_CONF4_VTLBSIZEEXT;
2686 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2687 				MIPS_CONF4_VTLBSIZEEXT_SHIFT;
2688 		} else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2689 			   MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT) {
2690 			config4 &= ~MIPS_CONF4_MMUSIZEEXT;
2691 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2692 				MIPS_CONF4_MMUSIZEEXT_SHIFT;
2693 		}
2694 		write_gc0_config4(config4);
2695 	}
2696 
2697 	/*
2698 	 * Set Guest.Wired.Limit = 0 (no limit up to Guest.MMUSize-1), unless it
2699 	 * would exceed Root.Wired.Limit (clearing Guest.Wired.Wired so write
2700 	 * not dropped)
2701 	 */
2702 	if (cpu_has_mips_r6) {
2703 		limit = (read_c0_wired() & MIPSR6_WIRED_LIMIT) >>
2704 						MIPSR6_WIRED_LIMIT_SHIFT;
2705 		if (size - 1 <= limit)
2706 			limit = 0;
2707 		write_gc0_wired(limit << MIPSR6_WIRED_LIMIT_SHIFT);
2708 	}
2709 
2710 	/* Read back MMUSize - 1 */
2711 	back_to_back_c0_hazard();
2712 	if (cpu_guest_has_conf1)
2713 		ret = (read_gc0_config1() & MIPS_CONF1_TLBS) >>
2714 						MIPS_CONF1_TLBS_SHIFT;
2715 	if (config4) {
2716 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2717 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT)
2718 			ret |= ((config4 & MIPS_CONF4_VTLBSIZEEXT) >>
2719 				MIPS_CONF4_VTLBSIZEEXT_SHIFT) <<
2720 				MIPS_CONF1_TLBS_SIZE;
2721 		else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2722 			 MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT)
2723 			ret |= ((config4 & MIPS_CONF4_MMUSIZEEXT) >>
2724 				MIPS_CONF4_MMUSIZEEXT_SHIFT) <<
2725 				MIPS_CONF1_TLBS_SIZE;
2726 	}
2727 	return ret + 1;
2728 }
2729 
2730 static int kvm_vz_hardware_enable(void)
2731 {
2732 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
2733 
2734 	/*
2735 	 * ImgTec cores tend to use a shared root/guest TLB. To avoid overlap of
2736 	 * root wired and guest entries, the guest TLB may need resizing.
2737 	 */
2738 	mmu_size = current_cpu_data.tlbsizevtlb;
2739 	ftlb_size = current_cpu_data.tlbsize - mmu_size;
2740 
2741 	/* Try switching to maximum guest VTLB size for flush */
2742 	guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
2743 	current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2744 	kvm_vz_local_flush_guesttlb_all();
2745 
2746 	/*
2747 	 * Reduce to make space for root wired entries and at least 2 root
2748 	 * non-wired entries. This does assume that long-term wired entries
2749 	 * won't be added later.
2750 	 */
2751 	guest_mmu_size = mmu_size - num_wired_entries() - 2;
2752 	guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
2753 	current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2754 
2755 	/*
2756 	 * Write the VTLB size, but if another CPU has already written, check it
2757 	 * matches or we won't provide a consistent view to the guest. If this
2758 	 * ever happens it suggests an asymmetric number of wired entries.
2759 	 */
2760 	if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
2761 	    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
2762 		 "Available guest VTLB size mismatch"))
2763 		return -EINVAL;
2764 
2765 	/*
2766 	 * Enable virtualization features granting guest direct control of
2767 	 * certain features:
2768 	 * CP0=1:	Guest coprocessor 0 context.
2769 	 * AT=Guest:	Guest MMU.
2770 	 * CG=1:	Hit (virtual address) CACHE operations (optional).
2771 	 * CF=1:	Guest Config registers.
2772 	 * CGI=1:	Indexed flush CACHE operations (optional).
