xref: /openbmc/linux/arch/mips/kvm/vz.c (revision 3213486f)
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 	case Hit_Invalidate_I:
1109 	case Hit_Invalidate_D:
1110 	case Hit_Writeback_Inv_D:
1111 		if (boot_cpu_type() == CPU_CAVIUM_OCTEON3) {
1112 			/* We can just flush entire icache */
1113 			local_flush_icache_range(0, 0);
1114 			return EMULATE_DONE;
1115 		}
1116 
1117 		/* So far, other platforms support guest hit cache ops */
1118 		break;
1119 	default:
1120 		break;
1121 	};
1122 
1123 	kvm_err("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1124 		curr_pc, vcpu->arch.gprs[31], cache, op, base, arch->gprs[base],
1125 		offset);
1126 	/* Rollback PC */
1127 	vcpu->arch.pc = curr_pc;
1128 
1129 	return EMULATE_FAIL;
1130 }
1131 
1132 static enum emulation_result kvm_trap_vz_handle_gpsi(u32 cause, u32 *opc,
1133 						     struct kvm_vcpu *vcpu)
1134 {
1135 	enum emulation_result er = EMULATE_DONE;
1136 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1137 	struct kvm_run *run = vcpu->run;
1138 	union mips_instruction inst;
1139 	int rd, rt, sel;
1140 	int err;
1141 
1142 	/*
1143 	 *  Fetch the instruction.
1144 	 */
1145 	if (cause & CAUSEF_BD)
1146 		opc += 1;
1147 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1148 	if (err)
1149 		return EMULATE_FAIL;
1150 
1151 	switch (inst.r_format.opcode) {
1152 	case cop0_op:
1153 		er = kvm_vz_gpsi_cop0(inst, opc, cause, run, vcpu);
1154 		break;
1155 #ifndef CONFIG_CPU_MIPSR6
1156 	case cache_op:
1157 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1158 		er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
1159 		break;
1160 #endif
1161 	case spec3_op:
1162 		switch (inst.spec3_format.func) {
1163 #ifdef CONFIG_CPU_MIPSR6
1164 		case cache6_op:
1165 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1166 			er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
1167 			break;
1168 #endif
1169 		case rdhwr_op:
1170 			if (inst.r_format.rs || (inst.r_format.re >> 3))
1171 				goto unknown;
1172 
1173 			rd = inst.r_format.rd;
1174 			rt = inst.r_format.rt;
1175 			sel = inst.r_format.re & 0x7;
1176 
1177 			switch (rd) {
1178 			case MIPS_HWR_CC:	/* Read count register */
1179 				arch->gprs[rt] =
1180 					(long)(int)kvm_mips_read_count(vcpu);
1181 				break;
1182 			default:
1183 				trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1184 					      KVM_TRACE_HWR(rd, sel), 0);
1185 				goto unknown;
1186 			};
1187 
1188 			trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1189 				      KVM_TRACE_HWR(rd, sel), arch->gprs[rt]);
1190 
1191 			er = update_pc(vcpu, cause);
1192 			break;
1193 		default:
1194 			goto unknown;
1195 		};
1196 		break;
1197 unknown:
1198 
1199 	default:
1200 		kvm_err("GPSI exception not supported (%p/%#x)\n",
1201 				opc, inst.word);
1202 		kvm_arch_vcpu_dump_regs(vcpu);
1203 		er = EMULATE_FAIL;
1204 		break;
1205 	}
1206 
1207 	return er;
1208 }
1209 
1210 static enum emulation_result kvm_trap_vz_handle_gsfc(u32 cause, u32 *opc,
1211 						     struct kvm_vcpu *vcpu)
1212 {
1213 	enum emulation_result er = EMULATE_DONE;
1214 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1215 	union mips_instruction inst;
1216 	int err;
1217 
1218 	/*
1219 	 *  Fetch the instruction.
1220 	 */
1221 	if (cause & CAUSEF_BD)
1222 		opc += 1;
1223 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1224 	if (err)
1225 		return EMULATE_FAIL;
1226 
1227 	/* complete MTC0 on behalf of guest and advance EPC */
1228 	if (inst.c0r_format.opcode == cop0_op &&
1229 	    inst.c0r_format.rs == mtc_op &&
1230 	    inst.c0r_format.z == 0) {
1231 		int rt = inst.c0r_format.rt;
1232 		int rd = inst.c0r_format.rd;
1233 		int sel = inst.c0r_format.sel;
1234 		unsigned int val = arch->gprs[rt];
1235 		unsigned int old_val, change;
1236 
1237 		trace_kvm_hwr(vcpu, KVM_TRACE_MTC0, KVM_TRACE_COP0(rd, sel),
1238 			      val);
1239 
1240 		if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1241 			/* FR bit should read as zero if no FPU */
1242 			if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1243 				val &= ~(ST0_CU1 | ST0_FR);
1244 
1245 			/*
1246 			 * Also don't allow FR to be set if host doesn't support
1247 			 * it.
1248 			 */
1249 			if (!(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
1250 				val &= ~ST0_FR;
1251 
1252 			old_val = read_gc0_status();
1253 			change = val ^ old_val;
1254 
1255 			if (change & ST0_FR) {
1256 				/*
1257 				 * FPU and Vector register state is made
1258 				 * UNPREDICTABLE by a change of FR, so don't
1259 				 * even bother saving it.
1260 				 */
1261 				kvm_drop_fpu(vcpu);
1262 			}
1263 
1264 			/*
1265 			 * If MSA state is already live, it is undefined how it
1266 			 * interacts with FR=0 FPU state, and we don't want to
1267 			 * hit reserved instruction exceptions trying to save
1268 			 * the MSA state later when CU=1 && FR=1, so play it
1269 			 * safe and save it first.
1270 			 */
1271 			if (change & ST0_CU1 && !(val & ST0_FR) &&
1272 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1273 				kvm_lose_fpu(vcpu);
1274 
1275 			write_gc0_status(val);
1276 		} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1277 			u32 old_cause = read_gc0_cause();
1278 			u32 change = old_cause ^ val;
1279 
1280 			/* DC bit enabling/disabling timer? */
1281 			if (change & CAUSEF_DC) {
1282 				if (val & CAUSEF_DC) {
1283 					kvm_vz_lose_htimer(vcpu);
1284 					kvm_mips_count_disable_cause(vcpu);
1285 				} else {
1286 					kvm_mips_count_enable_cause(vcpu);
1287 				}
1288 			}
1289 
1290 			/* Only certain bits are RW to the guest */
1291 			change &= (CAUSEF_DC | CAUSEF_IV | CAUSEF_WP |
1292 				   CAUSEF_IP0 | CAUSEF_IP1);
1293 
1294 			/* WP can only be cleared */
1295 			change &= ~CAUSEF_WP | old_cause;
1296 
1297 			write_gc0_cause(old_cause ^ change);
1298 		} else if ((rd == MIPS_CP0_STATUS) && (sel == 1)) { /* IntCtl */
1299 			write_gc0_intctl(val);
1300 		} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1301 			old_val = read_gc0_config5();
1302 			change = val ^ old_val;
1303 			/* Handle changes in FPU/MSA modes */
1304 			preempt_disable();
1305 
1306 			/*
1307 			 * Propagate FRE changes immediately if the FPU
1308 			 * context is already loaded.
1309 			 */
1310 			if (change & MIPS_CONF5_FRE &&
1311 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1312 				change_c0_config5(MIPS_CONF5_FRE, val);
1313 
1314 			preempt_enable();
1315 
1316 			val = old_val ^
1317 				(change & kvm_vz_config5_guest_wrmask(vcpu));
1318 			write_gc0_config5(val);
1319 		} else {
1320 			kvm_err("Handle GSFC, unsupported field change @ %p: %#x\n",
1321 			    opc, inst.word);
1322 			er = EMULATE_FAIL;
1323 		}
1324 
1325 		if (er != EMULATE_FAIL)
1326 			er = update_pc(vcpu, cause);
1327 	} else {
1328 		kvm_err("Handle GSFC, unrecognized instruction @ %p: %#x\n",
1329 			opc, inst.word);
1330 		er = EMULATE_FAIL;
1331 	}
1332 
1333 	return er;
1334 }
1335 
1336 static enum emulation_result kvm_trap_vz_handle_ghfc(u32 cause, u32 *opc,
1337 						     struct kvm_vcpu *vcpu)
1338 {
1339 	/*
1340 	 * Presumably this is due to MC (guest mode change), so lets trace some
1341 	 * relevant info.
1342 	 */
1343 	trace_kvm_guest_mode_change(vcpu);
1344 
1345 	return EMULATE_DONE;
1346 }
1347 
1348 static enum emulation_result kvm_trap_vz_handle_hc(u32 cause, u32 *opc,
1349 						   struct kvm_vcpu *vcpu)
1350 {
1351 	enum emulation_result er;
1352 	union mips_instruction inst;
1353 	unsigned long curr_pc;
1354 	int err;
1355 
1356 	if (cause & CAUSEF_BD)
1357 		opc += 1;
1358 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1359 	if (err)
1360 		return EMULATE_FAIL;
1361 
1362 	/*
1363 	 * Update PC and hold onto current PC in case there is
1364 	 * an error and we want to rollback the PC
1365 	 */
1366 	curr_pc = vcpu->arch.pc;
1367 	er = update_pc(vcpu, cause);
1368 	if (er == EMULATE_FAIL)
1369 		return er;
1370 
1371 	er = kvm_mips_emul_hypcall(vcpu, inst);
1372 	if (er == EMULATE_FAIL)
1373 		vcpu->arch.pc = curr_pc;
1374 
1375 	return er;
1376 }
1377 
1378 static enum emulation_result kvm_trap_vz_no_handler_guest_exit(u32 gexccode,
1379 							u32 cause,
1380 							u32 *opc,
1381 							struct kvm_vcpu *vcpu)
1382 {
1383 	u32 inst;
1384 
1385 	/*
1386 	 *  Fetch the instruction.
1387 	 */
1388 	if (cause & CAUSEF_BD)
1389 		opc += 1;
1390 	kvm_get_badinstr(opc, vcpu, &inst);
1391 
1392 	kvm_err("Guest Exception Code: %d not yet handled @ PC: %p, inst: 0x%08x  Status: %#x\n",
1393 		gexccode, opc, inst, read_gc0_status());
1394 
1395 	return EMULATE_FAIL;
1396 }
1397 
1398 static int kvm_trap_vz_handle_guest_exit(struct kvm_vcpu *vcpu)
1399 {
1400 	u32 *opc = (u32 *) vcpu->arch.pc;
1401 	u32 cause = vcpu->arch.host_cp0_cause;
1402 	enum emulation_result er = EMULATE_DONE;
1403 	u32 gexccode = (vcpu->arch.host_cp0_guestctl0 &
1404 			MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
1405 	int ret = RESUME_GUEST;
1406 
1407 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_GEXCCODE_BASE + gexccode);
1408 	switch (gexccode) {
1409 	case MIPS_GCTL0_GEXC_GPSI:
1410 		++vcpu->stat.vz_gpsi_exits;
1411 		er = kvm_trap_vz_handle_gpsi(cause, opc, vcpu);
1412 		break;
1413 	case MIPS_GCTL0_GEXC_GSFC:
1414 		++vcpu->stat.vz_gsfc_exits;
1415 		er = kvm_trap_vz_handle_gsfc(cause, opc, vcpu);
1416 		break;
1417 	case MIPS_GCTL0_GEXC_HC:
1418 		++vcpu->stat.vz_hc_exits;
1419 		er = kvm_trap_vz_handle_hc(cause, opc, vcpu);
1420 		break;
1421 	case MIPS_GCTL0_GEXC_GRR:
1422 		++vcpu->stat.vz_grr_exits;
1423 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1424 						       vcpu);
1425 		break;
1426 	case MIPS_GCTL0_GEXC_GVA:
1427 		++vcpu->stat.vz_gva_exits;
1428 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1429 						       vcpu);
1430 		break;
1431 	case MIPS_GCTL0_GEXC_GHFC:
1432 		++vcpu->stat.vz_ghfc_exits;
1433 		er = kvm_trap_vz_handle_ghfc(cause, opc, vcpu);
1434 		break;
1435 	case MIPS_GCTL0_GEXC_GPA:
1436 		++vcpu->stat.vz_gpa_exits;
1437 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1438 						       vcpu);
1439 		break;
1440 	default:
1441 		++vcpu->stat.vz_resvd_exits;
1442 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1443 						       vcpu);
1444 		break;
1445 
1446 	}
1447 
1448 	if (er == EMULATE_DONE) {
1449 		ret = RESUME_GUEST;
1450 	} else if (er == EMULATE_HYPERCALL) {
1451 		ret = kvm_mips_handle_hypcall(vcpu);
1452 	} else {
1453 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1454 		ret = RESUME_HOST;
1455 	}
1456 	return ret;
1457 }
1458 
1459 /**
1460  * kvm_trap_vz_handle_cop_unusuable() - Guest used unusable coprocessor.
1461  * @vcpu:	Virtual CPU context.
1462  *
1463  * Handle when the guest attempts to use a coprocessor which hasn't been allowed
1464  * by the root context.
1465  */
1466 static int kvm_trap_vz_handle_cop_unusable(struct kvm_vcpu *vcpu)
1467 {
1468 	struct kvm_run *run = vcpu->run;
1469 	u32 cause = vcpu->arch.host_cp0_cause;
1470 	enum emulation_result er = EMULATE_FAIL;
1471 	int ret = RESUME_GUEST;
1472 
1473 	if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 1) {
1474 		/*
1475 		 * If guest FPU not present, the FPU operation should have been
1476 		 * treated as a reserved instruction!
1477 		 * If FPU already in use, we shouldn't get this at all.
1478 		 */
1479 		if (WARN_ON(!kvm_mips_guest_has_fpu(&vcpu->arch) ||
1480 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1481 			preempt_enable();
1482 			return EMULATE_FAIL;
1483 		}
1484 
1485 		kvm_own_fpu(vcpu);
1486 		er = EMULATE_DONE;
1487 	}
1488 	/* other coprocessors not handled */
1489 
1490 	switch (er) {
1491 	case EMULATE_DONE:
1492 		ret = RESUME_GUEST;
1493 		break;
1494 
1495 	case EMULATE_FAIL:
1496 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1497 		ret = RESUME_HOST;
1498 		break;
1499 
1500 	default:
1501 		BUG();
1502 	}
1503 	return ret;
1504 }
1505 
1506 /**
1507  * kvm_trap_vz_handle_msa_disabled() - Guest used MSA while disabled in root.
1508  * @vcpu:	Virtual CPU context.
1509  *
1510  * Handle when the guest attempts to use MSA when it is disabled in the root
1511  * context.
1512  */
1513 static int kvm_trap_vz_handle_msa_disabled(struct kvm_vcpu *vcpu)
1514 {
1515 	struct kvm_run *run = vcpu->run;
1516 
1517 	/*
1518 	 * If MSA not present or not exposed to guest or FR=0, the MSA operation
1519 	 * should have been treated as a reserved instruction!
1520 	 * Same if CU1=1, FR=0.
1521 	 * If MSA already in use, we shouldn't get this at all.
1522 	 */
1523 	if (!kvm_mips_guest_has_msa(&vcpu->arch) ||
1524 	    (read_gc0_status() & (ST0_CU1 | ST0_FR)) == ST0_CU1 ||
1525 	    !(read_gc0_config5() & MIPS_CONF5_MSAEN) ||
1526 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1527 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1528 		return RESUME_HOST;
1529 	}
1530 
1531 	kvm_own_msa(vcpu);
1532 
1533 	return RESUME_GUEST;
1534 }
1535 
1536 static int kvm_trap_vz_handle_tlb_ld_miss(struct kvm_vcpu *vcpu)
1537 {
1538 	struct kvm_run *run = vcpu->run;
1539 	u32 *opc = (u32 *) vcpu->arch.pc;
1540 	u32 cause = vcpu->arch.host_cp0_cause;
1541 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1542 	union mips_instruction inst;
1543 	enum emulation_result er = EMULATE_DONE;
1544 	int err, ret = RESUME_GUEST;
1545 
1546 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, false)) {
1547 		/* A code fetch fault doesn't count as an MMIO */
1548 		if (kvm_is_ifetch_fault(&vcpu->arch)) {
1549 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1550 			return RESUME_HOST;
1551 		}
1552 
1553 		/* Fetch the instruction */
1554 		if (cause & CAUSEF_BD)
1555 			opc += 1;
1556 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1557 		if (err) {
1558 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1559 			return RESUME_HOST;
1560 		}
1561 
1562 		/* Treat as MMIO */
1563 		er = kvm_mips_emulate_load(inst, cause, run, vcpu);
1564 		if (er == EMULATE_FAIL) {
1565 			kvm_err("Guest Emulate Load from MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1566 				opc, badvaddr);
1567 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1568 		}
1569 	}
1570 
1571 	if (er == EMULATE_DONE) {
1572 		ret = RESUME_GUEST;
1573 	} else if (er == EMULATE_DO_MMIO) {
1574 		run->exit_reason = KVM_EXIT_MMIO;
1575 		ret = RESUME_HOST;
1576 	} else {
1577 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1578 		ret = RESUME_HOST;
1579 	}
1580 	return ret;
1581 }
1582 
1583 static int kvm_trap_vz_handle_tlb_st_miss(struct kvm_vcpu *vcpu)
1584 {
1585 	struct kvm_run *run = vcpu->run;
1586 	u32 *opc = (u32 *) vcpu->arch.pc;
1587 	u32 cause = vcpu->arch.host_cp0_cause;
1588 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1589 	union mips_instruction inst;
1590 	enum emulation_result er = EMULATE_DONE;
1591 	int err;
1592 	int ret = RESUME_GUEST;
1593 
1594 	/* Just try the access again if we couldn't do the translation */
1595 	if (kvm_vz_badvaddr_to_gpa(vcpu, badvaddr, &badvaddr))
1596 		return RESUME_GUEST;
1597 	vcpu->arch.host_cp0_badvaddr = badvaddr;
1598 
1599 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, true)) {
1600 		/* Fetch the instruction */
1601 		if (cause & CAUSEF_BD)
1602 			opc += 1;
1603 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1604 		if (err) {
1605 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1606 			return RESUME_HOST;
1607 		}
1608 
1609 		/* Treat as MMIO */
1610 		er = kvm_mips_emulate_store(inst, cause, run, vcpu);
1611 		if (er == EMULATE_FAIL) {
1612 			kvm_err("Guest Emulate Store to MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1613 				opc, badvaddr);
1614 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1615 		}
1616 	}
1617 
1618 	if (er == EMULATE_DONE) {
1619 		ret = RESUME_GUEST;
1620 	} else if (er == EMULATE_DO_MMIO) {
1621 		run->exit_reason = KVM_EXIT_MMIO;
1622 		ret = RESUME_HOST;
1623 	} else {
1624 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1625 		ret = RESUME_HOST;
1626 	}
1627 	return ret;
1628 }
1629 
1630 static u64 kvm_vz_get_one_regs[] = {
1631 	KVM_REG_MIPS_CP0_INDEX,