2773 	 */
2774 	write_c0_guestctl0(MIPS_GCTL0_CP0 |
2775 			   (MIPS_GCTL0_AT_GUEST << MIPS_GCTL0_AT_SHIFT) |
2776 			   MIPS_GCTL0_CG | MIPS_GCTL0_CF);
2777 	if (cpu_has_guestctl0ext)
2778 		set_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2779 
2780 	if (cpu_has_guestid) {
2781 		write_c0_guestctl1(0);
2782 		kvm_vz_local_flush_roottlb_all_guests();
2783 
2784 		GUESTID_MASK = current_cpu_data.guestid_mask;
2785 		GUESTID_FIRST_VERSION = GUESTID_MASK + 1;
2786 		GUESTID_VERSION_MASK = ~GUESTID_MASK;
2787 
2788 		current_cpu_data.guestid_cache = GUESTID_FIRST_VERSION;
2789 	}
2790 
2791 	/* clear any pending injected virtual guest interrupts */
2792 	if (cpu_has_guestctl2)
2793 		clear_c0_guestctl2(0x3f << 10);
2794 
2795 	return 0;
2796 }
2797 
2798 static void kvm_vz_hardware_disable(void)
2799 {
2800 	kvm_vz_local_flush_guesttlb_all();
2801 
2802 	if (cpu_has_guestid) {
2803 		write_c0_guestctl1(0);
2804 		kvm_vz_local_flush_roottlb_all_guests();
2805 	}
2806 }
2807 
2808 static int kvm_vz_check_extension(struct kvm *kvm, long ext)
2809 {
2810 	int r;
2811 
2812 	switch (ext) {
2813 	case KVM_CAP_MIPS_VZ:
2814 		/* we wouldn't be here unless cpu_has_vz */
2815 		r = 1;
2816 		break;
2817 #ifdef CONFIG_64BIT
2818 	case KVM_CAP_MIPS_64BIT:
2819 		/* We support 64-bit registers/operations and addresses */
2820 		r = 2;
2821 		break;
2822 #endif
2823 	default:
2824 		r = 0;
2825 		break;
2826 	}
2827 
2828 	return r;
2829 }
2830 
2831 static int kvm_vz_vcpu_init(struct kvm_vcpu *vcpu)
2832 {
2833 	int i;
2834 
2835 	for_each_possible_cpu(i)
2836 		vcpu->arch.vzguestid[i] = 0;
2837 
2838 	return 0;
2839 }
2840 
2841 static void kvm_vz_vcpu_uninit(struct kvm_vcpu *vcpu)
2842 {
2843 	int cpu;
2844 
2845 	/*
2846 	 * If the VCPU is freed and reused as another VCPU, we don't want the
2847 	 * matching pointer wrongly hanging around in last_vcpu[] or
2848 	 * last_exec_vcpu[].
2849 	 */
2850 	for_each_possible_cpu(cpu) {
2851 		if (last_vcpu[cpu] == vcpu)
2852 			last_vcpu[cpu] = NULL;
2853 		if (last_exec_vcpu[cpu] == vcpu)
2854 			last_exec_vcpu[cpu] = NULL;
2855 	}
2856 }
2857 
2858 static int kvm_vz_vcpu_setup(struct kvm_vcpu *vcpu)
2859 {
2860 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2861 	unsigned long count_hz = 100*1000*1000; /* default to 100 MHz */
2862 
2863 	/*
2864 	 * Start off the timer at the same frequency as the host timer, but the
2865 	 * soft timer doesn't handle frequencies greater than 1GHz yet.
2866 	 */
2867 	if (mips_hpt_frequency && mips_hpt_frequency <= NSEC_PER_SEC)
2868 		count_hz = mips_hpt_frequency;
2869 	kvm_mips_init_count(vcpu, count_hz);
2870 
2871 	/*
2872 	 * Initialize guest register state to valid architectural reset state.