1632 	KVM_REG_MIPS_CP0_ENTRYLO0,
1633 	KVM_REG_MIPS_CP0_ENTRYLO1,
1634 	KVM_REG_MIPS_CP0_CONTEXT,
1635 	KVM_REG_MIPS_CP0_PAGEMASK,
1636 	KVM_REG_MIPS_CP0_PAGEGRAIN,
1637 	KVM_REG_MIPS_CP0_WIRED,
1638 	KVM_REG_MIPS_CP0_HWRENA,
1639 	KVM_REG_MIPS_CP0_BADVADDR,
1640 	KVM_REG_MIPS_CP0_COUNT,
1641 	KVM_REG_MIPS_CP0_ENTRYHI,
1642 	KVM_REG_MIPS_CP0_COMPARE,
1643 	KVM_REG_MIPS_CP0_STATUS,
1644 	KVM_REG_MIPS_CP0_INTCTL,
1645 	KVM_REG_MIPS_CP0_CAUSE,
1646 	KVM_REG_MIPS_CP0_EPC,
1647 	KVM_REG_MIPS_CP0_PRID,
1648 	KVM_REG_MIPS_CP0_EBASE,
1649 	KVM_REG_MIPS_CP0_CONFIG,
1650 	KVM_REG_MIPS_CP0_CONFIG1,
1651 	KVM_REG_MIPS_CP0_CONFIG2,
1652 	KVM_REG_MIPS_CP0_CONFIG3,
1653 	KVM_REG_MIPS_CP0_CONFIG4,
1654 	KVM_REG_MIPS_CP0_CONFIG5,
1655 #ifdef CONFIG_64BIT
1656 	KVM_REG_MIPS_CP0_XCONTEXT,
1657 #endif
1658 	KVM_REG_MIPS_CP0_ERROREPC,
1659 
1660 	KVM_REG_MIPS_COUNT_CTL,
1661 	KVM_REG_MIPS_COUNT_RESUME,
1662 	KVM_REG_MIPS_COUNT_HZ,
1663 };
1664 
1665 static u64 kvm_vz_get_one_regs_contextconfig[] = {
1666 	KVM_REG_MIPS_CP0_CONTEXTCONFIG,
1667 #ifdef CONFIG_64BIT
1668 	KVM_REG_MIPS_CP0_XCONTEXTCONFIG,
1669 #endif
1670 };
1671 
1672 static u64 kvm_vz_get_one_regs_segments[] = {
1673 	KVM_REG_MIPS_CP0_SEGCTL0,
1674 	KVM_REG_MIPS_CP0_SEGCTL1,
1675 	KVM_REG_MIPS_CP0_SEGCTL2,
1676 };
1677 
1678 static u64 kvm_vz_get_one_regs_htw[] = {
1679 	KVM_REG_MIPS_CP0_PWBASE,
1680 	KVM_REG_MIPS_CP0_PWFIELD,
1681 	KVM_REG_MIPS_CP0_PWSIZE,
1682 	KVM_REG_MIPS_CP0_PWCTL,
1683 };
1684 
1685 static u64 kvm_vz_get_one_regs_kscratch[] = {
1686 	KVM_REG_MIPS_CP0_KSCRATCH1,
1687 	KVM_REG_MIPS_CP0_KSCRATCH2,
1688 	KVM_REG_MIPS_CP0_KSCRATCH3,
1689 	KVM_REG_MIPS_CP0_KSCRATCH4,
1690 	KVM_REG_MIPS_CP0_KSCRATCH5,
1691 	KVM_REG_MIPS_CP0_KSCRATCH6,
1692 };
1693 
1694 static unsigned long kvm_vz_num_regs(struct kvm_vcpu *vcpu)
1695 {
1696 	unsigned long ret;
1697 
1698 	ret = ARRAY_SIZE(kvm_vz_get_one_regs);
1699 	if (cpu_guest_has_userlocal)
1700 		++ret;
1701 	if (cpu_guest_has_badinstr)
1702 		++ret;
1703 	if (cpu_guest_has_badinstrp)
1704 		++ret;
1705 	if (cpu_guest_has_contextconfig)
1706 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1707 	if (cpu_guest_has_segments)
1708 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1709 	if (cpu_guest_has_htw)
1710 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1711 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar)
1712 		ret += 1 + ARRAY_SIZE(vcpu->arch.maar);
1713 	ret += __arch_hweight8(cpu_data[0].guest.kscratch_mask);
1714 
1715 	return ret;
1716 }
1717 
1718 static int kvm_vz_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
1719 {
1720 	u64 index;
1721 	unsigned int i;
1722 
1723 	if (copy_to_user(indices, kvm_vz_get_one_regs,
1724 			 sizeof(kvm_vz_get_one_regs)))
1725 		return -EFAULT;
1726 	indices += ARRAY_SIZE(kvm_vz_get_one_regs);
1727 
1728 	if (cpu_guest_has_userlocal) {
1729 		index = KVM_REG_MIPS_CP0_USERLOCAL;
1730 		if (copy_to_user(indices, &index, sizeof(index)))
1731 			return -EFAULT;
1732 		++indices;
1733 	}
1734 	if (cpu_guest_has_badinstr) {
1735 		index = KVM_REG_MIPS_CP0_BADINSTR;
1736 		if (copy_to_user(indices, &index, sizeof(index)))
1737 			return -EFAULT;
1738 		++indices;
1739 	}
1740 	if (cpu_guest_has_badinstrp) {
1741 		index = KVM_REG_MIPS_CP0_BADINSTRP;
1742 		if (copy_to_user(indices, &index, sizeof(index)))
1743 			return -EFAULT;
1744 		++indices;
1745 	}
1746 	if (cpu_guest_has_contextconfig) {
1747 		if (copy_to_user(indices, kvm_vz_get_one_regs_contextconfig,
1748 				 sizeof(kvm_vz_get_one_regs_contextconfig)))
1749 			return -EFAULT;
1750 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1751 	}
1752 	if (cpu_guest_has_segments) {
1753 		if (copy_to_user(indices, kvm_vz_get_one_regs_segments,
1754 				 sizeof(kvm_vz_get_one_regs_segments)))
1755 			return -EFAULT;
1756 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1757 	}
1758 	if (cpu_guest_has_htw) {
1759 		if (copy_to_user(indices, kvm_vz_get_one_regs_htw,
1760 				 sizeof(kvm_vz_get_one_regs_htw)))
1761 			return -EFAULT;
1762 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1763 	}
1764 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar) {
1765 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.maar); ++i) {
1766 			index = KVM_REG_MIPS_CP0_MAAR(i);
1767 			if (copy_to_user(indices, &index, sizeof(index)))
1768 				return -EFAULT;
1769 			++indices;
1770 		}
1771 
1772 		index = KVM_REG_MIPS_CP0_MAARI;
1773 		if (copy_to_user(indices, &index, sizeof(index)))
1774 			return -EFAULT;
1775 		++indices;
1776 	}
1777 	for (i = 0; i < 6; ++i) {
1778 		if (!cpu_guest_has_kscr(i + 2))
1779 			continue;
1780 
1781 		if (copy_to_user(indices, &kvm_vz_get_one_regs_kscratch[i],
1782 				 sizeof(kvm_vz_get_one_regs_kscratch[i])))
1783 			return -EFAULT;
1784 		++indices;
1785 	}
1786 
1787 	return 0;
1788 }
1789 
1790 static inline s64 entrylo_kvm_to_user(unsigned long v)
1791 {
1792 	s64 mask, ret = v;
1793 
1794 	if (BITS_PER_LONG == 32) {
1795 		/*
1796 		 * KVM API exposes 64-bit version of the register, so move the
1797 		 * RI/XI bits up into place.
1798 		 */
1799 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1800 		ret &= ~mask;
1801 		ret |= ((s64)v & mask) << 32;
1802 	}
1803 	return ret;
1804 }
1805 
1806 static inline unsigned long entrylo_user_to_kvm(s64 v)
1807 {
1808 	unsigned long mask, ret = v;
1809 
1810 	if (BITS_PER_LONG == 32) {
1811 		/*
1812 		 * KVM API exposes 64-bit versiono of the register, so move the
1813 		 * RI/XI bits down into place.
1814 		 */
1815 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1816 		ret &= ~mask;
1817 		ret |= (v >> 32) & mask;
1818 	}
1819 	return ret;
1820 }
1821 
1822 static int kvm_vz_get_one_reg(struct kvm_vcpu *vcpu,
1823 			      const struct kvm_one_reg *reg,
1824 			      s64 *v)
1825 {
1826 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1827 	unsigned int idx;
1828 
1829 	switch (reg->id) {
1830 	case KVM_REG_MIPS_CP0_INDEX:
1831 		*v = (long)read_gc0_index();
1832 		break;
1833 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1834 		*v = entrylo_kvm_to_user(read_gc0_entrylo0());
1835 		break;
1836 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1837 		*v = entrylo_kvm_to_user(read_gc0_entrylo1());
1838 		break;
1839 	case KVM_REG_MIPS_CP0_CONTEXT:
1840 		*v = (long)read_gc0_context();
1841 		break;
1842 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1843 		if (!cpu_guest_has_contextconfig)
1844 			return -EINVAL;
1845 		*v = read_gc0_contextconfig();
1846 		break;
1847 	case KVM_REG_MIPS_CP0_USERLOCAL:
1848 		if (!cpu_guest_has_userlocal)
1849 			return -EINVAL;
1850 		*v = read_gc0_userlocal();
1851 		break;
1852 #ifdef CONFIG_64BIT
1853 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1854 		if (!cpu_guest_has_contextconfig)
1855 			return -EINVAL;
1856 		*v = read_gc0_xcontextconfig();
1857 		break;
1858 #endif
1859 	case KVM_REG_MIPS_CP0_PAGEMASK:
1860 		*v = (long)read_gc0_pagemask();
1861 		break;
1862 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1863 		*v = (long)read_gc0_pagegrain();
1864 		break;
1865 	case KVM_REG_MIPS_CP0_SEGCTL0:
1866 		if (!cpu_guest_has_segments)
1867 			return -EINVAL;
1868 		*v = read_gc0_segctl0();
1869 		break;
1870 	case KVM_REG_MIPS_CP0_SEGCTL1:
1871 		if (!cpu_guest_has_segments)
1872 			return -EINVAL;
1873 		*v = read_gc0_segctl1();
1874 		break;
1875 	case KVM_REG_MIPS_CP0_SEGCTL2:
1876 		if (!cpu_guest_has_segments)
1877 			return -EINVAL;
1878 		*v = read_gc0_segctl2();
1879 		break;
1880 	case KVM_REG_MIPS_CP0_PWBASE:
1881 		if (!cpu_guest_has_htw)
1882 			return -EINVAL;
1883 		*v = read_gc0_pwbase();
1884 		break;
1885 	case KVM_REG_MIPS_CP0_PWFIELD:
1886 		if (!cpu_guest_has_htw)
1887 			return -EINVAL;
1888 		*v = read_gc0_pwfield();
1889 		break;
1890 	case KVM_REG_MIPS_CP0_PWSIZE:
1891 		if (!cpu_guest_has_htw)
1892 			return -EINVAL;
1893 		*v = read_gc0_pwsize();
1894 		break;
1895 	case KVM_REG_MIPS_CP0_WIRED:
1896 		*v = (long)read_gc0_wired();
1897 		break;
1898 	case KVM_REG_MIPS_CP0_PWCTL:
1899 		if (!cpu_guest_has_htw)
1900 			return -EINVAL;
1901 		*v = read_gc0_pwctl();
1902 		break;
1903 	case KVM_REG_MIPS_CP0_HWRENA:
1904 		*v = (long)read_gc0_hwrena();
1905 		break;
1906 	case KVM_REG_MIPS_CP0_BADVADDR:
1907 		*v = (long)read_gc0_badvaddr();
1908 		break;
1909 	case KVM_REG_MIPS_CP0_BADINSTR:
1910 		if (!cpu_guest_has_badinstr)
1911 			return -EINVAL;
1912 		*v = read_gc0_badinstr();
1913 		break;
1914 	case KVM_REG_MIPS_CP0_BADINSTRP:
1915 		if (!cpu_guest_has_badinstrp)
1916 			return -EINVAL;
1917 		*v = read_gc0_badinstrp();
1918 		break;
1919 	case KVM_REG_MIPS_CP0_COUNT:
1920 		*v = kvm_mips_read_count(vcpu);
1921 		break;
1922 	case KVM_REG_MIPS_CP0_ENTRYHI:
1923 		*v = (long)read_gc0_entryhi();
1924 		break;
1925 	case KVM_REG_MIPS_CP0_COMPARE:
1926 		*v = (long)read_gc0_compare();
1927 		break;
1928 	case KVM_REG_MIPS_CP0_STATUS:
1929 		*v = (long)read_gc0_status();
1930 		break;
1931 	case KVM_REG_MIPS_CP0_INTCTL:
1932 		*v = read_gc0_intctl();
1933 		break;
1934 	case KVM_REG_MIPS_CP0_CAUSE:
1935 		*v = (long)read_gc0_cause();
1936 		break;
1937 	case KVM_REG_MIPS_CP0_EPC:
1938 		*v = (long)read_gc0_epc();
1939 		break;
1940 	case KVM_REG_MIPS_CP0_PRID:
1941 		switch (boot_cpu_type()) {
1942 		case CPU_CAVIUM_OCTEON3:
1943 			/* Octeon III has a read-only guest.PRid */
1944 			*v = read_gc0_prid();
1945 			break;
1946 		default:
1947 			*v = (long)kvm_read_c0_guest_prid(cop0);
1948 			break;
1949 		};
1950 		break;
1951 	case KVM_REG_MIPS_CP0_EBASE:
1952 		*v = kvm_vz_read_gc0_ebase();
1953 		break;
1954 	case KVM_REG_MIPS_CP0_CONFIG:
1955 		*v = read_gc0_config();
1956 		break;
1957 	case KVM_REG_MIPS_CP0_CONFIG1:
1958 		if (!cpu_guest_has_conf1)
1959 			return -EINVAL;
1960 		*v = read_gc0_config1();
1961 		break;
1962 	case KVM_REG_MIPS_CP0_CONFIG2:
1963 		if (!cpu_guest_has_conf2)
1964 			return -EINVAL;
1965 		*v = read_gc0_config2();
1966 		break;
1967 	case KVM_REG_MIPS_CP0_CONFIG3:
1968 		if (!cpu_guest_has_conf3)
1969 			return -EINVAL;
1970 		*v = read_gc0_config3();
1971 		break;
1972 	case KVM_REG_MIPS_CP0_CONFIG4:
1973 		if (!cpu_guest_has_conf4)
1974 			return -EINVAL;
1975 		*v = read_gc0_config4();
1976 		break;
1977 	case KVM_REG_MIPS_CP0_CONFIG5:
1978 		if (!cpu_guest_has_conf5)
1979 			return -EINVAL;
1980 		*v = read_gc0_config5();
1981 		break;
1982 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
1983 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
1984 			return -EINVAL;
1985 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
1986 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
1987 			return -EINVAL;
1988 		*v = vcpu->arch.maar[idx];
1989 		break;
1990 	case KVM_REG_MIPS_CP0_MAARI:
1991 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
1992 			return -EINVAL;
1993 		*v = kvm_read_sw_gc0_maari(vcpu->arch.cop0);
1994 		break;
1995 #ifdef CONFIG_64BIT
1996 	case KVM_REG_MIPS_CP0_XCONTEXT:
1997 		*v = read_gc0_xcontext();
1998 		break;
1999 #endif
2000 	case KVM_REG_MIPS_CP0_ERROREPC:
2001 		*v = (long)read_gc0_errorepc();
2002 		break;
2003 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2004 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2005 		if (!cpu_guest_has_kscr(idx))
2006 			return -EINVAL;
2007 		switch (idx) {
2008 		case 2:
2009 			*v = (long)read_gc0_kscratch1();
2010 			break;
2011 		case 3:
2012 			*v = (long)read_gc0_kscratch2();
2013 			break;
2014 		case 4:
2015 			*v = (long)read_gc0_kscratch3();
2016 			break;
2017 		case 5:
2018 			*v = (long)read_gc0_kscratch4();
2019 			break;
2020 		case 6:
2021 			*v = (long)read_gc0_kscratch5();
2022 			break;
2023 		case 7:
2024 			*v = (long)read_gc0_kscratch6();
2025 			break;
2026 		}
2027 		break;
2028 	case KVM_REG_MIPS_COUNT_CTL:
2029 		*v = vcpu->arch.count_ctl;
2030 		break;
2031 	case KVM_REG_MIPS_COUNT_RESUME:
2032 		*v = ktime_to_ns(vcpu->arch.count_resume);
2033 		break;
2034 	case KVM_REG_MIPS_COUNT_HZ:
2035 		*v = vcpu->arch.count_hz;
2036 		break;
2037 	default:
2038 		return -EINVAL;
2039 	}
2040 	return 0;
2041 }
2042 
2043 static int kvm_vz_set_one_reg(struct kvm_vcpu *vcpu,
2044 			      const struct kvm_one_reg *reg,
2045 			      s64 v)
2046 {
2047 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2048 	unsigned int idx;
2049 	int ret = 0;
2050 	unsigned int cur, change;
2051 
2052 	switch (reg->id) {
2053 	case KVM_REG_MIPS_CP0_INDEX:
2054 		write_gc0_index(v);
2055 		break;
2056 	case KVM_REG_MIPS_CP0_ENTRYLO0:
2057 		write_gc0_entrylo0(entrylo_user_to_kvm(v));
2058 		break;
2059 	case KVM_REG_MIPS_CP0_ENTRYLO1:
2060 		write_gc0_entrylo1(entrylo_user_to_kvm(v));
2061 		break;
2062 	case KVM_REG_MIPS_CP0_CONTEXT:
2063 		write_gc0_context(v);
2064 		break;
2065 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
2066 		if (!cpu_guest_has_contextconfig)
2067 			return -EINVAL;
2068 		write_gc0_contextconfig(v);
2069 		break;
2070 	case KVM_REG_MIPS_CP0_USERLOCAL:
2071 		if (!cpu_guest_has_userlocal)
2072 			return -EINVAL;
2073 		write_gc0_userlocal(v);
2074 		break;
2075 #ifdef CONFIG_64BIT
2076 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
2077 		if (!cpu_guest_has_contextconfig)
2078 			return -EINVAL;
2079 		write_gc0_xcontextconfig(v);
2080 		break;
2081 #endif
2082 	case KVM_REG_MIPS_CP0_PAGEMASK:
2083 		write_gc0_pagemask(v);
2084 		break;
2085 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
2086 		write_gc0_pagegrain(v);
2087 		break;
2088 	case KVM_REG_MIPS_CP0_SEGCTL0:
2089 		if (!cpu_guest_has_segments)
2090 			return -EINVAL;
2091 		write_gc0_segctl0(v);
2092 		break;
2093 	case KVM_REG_MIPS_CP0_SEGCTL1:
2094 		if (!cpu_guest_has_segments)
2095 			return -EINVAL;
2096 		write_gc0_segctl1(v);
2097 		break;
2098 	case KVM_REG_MIPS_CP0_SEGCTL2:
2099 		if (!cpu_guest_has_segments)
2100 			return -EINVAL;
2101 		write_gc0_segctl2(v);
2102 		break;
2103 	case KVM_REG_MIPS_CP0_PWBASE:
2104 		if (!cpu_guest_has_htw)
2105 			return -EINVAL;
2106 		write_gc0_pwbase(v);
2107 		break;
2108 	case KVM_REG_MIPS_CP0_PWFIELD:
2109 		if (!cpu_guest_has_htw)
2110 			return -EINVAL;
2111 		write_gc0_pwfield(v);
2112 		break;
2113 	case KVM_REG_MIPS_CP0_PWSIZE:
2114 		if (!cpu_guest_has_htw)
2115 			return -EINVAL;
2116 		write_gc0_pwsize(v);
2117 		break;
2118 	case KVM_REG_MIPS_CP0_WIRED:
2119 		change_gc0_wired(MIPSR6_WIRED_WIRED, v);
2120 		break;
2121 	case KVM_REG_MIPS_CP0_PWCTL:
2122 		if (!cpu_guest_has_htw)
2123 			return -EINVAL;
2124 		write_gc0_pwctl(v);
2125 		break;
2126 	case KVM_REG_MIPS_CP0_HWRENA:
2127 		write_gc0_hwrena(v);
2128 		break;
2129 	case KVM_REG_MIPS_CP0_BADVADDR:
2130 		write_gc0_badvaddr(v);
2131 		break;
2132 	case KVM_REG_MIPS_CP0_BADINSTR:
2133 		if (!cpu_guest_has_badinstr)
2134 			return -EINVAL;
2135 		write_gc0_badinstr(v);
2136 		break;
2137 	case KVM_REG_MIPS_CP0_BADINSTRP:
2138 		if (!cpu_guest_has_badinstrp)
2139 			return -EINVAL;
2140 		write_gc0_badinstrp(v);
2141 		break;
2142 	case KVM_REG_MIPS_CP0_COUNT:
2143 		kvm_mips_write_count(vcpu, v);
2144 		break;
2145 	case KVM_REG_MIPS_CP0_ENTRYHI:
2146 		write_gc0_entryhi(v);
2147 		break;
2148 	case KVM_REG_MIPS_CP0_COMPARE:
2149 		kvm_mips_write_compare(vcpu, v, false);
2150 		break;
2151 	case KVM_REG_MIPS_CP0_STATUS:
2152 		write_gc0_status(v);
2153 		break;
2154 	case KVM_REG_MIPS_CP0_INTCTL:
2155 		write_gc0_intctl(v);
2156 		break;
2157 	case KVM_REG_MIPS_CP0_CAUSE:
2158 		/*
2159 		 * If the timer is stopped or started (DC bit) it must look
2160 		 * atomic with changes to the timer interrupt pending bit (TI).
2161 		 * A timer interrupt should not happen in between.