2873 	 */
2874 
2875 	/* PageGrain */
2876 	if (cpu_has_mips_r6)
2877 		kvm_write_sw_gc0_pagegrain(cop0, PG_RIE | PG_XIE | PG_IEC);
2878 	/* Wired */
2879 	if (cpu_has_mips_r6)
2880 		kvm_write_sw_gc0_wired(cop0,
2881 				       read_gc0_wired() & MIPSR6_WIRED_LIMIT);
2882 	/* Status */
2883 	kvm_write_sw_gc0_status(cop0, ST0_BEV | ST0_ERL);
2884 	if (cpu_has_mips_r6)
2885 		kvm_change_sw_gc0_status(cop0, ST0_FR, read_gc0_status());
2886 	/* IntCtl */
2887 	kvm_write_sw_gc0_intctl(cop0, read_gc0_intctl() &
2888 				(INTCTLF_IPFDC | INTCTLF_IPPCI | INTCTLF_IPTI));
2889 	/* PRId */
2890 	kvm_write_sw_gc0_prid(cop0, boot_cpu_data.processor_id);
2891 	/* EBase */
2892 	kvm_write_sw_gc0_ebase(cop0, (s32)0x80000000 | vcpu->vcpu_id);
2893 	/* Config */
2894 	kvm_save_gc0_config(cop0);
2895 	/* architecturally writable (e.g. from guest) */
2896 	kvm_change_sw_gc0_config(cop0, CONF_CM_CMASK,
2897 				 _page_cachable_default >> _CACHE_SHIFT);
2898 	/* architecturally read only, but maybe writable from root */
2899 	kvm_change_sw_gc0_config(cop0, MIPS_CONF_MT, read_c0_config());
2900 	if (cpu_guest_has_conf1) {
2901 		kvm_set_sw_gc0_config(cop0, MIPS_CONF_M);
2902 		/* Config1 */
2903 		kvm_save_gc0_config1(cop0);
2904 		/* architecturally read only, but maybe writable from root */
2905 		kvm_clear_sw_gc0_config1(cop0, MIPS_CONF1_C2	|
2906 					       MIPS_CONF1_MD	|
2907 					       MIPS_CONF1_PC	|
2908 					       MIPS_CONF1_WR	|
2909 					       MIPS_CONF1_CA	|
2910 					       MIPS_CONF1_FP);
2911 	}
2912 	if (cpu_guest_has_conf2) {
2913 		kvm_set_sw_gc0_config1(cop0, MIPS_CONF_M);
2914 		/* Config2 */
2915 		kvm_save_gc0_config2(cop0);
2916 	}
2917 	if (cpu_guest_has_conf3) {
2918 		kvm_set_sw_gc0_config2(cop0, MIPS_CONF_M);
2919 		/* Config3 */
2920 		kvm_save_gc0_config3(cop0);
2921 		/* architecturally writable (e.g. from guest) */
2922 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_ISA_OE);
2923 		/* architecturally read only, but maybe writable from root */
2924 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_MSA	|
2925 					       MIPS_CONF3_BPG	|
2926 					       MIPS_CONF3_ULRI	|
2927 					       MIPS_CONF3_DSP	|
2928 					       MIPS_CONF3_CTXTC	|
2929 					       MIPS_CONF3_ITL	|
2930 					       MIPS_CONF3_LPA	|
2931 					       MIPS_CONF3_VEIC	|
2932 					       MIPS_CONF3_VINT	|
2933 					       MIPS_CONF3_SP	|
2934 					       MIPS_CONF3_CDMM	|
2935 					       MIPS_CONF3_MT	|
2936 					       MIPS_CONF3_SM	|
2937 					       MIPS_CONF3_TL);
2938 	}
2939 	if (cpu_guest_has_conf4) {