2162 		 */
2163 		if ((read_gc0_cause() ^ v) & CAUSEF_DC) {
2164 			if (v & CAUSEF_DC) {
2165 				/* disable timer first */
2166 				kvm_mips_count_disable_cause(vcpu);
2167 				change_gc0_cause((u32)~CAUSEF_DC, v);
2168 			} else {
2169 				/* enable timer last */
2170 				change_gc0_cause((u32)~CAUSEF_DC, v);
2171 				kvm_mips_count_enable_cause(vcpu);
2172 			}
2173 		} else {
2174 			write_gc0_cause(v);
2175 		}
2176 		break;
2177 	case KVM_REG_MIPS_CP0_EPC:
2178 		write_gc0_epc(v);
2179 		break;
2180 	case KVM_REG_MIPS_CP0_PRID:
2181 		switch (boot_cpu_type()) {
2182 		case CPU_CAVIUM_OCTEON3:
2183 			/* Octeon III has a guest.PRid, but its read-only */
2184 			break;
2185 		default:
2186 			kvm_write_c0_guest_prid(cop0, v);
2187 			break;
2188 		};
2189 		break;
2190 	case KVM_REG_MIPS_CP0_EBASE:
2191 		kvm_vz_write_gc0_ebase(v);
2192 		break;
2193 	case KVM_REG_MIPS_CP0_CONFIG:
2194 		cur = read_gc0_config();
2195 		change = (cur ^ v) & kvm_vz_config_user_wrmask(vcpu);
2196 		if (change) {
2197 			v = cur ^ change;
2198 			write_gc0_config(v);
2199 		}
2200 		break;
2201 	case KVM_REG_MIPS_CP0_CONFIG1:
2202 		if (!cpu_guest_has_conf1)
2203 			break;
2204 		cur = read_gc0_config1();
2205 		change = (cur ^ v) & kvm_vz_config1_user_wrmask(vcpu);
2206 		if (change) {
2207 			v = cur ^ change;
2208 			write_gc0_config1(v);
2209 		}
2210 		break;
2211 	case KVM_REG_MIPS_CP0_CONFIG2:
2212 		if (!cpu_guest_has_conf2)
2213 			break;
2214 		cur = read_gc0_config2();
2215 		change = (cur ^ v) & kvm_vz_config2_user_wrmask(vcpu);
2216 		if (change) {
2217 			v = cur ^ change;
2218 			write_gc0_config2(v);
2219 		}
2220 		break;
2221 	case KVM_REG_MIPS_CP0_CONFIG3:
2222 		if (!cpu_guest_has_conf3)
2223 			break;
2224 		cur = read_gc0_config3();
2225 		change = (cur ^ v) & kvm_vz_config3_user_wrmask(vcpu);
2226 		if (change) {
2227 			v = cur ^ change;
2228 			write_gc0_config3(v);
2229 		}
2230 		break;
2231 	case KVM_REG_MIPS_CP0_CONFIG4:
2232 		if (!cpu_guest_has_conf4)
2233 			break;
2234 		cur = read_gc0_config4();
2235 		change = (cur ^ v) & kvm_vz_config4_user_wrmask(vcpu);
2236 		if (change) {
2237 			v = cur ^ change;
2238 			write_gc0_config4(v);
2239 		}
2240 		break;
2241 	case KVM_REG_MIPS_CP0_CONFIG5:
2242 		if (!cpu_guest_has_conf5)
2243 			break;
2244 		cur = read_gc0_config5();
2245 		change = (cur ^ v) & kvm_vz_config5_user_wrmask(vcpu);
2246 		if (change) {
2247 			v = cur ^ change;
2248 			write_gc0_config5(v);
2249 		}
2250 		break;
2251 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2252 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2253 			return -EINVAL;
2254 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2255 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2256 			return -EINVAL;
2257 		vcpu->arch.maar[idx] = mips_process_maar(dmtc_op, v);
2258 		break;
2259 	case KVM_REG_MIPS_CP0_MAARI:
2260 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2261 			return -EINVAL;
2262 		kvm_write_maari(vcpu, v);
2263 		break;
2264 #ifdef CONFIG_64BIT
2265 	case KVM_REG_MIPS_CP0_XCONTEXT:
2266 		write_gc0_xcontext(v);
2267 		break;
2268 #endif
2269 	case KVM_REG_MIPS_CP0_ERROREPC:
2270 		write_gc0_errorepc(v);
2271 		break;
2272 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2273 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2274 		if (!cpu_guest_has_kscr(idx))
2275 			return -EINVAL;
2276 		switch (idx) {
2277 		case 2:
2278 			write_gc0_kscratch1(v);
2279 			break;
2280 		case 3:
2281 			write_gc0_kscratch2(v);
2282 			break;
2283 		case 4:
2284 			write_gc0_kscratch3(v);
2285 			break;
2286 		case 5:
2287 			write_gc0_kscratch4(v);
2288 			break;
2289 		case 6:
2290 			write_gc0_kscratch5(v);
2291 			break;
2292 		case 7:
2293 			write_gc0_kscratch6(v);
2294 			break;
2295 		}
2296 		break;
2297 	case KVM_REG_MIPS_COUNT_CTL:
2298 		ret = kvm_mips_set_count_ctl(vcpu, v);
2299 		break;
2300 	case KVM_REG_MIPS_COUNT_RESUME:
2301 		ret = kvm_mips_set_count_resume(vcpu, v);
2302 		break;
2303 	case KVM_REG_MIPS_COUNT_HZ:
2304 		ret = kvm_mips_set_count_hz(vcpu, v);
2305 		break;
2306 	default:
2307 		return -EINVAL;
2308 	}
2309 	return ret;
2310 }
2311 
2312 #define guestid_cache(cpu)	(cpu_data[cpu].guestid_cache)
2313 static void kvm_vz_get_new_guestid(unsigned long cpu, struct kvm_vcpu *vcpu)
2314 {
2315 	unsigned long guestid = guestid_cache(cpu);
2316 
2317 	if (!(++guestid & GUESTID_MASK)) {
2318 		if (cpu_has_vtag_icache)
2319 			flush_icache_all();
2320 
2321 		if (!guestid)		/* fix version if needed */
2322 			guestid = GUESTID_FIRST_VERSION;
2323 
2324 		++guestid;		/* guestid 0 reserved for root */
2325 
2326 		/* start new guestid cycle */
2327 		kvm_vz_local_flush_roottlb_all_guests();
2328 		kvm_vz_local_flush_guesttlb_all();
2329 	}
2330 
2331 	guestid_cache(cpu) = guestid;
2332 }
2333 
2334 /* Returns 1 if the guest TLB may be clobbered */
2335 static int kvm_vz_check_requests(struct kvm_vcpu *vcpu, int cpu)
2336 {
2337 	int ret = 0;
2338 	int i;
2339 
2340 	if (!kvm_request_pending(vcpu))
2341 		return 0;
2342 
2343 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2344 		if (cpu_has_guestid) {
2345 			/* Drop all GuestIDs for this VCPU */
2346 			for_each_possible_cpu(i)
2347 				vcpu->arch.vzguestid[i] = 0;
2348 			/* This will clobber guest TLB contents too */
2349 			ret = 1;
2350 		}
2351 		/*
2352 		 * For Root ASID Dealias (RAD) we don't do anything here, but we
2353 		 * still need the request to ensure we recheck asid_flush_mask.
2354 		 * We can still return 0 as only the root TLB will be affected
2355 		 * by a root ASID flush.
2356 		 */
2357 	}
2358 
2359 	return ret;
2360 }
2361 
2362 static void kvm_vz_vcpu_save_wired(struct kvm_vcpu *vcpu)
2363 {
2364 	unsigned int wired = read_gc0_wired();
2365 	struct kvm_mips_tlb *tlbs;
2366 	int i;
2367 
2368 	/* Expand the wired TLB array if necessary */
2369 	wired &= MIPSR6_WIRED_WIRED;
2370 	if (wired > vcpu->arch.wired_tlb_limit) {
2371 		tlbs = krealloc(vcpu->arch.wired_tlb, wired *
2372 				sizeof(*vcpu->arch.wired_tlb), GFP_ATOMIC);
2373 		if (WARN_ON(!tlbs)) {
2374 			/* Save whatever we can */
2375 			wired = vcpu->arch.wired_tlb_limit;
2376 		} else {
2377 			vcpu->arch.wired_tlb = tlbs;
2378 			vcpu->arch.wired_tlb_limit = wired;
2379 		}
2380 	}
2381 
2382 	if (wired)
2383 		/* Save wired entries from the guest TLB */
2384 		kvm_vz_save_guesttlb(vcpu->arch.wired_tlb, 0, wired);
2385 	/* Invalidate any dropped entries since last time */
2386 	for (i = wired; i < vcpu->arch.wired_tlb_used; ++i) {
2387 		vcpu->arch.wired_tlb[i].tlb_hi = UNIQUE_GUEST_ENTRYHI(i);
2388 		vcpu->arch.wired_tlb[i].tlb_lo[0] = 0;
2389 		vcpu->arch.wired_tlb[i].tlb_lo[1] = 0;
2390 		vcpu->arch.wired_tlb[i].tlb_mask = 0;
2391 	}
2392 	vcpu->arch.wired_tlb_used = wired;
2393 }
2394 
2395 static void kvm_vz_vcpu_load_wired(struct kvm_vcpu *vcpu)
2396 {
2397 	/* Load wired entries into the guest TLB */
2398 	if (vcpu->arch.wired_tlb)
2399 		kvm_vz_load_guesttlb(vcpu->arch.wired_tlb, 0,
2400 				     vcpu->arch.wired_tlb_used);
2401 }
2402 
2403 static void kvm_vz_vcpu_load_tlb(struct kvm_vcpu *vcpu, int cpu)
2404 {
2405 	struct kvm *kvm = vcpu->kvm;
2406 	struct mm_struct *gpa_mm = &kvm->arch.gpa_mm;
2407 	bool migrated;
2408 
2409 	/*
2410 	 * Are we entering guest context on a different CPU to last time?
2411 	 * If so, the VCPU's guest TLB state on this CPU may be stale.
2412 	 */
2413 	migrated = (vcpu->arch.last_exec_cpu != cpu);
2414 	vcpu->arch.last_exec_cpu = cpu;
2415 
2416 	/*
2417 	 * A vcpu's GuestID is set in GuestCtl1.ID when the vcpu is loaded and
2418 	 * remains set until another vcpu is loaded in.  As a rule GuestRID
2419 	 * remains zeroed when in root context unless the kernel is busy
2420 	 * manipulating guest tlb entries.
2421 	 */
2422 	if (cpu_has_guestid) {
2423 		/*
2424 		 * Check if our GuestID is of an older version and thus invalid.
2425 		 *
2426 		 * We also discard the stored GuestID if we've executed on
2427 		 * another CPU, as the guest mappings may have changed without
2428 		 * hypervisor knowledge.
2429 		 */
2430 		if (migrated ||
2431 		    (vcpu->arch.vzguestid[cpu] ^ guestid_cache(cpu)) &
2432 					GUESTID_VERSION_MASK) {
2433 			kvm_vz_get_new_guestid(cpu, vcpu);
2434 			vcpu->arch.vzguestid[cpu] = guestid_cache(cpu);
2435 			trace_kvm_guestid_change(vcpu,
2436 						 vcpu->arch.vzguestid[cpu]);
2437 		}
2438 
2439 		/* Restore GuestID */
2440 		change_c0_guestctl1(GUESTID_MASK, vcpu->arch.vzguestid[cpu]);
2441 	} else {
2442 		/*
2443 		 * The Guest TLB only stores a single guest's TLB state, so
2444 		 * flush it if another VCPU has executed on this CPU.
2445 		 *
2446 		 * We also flush if we've executed on another CPU, as the guest
2447 		 * mappings may have changed without hypervisor knowledge.
2448 		 */
2449 		if (migrated || last_exec_vcpu[cpu] != vcpu)
2450 			kvm_vz_local_flush_guesttlb_all();
2451 		last_exec_vcpu[cpu] = vcpu;
2452 
2453 		/*
2454 		 * Root ASID dealiases guest GPA mappings in the root TLB.
2455 		 * Allocate new root ASID if needed.
2456 		 */
2457 		if (cpumask_test_and_clear_cpu(cpu, &kvm->arch.asid_flush_mask))
2458 			get_new_mmu_context(gpa_mm);
2459 		else
2460 			check_mmu_context(gpa_mm);
2461 	}
2462 }
2463 
2464 static int kvm_vz_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2465 {
2466 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2467 	bool migrated, all;
2468 
2469 	/*
2470 	 * Have we migrated to a different CPU?
2471 	 * If so, any old guest TLB state may be stale.
2472 	 */
2473 	migrated = (vcpu->arch.last_sched_cpu != cpu);
2474 
2475 	/*
2476 	 * Was this the last VCPU to run on this CPU?
2477 	 * If not, any old guest state from this VCPU will have been clobbered.
2478 	 */
2479 	all = migrated || (last_vcpu[cpu] != vcpu);
2480 	last_vcpu[cpu] = vcpu;
2481 
2482 	/*
2483 	 * Restore CP0_Wired unconditionally as we clear it after use, and
2484 	 * restore wired guest TLB entries (while in guest context).
2485 	 */
2486 	kvm_restore_gc0_wired(cop0);
2487 	if (current->flags & PF_VCPU) {
2488 		tlbw_use_hazard();
2489 		kvm_vz_vcpu_load_tlb(vcpu, cpu);
2490 		kvm_vz_vcpu_load_wired(vcpu);
2491 	}
2492 
2493 	/*
2494 	 * Restore timer state regardless, as e.g. Cause.TI can change over time
2495 	 * if left unmaintained.
2496 	 */
2497 	kvm_vz_restore_timer(vcpu);
2498 
2499 	/* Set MC bit if we want to trace guest mode changes */
2500 	if (kvm_trace_guest_mode_change)
2501 		set_c0_guestctl0(MIPS_GCTL0_MC);
2502 	else
2503 		clear_c0_guestctl0(MIPS_GCTL0_MC);
2504 
2505 	/* Don't bother restoring registers multiple times unless necessary */
2506 	if (!all)
2507 		return 0;
2508 
2509 	/*
2510 	 * Restore config registers first, as some implementations restrict
2511 	 * writes to other registers when the corresponding feature bits aren't
2512 	 * set. For example Status.CU1 cannot be set unless Config1.FP is set.
2513 	 */
2514 	kvm_restore_gc0_config(cop0);
2515 	if (cpu_guest_has_conf1)
2516 		kvm_restore_gc0_config1(cop0);
2517 	if (cpu_guest_has_conf2)
2518 		kvm_restore_gc0_config2(cop0);
2519 	if (cpu_guest_has_conf3)
2520 		kvm_restore_gc0_config3(cop0);
2521 	if (cpu_guest_has_conf4)
2522 		kvm_restore_gc0_config4(cop0);
2523 	if (cpu_guest_has_conf5)
2524 		kvm_restore_gc0_config5(cop0);
2525 	if (cpu_guest_has_conf6)
2526 		kvm_restore_gc0_config6(cop0);
2527 	if (cpu_guest_has_conf7)
2528 		kvm_restore_gc0_config7(cop0);
2529 
2530 	kvm_restore_gc0_index(cop0);
2531 	kvm_restore_gc0_entrylo0(cop0);
2532 	kvm_restore_gc0_entrylo1(cop0);
2533 	kvm_restore_gc0_context(cop0);
2534 	if (cpu_guest_has_contextconfig)
2535 		kvm_restore_gc0_contextconfig(cop0);
2536 #ifdef CONFIG_64BIT
2537 	kvm_restore_gc0_xcontext(cop0);
2538 	if (cpu_guest_has_contextconfig)
2539 		kvm_restore_gc0_xcontextconfig(cop0);
2540 #endif
2541 	kvm_restore_gc0_pagemask(cop0);
2542 	kvm_restore_gc0_pagegrain(cop0);
2543 	kvm_restore_gc0_hwrena(cop0);
2544 	kvm_restore_gc0_badvaddr(cop0);
2545 	kvm_restore_gc0_entryhi(cop0);
2546 	kvm_restore_gc0_status(cop0);
2547 	kvm_restore_gc0_intctl(cop0);
2548 	kvm_restore_gc0_epc(cop0);
2549 	kvm_vz_write_gc0_ebase(kvm_read_sw_gc0_ebase(cop0));
2550 	if (cpu_guest_has_userlocal)
2551 		kvm_restore_gc0_userlocal(cop0);
2552 
2553 	kvm_restore_gc0_errorepc(cop0);
2554 
2555 	/* restore KScratch registers if enabled in guest */
2556 	if (cpu_guest_has_conf4) {
2557 		if (cpu_guest_has_kscr(2))
2558 			kvm_restore_gc0_kscratch1(cop0);
2559 		if (cpu_guest_has_kscr(3))
2560 			kvm_restore_gc0_kscratch2(cop0);
2561 		if (cpu_guest_has_kscr(4))
2562 			kvm_restore_gc0_kscratch3(cop0);
2563 		if (cpu_guest_has_kscr(5))
2564 			kvm_restore_gc0_kscratch4(cop0);
2565 		if (cpu_guest_has_kscr(6))
2566 			kvm_restore_gc0_kscratch5(cop0);
2567 		if (cpu_guest_has_kscr(7))
2568 			kvm_restore_gc0_kscratch6(cop0);
2569 	}
2570 
2571 	if (cpu_guest_has_badinstr)
2572 		kvm_restore_gc0_badinstr(cop0);
2573 	if (cpu_guest_has_badinstrp)
2574 		kvm_restore_gc0_badinstrp(cop0);
2575 
2576 	if (cpu_guest_has_segments) {
2577 		kvm_restore_gc0_segctl0(cop0);
2578 		kvm_restore_gc0_segctl1(cop0);
2579 		kvm_restore_gc0_segctl2(cop0);
2580 	}
2581 
2582 	/* restore HTW registers */
2583 	if (cpu_guest_has_htw) {
2584 		kvm_restore_gc0_pwbase(cop0);
2585 		kvm_restore_gc0_pwfield(cop0);
2586 		kvm_restore_gc0_pwsize(cop0);
2587 		kvm_restore_gc0_pwctl(cop0);
2588 	}
2589 
2590 	/* restore Root.GuestCtl2 from unused Guest guestctl2 register */
2591 	if (cpu_has_guestctl2)
2592 		write_c0_guestctl2(
2593 			cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL]);
2594 
2595 	/*
2596 	 * We should clear linked load bit to break interrupted atomics. This
2597 	 * prevents a SC on the next VCPU from succeeding by matching a LL on
2598 	 * the previous VCPU.
2599 	 */
2600 	if (cpu_guest_has_rw_llb)
2601 		write_gc0_lladdr(0);
2602 
2603 	return 0;
2604 }
2605 
2606 static int kvm_vz_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
2607 {
2608 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2609 
2610 	if (current->flags & PF_VCPU)
2611 		kvm_vz_vcpu_save_wired(vcpu);
2612 
2613 	kvm_lose_fpu(vcpu);
2614 
2615 	kvm_save_gc0_index(cop0);
2616 	kvm_save_gc0_entrylo0(cop0);
2617 	kvm_save_gc0_entrylo1(cop0);
2618 	kvm_save_gc0_context(cop0);
2619 	if (cpu_guest_has_contextconfig)
2620 		kvm_save_gc0_contextconfig(cop0);
2621 #ifdef CONFIG_64BIT
2622 	kvm_save_gc0_xcontext(cop0);
2623 	if (cpu_guest_has_contextconfig)
2624 		kvm_save_gc0_xcontextconfig(cop0);
2625 #endif
2626 	kvm_save_gc0_pagemask(cop0);
2627 	kvm_save_gc0_pagegrain(cop0);
2628 	kvm_save_gc0_wired(cop0);
2629 	/* allow wired TLB entries to be overwritten */
2630 	clear_gc0_wired(MIPSR6_WIRED_WIRED);
2631 	kvm_save_gc0_hwrena(cop0);
2632 	kvm_save_gc0_badvaddr(cop0);
2633 	kvm_save_gc0_entryhi(cop0);
2634 	kvm_save_gc0_status(cop0);
2635 	kvm_save_gc0_intctl(cop0);
2636 	kvm_save_gc0_epc(cop0);
2637 	kvm_write_sw_gc0_ebase(cop0, kvm_vz_read_gc0_ebase());
2638 	if (cpu_guest_has_userlocal)
2639 		kvm_save_gc0_userlocal(cop0);
2640 
2641 	/* only save implemented config registers */
2642 	kvm_save_gc0_config(cop0);
2643 	if (cpu_guest_has_conf1)
2644 		kvm_save_gc0_config1(cop0);
2645 	if (cpu_guest_has_conf2)
2646 		kvm_save_gc0_config2(cop0);
2647 	if (cpu_guest_has_conf3)
2648 		kvm_save_gc0_config3(cop0);
2649 	if (cpu_guest_has_conf4)
2650 		kvm_save_gc0_config4(cop0);
2651 	if (cpu_guest_has_conf5)
2652 		kvm_save_gc0_config5(cop0);
2653 	if (cpu_guest_has_conf6)
2654 		kvm_save_gc0_config6(cop0);
2655 	if (cpu_guest_has_conf7)
2656 		kvm_save_gc0_config7(cop0);
2657 
2658 	kvm_save_gc0_errorepc(cop0);
2659 
2660 	/* save KScratch registers if enabled in guest */
2661 	if (cpu_guest_has_conf4) {
2662 		if (cpu_guest_has_kscr(2))
2663 			kvm_save_gc0_kscratch1(cop0);
2664 		if (cpu_guest_has_kscr(3))
2665 			kvm_save_gc0_kscratch2(cop0);
2666 		if (cpu_guest_has_kscr(4))
2667 			kvm_save_gc0_kscratch3(cop0);
2668 		if (cpu_guest_has_kscr(5))
2669 			kvm_save_gc0_kscratch4(cop0);
2670 		if (cpu_guest_has_kscr(6))
2671 			kvm_save_gc0_kscratch5(cop0);
2672 		if (cpu_guest_has_kscr(7))
2673 			kvm_save_gc0_kscratch6(cop0);
2674 	}
2675 
2676 	if (cpu_guest_has_badinstr)
2677 		kvm_save_gc0_badinstr(cop0);
2678 	if (cpu_guest_has_badinstrp)
2679 		kvm_save_gc0_badinstrp(cop0);
2680 
2681 	if (cpu_guest_has_segments) {
2682 		kvm_save_gc0_segctl0(cop0);
2683 		kvm_save_gc0_segctl1(cop0);
2684 		kvm_save_gc0_segctl2(cop0);
2685 	}
2686 
2687 	/* save HTW registers if enabled in guest */
2688 	if (cpu_guest_has_htw &&
2689 	    kvm_read_sw_gc0_config3(cop0) & MIPS_CONF3_PW) {
2690 		kvm_save_gc0_pwbase(cop0);
2691 		kvm_save_gc0_pwfield(cop0);
2692 		kvm_save_gc0_pwsize(cop0);
2693 		kvm_save_gc0_pwctl(cop0);
2694 	}
2695 
2696 	kvm_vz_save_timer(vcpu);
2697 
2698 	/* save Root.GuestCtl2 in unused Guest guestctl2 register */
2699 	if (cpu_has_guestctl2)
2700 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] =
2701 			read_c0_guestctl2();
2702 
2703 	return 0;
2704 }
2705 
2706 /**
2707  * kvm_vz_resize_guest_vtlb() - Attempt to resize guest VTLB.
2708  * @size:	Number of guest VTLB entries (0 < @size <= root VTLB entries).
2709  *
2710  * Attempt to resize the guest VTLB by writing guest Config registers. This is
2711  * necessary for cores with a shared root/guest TLB to avoid overlap with wired
2712  * entries in the root VTLB.
2713  *
2714  * Returns:	The resulting guest VTLB size.