2940 		kvm_set_sw_gc0_config3(cop0, MIPS_CONF_M);
2941 		/* Config4 */
2942 		kvm_save_gc0_config4(cop0);
2943 	}
2944 	if (cpu_guest_has_conf5) {
2945 		kvm_set_sw_gc0_config4(cop0, MIPS_CONF_M);
2946 		/* Config5 */
2947 		kvm_save_gc0_config5(cop0);
2948 		/* architecturally writable (e.g. from guest) */
2949 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_K	|
2950 					       MIPS_CONF5_CV	|
2951 					       MIPS_CONF5_MSAEN	|
2952 					       MIPS_CONF5_UFE	|
2953 					       MIPS_CONF5_FRE	|
2954 					       MIPS_CONF5_SBRI	|
2955 					       MIPS_CONF5_UFR);
2956 		/* architecturally read only, but maybe writable from root */
2957 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_MRP);
2958 	}
2959 
2960 	if (cpu_guest_has_contextconfig) {
2961 		/* ContextConfig */
2962 		kvm_write_sw_gc0_contextconfig(cop0, 0x007ffff0);
2963 #ifdef CONFIG_64BIT
2964 		/* XContextConfig */
2965 		/* bits SEGBITS-13+3:4 set */
2966 		kvm_write_sw_gc0_xcontextconfig(cop0,
2967 					((1ull << (cpu_vmbits - 13)) - 1) << 4);
2968 #endif
2969 	}
2970 
2971 	/* Implementation dependent, use the legacy layout */
2972 	if (cpu_guest_has_segments) {
2973 		/* SegCtl0, SegCtl1, SegCtl2 */
2974 		kvm_write_sw_gc0_segctl0(cop0, 0x00200010);
2975 		kvm_write_sw_gc0_segctl1(cop0, 0x00000002 |
2976 				(_page_cachable_default >> _CACHE_SHIFT) <<
2977 						(16 + MIPS_SEGCFG_C_SHIFT));
2978 		kvm_write_sw_gc0_segctl2(cop0, 0x00380438);
2979 	}
2980 
2981 	/* reset HTW registers */
2982 	if (cpu_guest_has_htw && cpu_has_mips_r6) {
2983 		/* PWField */
2984 		kvm_write_sw_gc0_pwfield(cop0, 0x0c30c302);
2985 		/* PWSize */
2986 		kvm_write_sw_gc0_pwsize(cop0, 1 << MIPS_PWSIZE_PTW_SHIFT);
2987 	}
2988 
2989 	/* start with no pending virtual guest interrupts */
2990 	if (cpu_has_guestctl2)
2991 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] = 0;
2992 
2993 	/* Put PC at reset vector */
2994 	vcpu->arch.pc = CKSEG1ADDR(0x1fc00000);
2995 
2996 	return 0;
2997 }
2998 
2999 static void kvm_vz_flush_shadow_all(struct kvm *kvm)
3000 {
3001 	if (cpu_has_guestid) {
3002 		/* Flush GuestID for each VCPU individually */
3003 		kvm_flush_remote_tlbs(kvm);
3004 	} else {
3005 		/*
3006 		 * For each CPU there is a single GPA ASID used by all VCPUs in
3007 		 * the VM, so it doesn't make sense for the VCPUs to handle
3008 		 * invalidation of these ASIDs individually.
3009 		 *
3010 		 * Instead mark all CPUs as needing ASID invalidation in
3011 		 * asid_flush_mask, and just use kvm_flush_remote_tlbs(kvm) to
3012 		 * kick any running VCPUs so they check asid_flush_mask.