2715  */
2716 static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
2717 {
2718 	unsigned int config4 = 0, ret = 0, limit;
2719 
2720 	/* Write MMUSize - 1 into guest Config registers */
2721 	if (cpu_guest_has_conf1)
2722 		change_gc0_config1(MIPS_CONF1_TLBS,
2723 				   (size - 1) << MIPS_CONF1_TLBS_SHIFT);
2724 	if (cpu_guest_has_conf4) {
2725 		config4 = read_gc0_config4();
2726 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2727 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT) {
2728 			config4 &= ~MIPS_CONF4_VTLBSIZEEXT;
2729 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2730 				MIPS_CONF4_VTLBSIZEEXT_SHIFT;
2731 		} else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2732 			   MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT) {
2733 			config4 &= ~MIPS_CONF4_MMUSIZEEXT;
2734 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2735 				MIPS_CONF4_MMUSIZEEXT_SHIFT;
2736 		}
2737 		write_gc0_config4(config4);
2738 	}
2739 
2740 	/*
2741 	 * Set Guest.Wired.Limit = 0 (no limit up to Guest.MMUSize-1), unless it
2742 	 * would exceed Root.Wired.Limit (clearing Guest.Wired.Wired so write
2743 	 * not dropped)
2744 	 */
2745 	if (cpu_has_mips_r6) {
2746 		limit = (read_c0_wired() & MIPSR6_WIRED_LIMIT) >>
2747 						MIPSR6_WIRED_LIMIT_SHIFT;
2748 		if (size - 1 <= limit)
2749 			limit = 0;
2750 		write_gc0_wired(limit << MIPSR6_WIRED_LIMIT_SHIFT);
2751 	}
2752 
2753 	/* Read back MMUSize - 1 */
2754 	back_to_back_c0_hazard();
2755 	if (cpu_guest_has_conf1)
2756 		ret = (read_gc0_config1() & MIPS_CONF1_TLBS) >>
2757 						MIPS_CONF1_TLBS_SHIFT;
2758 	if (config4) {
2759 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2760 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT)
2761 			ret |= ((config4 & MIPS_CONF4_VTLBSIZEEXT) >>
2762 				MIPS_CONF4_VTLBSIZEEXT_SHIFT) <<
2763 				MIPS_CONF1_TLBS_SIZE;
2764 		else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2765 			 MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT)
2766 			ret |= ((config4 & MIPS_CONF4_MMUSIZEEXT) >>
2767 				MIPS_CONF4_MMUSIZEEXT_SHIFT) <<
2768 				MIPS_CONF1_TLBS_SIZE;
2769 	}
2770 	return ret + 1;
2771 }
2772 
2773 static int kvm_vz_hardware_enable(void)
2774 {
2775 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
2776 	u64 guest_cvmctl, cvmvmconfig;
2777 
2778 	switch (current_cpu_type()) {
2779 	case CPU_CAVIUM_OCTEON3:
2780 		/* Set up guest timer/perfcount IRQ lines */
2781 		guest_cvmctl = read_gc0_cvmctl();
2782 		guest_cvmctl &= ~CVMCTL_IPTI;
2783 		guest_cvmctl |= 7ull << CVMCTL_IPTI_SHIFT;
2784 		guest_cvmctl &= ~CVMCTL_IPPCI;
2785 		guest_cvmctl |= 6ull << CVMCTL_IPPCI_SHIFT;
2786 		write_gc0_cvmctl(guest_cvmctl);
2787 
2788 		cvmvmconfig = read_c0_cvmvmconfig();
2789 		/* No I/O hole translation. */
2790 		cvmvmconfig |= CVMVMCONF_DGHT;
2791 		/* Halve the root MMU size */
2792 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2793 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
2794 		guest_mmu_size = mmu_size / 2;
2795 		mmu_size -= guest_mmu_size;
2796 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
2797 		cvmvmconfig |= mmu_size - 1;
2798 		write_c0_cvmvmconfig(cvmvmconfig);
2799 
2800 		/* Update our records */
2801 		current_cpu_data.tlbsize = mmu_size;
2802 		current_cpu_data.tlbsizevtlb = mmu_size;
2803 		current_cpu_data.guest.tlbsize = guest_mmu_size;
2804 
2805 		/* Flush moved entries in new (guest) context */
2806 		kvm_vz_local_flush_guesttlb_all();
2807 		break;
2808 	default:
2809 		/*
2810 		 * ImgTec cores tend to use a shared root/guest TLB. To avoid
2811 		 * overlap of root wired and guest entries, the guest TLB may
2812 		 * need resizing.
2813 		 */
2814 		mmu_size = current_cpu_data.tlbsizevtlb;
2815 		ftlb_size = current_cpu_data.tlbsize - mmu_size;
2816 
2817 		/* Try switching to maximum guest VTLB size for flush */
2818 		guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
2819 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2820 		kvm_vz_local_flush_guesttlb_all();
2821 
2822 		/*
2823 		 * Reduce to make space for root wired entries and at least 2
2824 		 * root non-wired entries. This does assume that long-term wired
2825 		 * entries won't be added later.
2826 		 */
2827 		guest_mmu_size = mmu_size - num_wired_entries() - 2;
2828 		guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
2829 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2830 
2831 		/*
2832 		 * Write the VTLB size, but if another CPU has already written,
2833 		 * check it matches or we won't provide a consistent view to the
2834 		 * guest. If this ever happens it suggests an asymmetric number
2835 		 * of wired entries.
2836 		 */
2837 		if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
2838 		    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
2839 			 "Available guest VTLB size mismatch"))
2840 			return -EINVAL;
2841 		break;
2842 	}
2843 
2844 	/*
2845 	 * Enable virtualization features granting guest direct control of
2846 	 * certain features:
2847 	 * CP0=1:	Guest coprocessor 0 context.
2848 	 * AT=Guest:	Guest MMU.
2849 	 * CG=1:	Hit (virtual address) CACHE operations (optional).
2850 	 * CF=1:	Guest Config registers.
2851 	 * CGI=1:	Indexed flush CACHE operations (optional).
2852 	 */
2853 	write_c0_guestctl0(MIPS_GCTL0_CP0 |
2854 			   (MIPS_GCTL0_AT_GUEST << MIPS_GCTL0_AT_SHIFT) |
2855 			   MIPS_GCTL0_CG | MIPS_GCTL0_CF);
2856 	if (cpu_has_guestctl0ext)
2857 		set_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2858 
2859 	if (cpu_has_guestid) {
2860 		write_c0_guestctl1(0);
2861 		kvm_vz_local_flush_roottlb_all_guests();
2862 
2863 		GUESTID_MASK = current_cpu_data.guestid_mask;
2864 		GUESTID_FIRST_VERSION = GUESTID_MASK + 1;
2865 		GUESTID_VERSION_MASK = ~GUESTID_MASK;
2866 
2867 		current_cpu_data.guestid_cache = GUESTID_FIRST_VERSION;
2868 	}
2869 
2870 	/* clear any pending injected virtual guest interrupts */
2871 	if (cpu_has_guestctl2)
2872 		clear_c0_guestctl2(0x3f << 10);
2873 
2874 	return 0;
2875 }
2876 
2877 static void kvm_vz_hardware_disable(void)
2878 {
2879 	u64 cvmvmconfig;
2880 	unsigned int mmu_size;
2881 
2882 	/* Flush any remaining guest TLB entries */
2883 	kvm_vz_local_flush_guesttlb_all();
2884 
2885 	switch (current_cpu_type()) {
2886 	case CPU_CAVIUM_OCTEON3:
2887 		/*
2888 		 * Allocate whole TLB for root. Existing guest TLB entries will
2889 		 * change ownership to the root TLB. We should be safe though as
2890 		 * they've already been flushed above while in guest TLB.
2891 		 */
2892 		cvmvmconfig = read_c0_cvmvmconfig();
2893 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2894 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
2895 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
2896 		cvmvmconfig |= mmu_size - 1;
2897 		write_c0_cvmvmconfig(cvmvmconfig);
2898 
2899 		/* Update our records */
2900 		current_cpu_data.tlbsize = mmu_size;
2901 		current_cpu_data.tlbsizevtlb = mmu_size;
2902 		current_cpu_data.guest.tlbsize = 0;
2903 
2904 		/* Flush moved entries in new (root) context */
2905 		local_flush_tlb_all();
2906 		break;
2907 	}
2908 
2909 	if (cpu_has_guestid) {
2910 		write_c0_guestctl1(0);
2911 		kvm_vz_local_flush_roottlb_all_guests();
2912 	}
2913 }
2914 
2915 static int kvm_vz_check_extension(struct kvm *kvm, long ext)
2916 {
2917 	int r;
2918 
2919 	switch (ext) {
2920 	case KVM_CAP_MIPS_VZ:
2921 		/* we wouldn't be here unless cpu_has_vz */
2922 		r = 1;
2923 		break;
2924 #ifdef CONFIG_64BIT
2925 	case KVM_CAP_MIPS_64BIT:
2926 		/* We support 64-bit registers/operations and addresses */
2927 		r = 2;
2928 		break;
2929 #endif
2930 	default:
2931 		r = 0;
2932 		break;
2933 	}
2934 
2935 	return r;
2936 }
2937 
2938 static int kvm_vz_vcpu_init(struct kvm_vcpu *vcpu)
2939 {
2940 	int i;
2941 
2942 	for_each_possible_cpu(i)
2943 		vcpu->arch.vzguestid[i] = 0;
2944 
2945 	return 0;
2946 }
2947 
2948 static void kvm_vz_vcpu_uninit(struct kvm_vcpu *vcpu)
2949 {
2950 	int cpu;
2951 
2952 	/*
2953 	 * If the VCPU is freed and reused as another VCPU, we don't want the
2954 	 * matching pointer wrongly hanging around in last_vcpu[] or
2955 	 * last_exec_vcpu[].
2956 	 */
2957 	for_each_possible_cpu(cpu) {
2958 		if (last_vcpu[cpu] == vcpu)
2959 			last_vcpu[cpu] = NULL;
2960 		if (last_exec_vcpu[cpu] == vcpu)
2961 			last_exec_vcpu[cpu] = NULL;
2962 	}
2963 }
2964 
2965 static int kvm_vz_vcpu_setup(struct kvm_vcpu *vcpu)
2966 {
2967 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2968 	unsigned long count_hz = 100*1000*1000; /* default to 100 MHz */
2969 
2970 	/*
2971 	 * Start off the timer at the same frequency as the host timer, but the
2972 	 * soft timer doesn't handle frequencies greater than 1GHz yet.
2973 	 */
2974 	if (mips_hpt_frequency && mips_hpt_frequency <= NSEC_PER_SEC)
2975 		count_hz = mips_hpt_frequency;
2976 	kvm_mips_init_count(vcpu, count_hz);
2977 
2978 	/*
2979 	 * Initialize guest register state to valid architectural reset state.