3013 		 */
3014 		cpumask_setall(&kvm->arch.asid_flush_mask);
3015 		kvm_flush_remote_tlbs(kvm);
3016 	}
3017 }
3018 
3019 static void kvm_vz_flush_shadow_memslot(struct kvm *kvm,
3020 					const struct kvm_memory_slot *slot)
3021 {
3022 	kvm_vz_flush_shadow_all(kvm);
3023 }
3024 
3025 static void kvm_vz_vcpu_reenter(struct kvm_run *run, struct kvm_vcpu *vcpu)
3026 {
3027 	int cpu = smp_processor_id();
3028 	int preserve_guest_tlb;
3029 
3030 	preserve_guest_tlb = kvm_vz_check_requests(vcpu, cpu);
3031 
3032 	if (preserve_guest_tlb)
3033 		kvm_vz_vcpu_save_wired(vcpu);
3034 
3035 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3036 
3037 	if (preserve_guest_tlb)
3038 		kvm_vz_vcpu_load_wired(vcpu);
3039 }
3040 
3041 static int kvm_vz_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
3042 {
3043 	int cpu = smp_processor_id();
3044 	int r;
3045 
3046 	kvm_vz_acquire_htimer(vcpu);
3047 	/* Check if we have any exceptions/interrupts pending */
3048 	kvm_mips_deliver_interrupts(vcpu, read_gc0_cause());
3049 
3050 	kvm_vz_check_requests(vcpu, cpu);
3051 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3052 	kvm_vz_vcpu_load_wired(vcpu);
3053 
3054 	r = vcpu->arch.vcpu_run(run, vcpu);
3055 
3056 	kvm_vz_vcpu_save_wired(vcpu);
3057 
3058 	return r;
3059 }
3060 
3061 static struct kvm_mips_callbacks kvm_vz_callbacks = {
3062 	.handle_cop_unusable = kvm_trap_vz_handle_cop_unusable,
3063 	.handle_tlb_mod = kvm_trap_vz_handle_tlb_st_miss,
3064 	.handle_tlb_ld_miss = kvm_trap_vz_handle_tlb_ld_miss,
3065 	.handle_tlb_st_miss = kvm_trap_vz_handle_tlb_st_miss,
3066 	.handle_addr_err_st = kvm_trap_vz_no_handler,
3067 	.handle_addr_err_ld = kvm_trap_vz_no_handler,
3068 	.handle_syscall = kvm_trap_vz_no_handler,
3069 	.handle_res_inst = kvm_trap_vz_no_handler,
3070 	.handle_break = kvm_trap_vz_no_handler,
3071 	.handle_msa_disabled = kvm_trap_vz_handle_msa_disabled,
3072 	.handle_guest_exit = kvm_trap_vz_handle_guest_exit,
3073 
3074 	.hardware_enable = kvm_vz_hardware_enable,
3075 	.hardware_disable = kvm_vz_hardware_disable,
3076 	.check_extension = kvm_vz_check_extension,
3077 	.vcpu_init = kvm_vz_vcpu_init,
3078 	.vcpu_uninit = kvm_vz_vcpu_uninit,
3079 	.vcpu_setup = kvm_vz_vcpu_setup,
3080 	.flush_shadow_all = kvm_vz_flush_shadow_all,
3081 	.flush_shadow_memslot = kvm_vz_flush_shadow_memslot,
3082 	.gva_to_gpa = kvm_vz_gva_to_gpa_cb,
3083 	.queue_timer_int = kvm_vz_queue_timer_int_cb,
3084 	.dequeue_timer_int = kvm_vz_dequeue_timer_int_cb,
3085 	.queue_io_int = kvm_vz_queue_io_int_cb,
3086 	.dequeue_io_int = kvm_vz_dequeue_io_int_cb,
3087 	.irq_deliver = kvm_vz_irq_deliver_cb,
3088 	.irq_clear = kvm_vz_irq_clear_cb,
3089 	.num_regs = kvm_vz_num_regs,
3090 	.copy_reg_indices = kvm_vz_copy_reg_indices,
3091 	.get_one_reg = kvm_vz_get_one_reg,
3092 	.set_one_reg = kvm_vz_set_one_reg,
3093 	.vcpu_load = kvm_vz_vcpu_load,
3094 	.vcpu_put = kvm_vz_vcpu_put,
3095 	.vcpu_run = kvm_vz_vcpu_run,
3096 	.vcpu_reenter = kvm_vz_vcpu_reenter,
3097 };
3098 
3099 int kvm_mips_emulation_init(struct kvm_mips_callbacks **install_callbacks)
3100 {
3101 	if (!cpu_has_vz)
3102 		return -ENODEV;
3103 
3104 	/*
3105 	 * VZ requires at least 2 KScratch registers, so it should have been
3106 	 * possible to allocate pgd_reg.
3107 	 */
3108 	if (WARN(pgd_reg == -1,
3109 		 "pgd_reg not allocated even though cpu_has_vz\n"))
3110 		return -ENODEV;
3111 
3112 	pr_info("Starting KVM with MIPS VZ extensions\n");
3113 
3114 	*install_callbacks = &kvm_vz_callbacks;
3115 	return 0;
3116 }
3117