2980 	 */
2981 
2982 	/* PageGrain */
2983 	if (cpu_has_mips_r6)
2984 		kvm_write_sw_gc0_pagegrain(cop0, PG_RIE | PG_XIE | PG_IEC);
2985 	/* Wired */
2986 	if (cpu_has_mips_r6)
2987 		kvm_write_sw_gc0_wired(cop0,
2988 				       read_gc0_wired() & MIPSR6_WIRED_LIMIT);
2989 	/* Status */
2990 	kvm_write_sw_gc0_status(cop0, ST0_BEV | ST0_ERL);
2991 	if (cpu_has_mips_r6)
2992 		kvm_change_sw_gc0_status(cop0, ST0_FR, read_gc0_status());
2993 	/* IntCtl */
2994 	kvm_write_sw_gc0_intctl(cop0, read_gc0_intctl() &
2995 				(INTCTLF_IPFDC | INTCTLF_IPPCI | INTCTLF_IPTI));
2996 	/* PRId */
2997 	kvm_write_sw_gc0_prid(cop0, boot_cpu_data.processor_id);
2998 	/* EBase */
2999 	kvm_write_sw_gc0_ebase(cop0, (s32)0x80000000 | vcpu->vcpu_id);
3000 	/* Config */
3001 	kvm_save_gc0_config(cop0);
3002 	/* architecturally writable (e.g. from guest) */
3003 	kvm_change_sw_gc0_config(cop0, CONF_CM_CMASK,
3004 				 _page_cachable_default >> _CACHE_SHIFT);
3005 	/* architecturally read only, but maybe writable from root */
3006 	kvm_change_sw_gc0_config(cop0, MIPS_CONF_MT, read_c0_config());
3007 	if (cpu_guest_has_conf1) {
3008 		kvm_set_sw_gc0_config(cop0, MIPS_CONF_M);
3009 		/* Config1 */
3010 		kvm_save_gc0_config1(cop0);
3011 		/* architecturally read only, but maybe writable from root */
3012 		kvm_clear_sw_gc0_config1(cop0, MIPS_CONF1_C2	|
3013 					       MIPS_CONF1_MD	|
3014 					       MIPS_CONF1_PC	|
3015 					       MIPS_CONF1_WR	|
3016 					       MIPS_CONF1_CA	|
3017 					       MIPS_CONF1_FP);
3018 	}
3019 	if (cpu_guest_has_conf2) {
3020 		kvm_set_sw_gc0_config1(cop0, MIPS_CONF_M);
3021 		/* Config2 */
3022 		kvm_save_gc0_config2(cop0);
3023 	}
3024 	if (cpu_guest_has_conf3) {
3025 		kvm_set_sw_gc0_config2(cop0, MIPS_CONF_M);
3026 		/* Config3 */
3027 		kvm_save_gc0_config3(cop0);
3028 		/* architecturally writable (e.g. from guest) */
3029 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_ISA_OE);
3030 		/* architecturally read only, but maybe writable from root */
3031 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_MSA	|
3032 					       MIPS_CONF3_BPG	|
3033 					       MIPS_CONF3_ULRI	|
3034 					       MIPS_CONF3_DSP	|
3035 					       MIPS_CONF3_CTXTC	|
3036 					       MIPS_CONF3_ITL	|
3037 					       MIPS_CONF3_LPA	|
3038 					       MIPS_CONF3_VEIC	|
3039 					       MIPS_CONF3_VINT	|
3040 					       MIPS_CONF3_SP	|
3041 					       MIPS_CONF3_CDMM	|
3042 					       MIPS_CONF3_MT	|
3043 					       MIPS_CONF3_SM	|
3044 					       MIPS_CONF3_TL);
3045 	}
3046 	if (cpu_guest_has_conf4) {
3047 		kvm_set_sw_gc0_config3(cop0, MIPS_CONF_M);
3048 		/* Config4 */
3049 		kvm_save_gc0_config4(cop0);
3050 	}
3051 	if (cpu_guest_has_conf5) {
3052 		kvm_set_sw_gc0_config4(cop0, MIPS_CONF_M);
3053 		/* Config5 */
3054 		kvm_save_gc0_config5(cop0);
3055 		/* architecturally writable (e.g. from guest) */
3056 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_K	|
3057 					       MIPS_CONF5_CV	|
3058 					       MIPS_CONF5_MSAEN	|
3059 					       MIPS_CONF5_UFE	|
3060 					       MIPS_CONF5_FRE	|
3061 					       MIPS_CONF5_SBRI	|
3062 					       MIPS_CONF5_UFR);
3063 		/* architecturally read only, but maybe writable from root */
3064 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_MRP);
3065 	}
3066 
3067 	if (cpu_guest_has_contextconfig) {
3068 		/* ContextConfig */
3069 		kvm_write_sw_gc0_contextconfig(cop0, 0x007ffff0);
3070 #ifdef CONFIG_64BIT
3071 		/* XContextConfig */
3072 		/* bits SEGBITS-13+3:4 set */
3073 		kvm_write_sw_gc0_xcontextconfig(cop0,
3074 					((1ull << (cpu_vmbits - 13)) - 1) << 4);
3075 #endif
3076 	}
3077 
3078 	/* Implementation dependent, use the legacy layout */
3079 	if (cpu_guest_has_segments) {
3080 		/* SegCtl0, SegCtl1, SegCtl2 */
3081 		kvm_write_sw_gc0_segctl0(cop0, 0x00200010);
3082 		kvm_write_sw_gc0_segctl1(cop0, 0x00000002 |
3083 				(_page_cachable_default >> _CACHE_SHIFT) <<
3084 						(16 + MIPS_SEGCFG_C_SHIFT));
3085 		kvm_write_sw_gc0_segctl2(cop0, 0x00380438);
3086 	}
3087 
3088 	/* reset HTW registers */
3089 	if (cpu_guest_has_htw && cpu_has_mips_r6) {
3090 		/* PWField */
3091 		kvm_write_sw_gc0_pwfield(cop0, 0x0c30c302);
3092 		/* PWSize */
3093 		kvm_write_sw_gc0_pwsize(cop0, 1 << MIPS_PWSIZE_PTW_SHIFT);
3094 	}
3095 
3096 	/* start with no pending virtual guest interrupts */
3097 	if (cpu_has_guestctl2)
3098 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] = 0;
3099 
3100 	/* Put PC at reset vector */
3101 	vcpu->arch.pc = CKSEG1ADDR(0x1fc00000);
3102 
3103 	return 0;
3104 }
3105 
3106 static void kvm_vz_flush_shadow_all(struct kvm *kvm)
3107 {
3108 	if (cpu_has_guestid) {
3109 		/* Flush GuestID for each VCPU individually */
3110 		kvm_flush_remote_tlbs(kvm);
3111 	} else {
3112 		/*
3113 		 * For each CPU there is a single GPA ASID used by all VCPUs in
3114 		 * the VM, so it doesn't make sense for the VCPUs to handle
3115 		 * invalidation of these ASIDs individually.
3116 		 *
3117 		 * Instead mark all CPUs as needing ASID invalidation in
3118 		 * asid_flush_mask, and just use kvm_flush_remote_tlbs(kvm) to
3119 		 * kick any running VCPUs so they check asid_flush_mask.
3120 		 */
3121 		cpumask_setall(&kvm->arch.asid_flush_mask);
3122 		kvm_flush_remote_tlbs(kvm);
3123 	}
3124 }
3125 
3126 static void kvm_vz_flush_shadow_memslot(struct kvm *kvm,
3127 					const struct kvm_memory_slot *slot)
3128 {
3129 	kvm_vz_flush_shadow_all(kvm);
3130 }
3131 
3132 static void kvm_vz_vcpu_reenter(struct kvm_run *run, struct kvm_vcpu *vcpu)
3133 {
3134 	int cpu = smp_processor_id();
3135 	int preserve_guest_tlb;
3136 
3137 	preserve_guest_tlb = kvm_vz_check_requests(vcpu, cpu);
3138 
3139 	if (preserve_guest_tlb)
3140 		kvm_vz_vcpu_save_wired(vcpu);
3141 
3142 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3143 
3144 	if (preserve_guest_tlb)
3145 		kvm_vz_vcpu_load_wired(vcpu);
3146 }
3147 
3148 static int kvm_vz_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
3149 {
3150 	int cpu = smp_processor_id();
3151 	int r;
3152 
3153 	kvm_vz_acquire_htimer(vcpu);
3154 	/* Check if we have any exceptions/interrupts pending */
3155 	kvm_mips_deliver_interrupts(vcpu, read_gc0_cause());
3156 
3157 	kvm_vz_check_requests(vcpu, cpu);
3158 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3159 	kvm_vz_vcpu_load_wired(vcpu);
3160 
3161 	r = vcpu->arch.vcpu_run(run, vcpu);
3162 
3163 	kvm_vz_vcpu_save_wired(vcpu);
3164 
3165 	return r;
3166 }
3167 
3168 static struct kvm_mips_callbacks kvm_vz_callbacks = {
3169 	.handle_cop_unusable = kvm_trap_vz_handle_cop_unusable,
3170 	.handle_tlb_mod = kvm_trap_vz_handle_tlb_st_miss,
3171 	.handle_tlb_ld_miss = kvm_trap_vz_handle_tlb_ld_miss,
3172 	.handle_tlb_st_miss = kvm_trap_vz_handle_tlb_st_miss,
3173 	.handle_addr_err_st = kvm_trap_vz_no_handler,
3174 	.handle_addr_err_ld = kvm_trap_vz_no_handler,
3175 	.handle_syscall = kvm_trap_vz_no_handler,
3176 	.handle_res_inst = kvm_trap_vz_no_handler,
3177 	.handle_break = kvm_trap_vz_no_handler,
3178 	.handle_msa_disabled = kvm_trap_vz_handle_msa_disabled,
3179 	.handle_guest_exit = kvm_trap_vz_handle_guest_exit,
3180 
3181 	.hardware_enable = kvm_vz_hardware_enable,
3182 	.hardware_disable = kvm_vz_hardware_disable,
3183 	.check_extension = kvm_vz_check_extension,
3184 	.vcpu_init = kvm_vz_vcpu_init,
3185 	.vcpu_uninit = kvm_vz_vcpu_uninit,
3186 	.vcpu_setup = kvm_vz_vcpu_setup,
3187 	.flush_shadow_all = kvm_vz_flush_shadow_all,
3188 	.flush_shadow_memslot = kvm_vz_flush_shadow_memslot,
3189 	.gva_to_gpa = kvm_vz_gva_to_gpa_cb,
3190 	.queue_timer_int = kvm_vz_queue_timer_int_cb,
3191 	.dequeue_timer_int = kvm_vz_dequeue_timer_int_cb,
3192 	.queue_io_int = kvm_vz_queue_io_int_cb,
3193 	.dequeue_io_int = kvm_vz_dequeue_io_int_cb,
3194 	.irq_deliver = kvm_vz_irq_deliver_cb,
3195 	.irq_clear = kvm_vz_irq_clear_cb,
3196 	.num_regs = kvm_vz_num_regs,
3197 	.copy_reg_indices = kvm_vz_copy_reg_indices,
3198 	.get_one_reg = kvm_vz_get_one_reg,
3199 	.set_one_reg = kvm_vz_set_one_reg,
3200 	.vcpu_load = kvm_vz_vcpu_load,
3201 	.vcpu_put = kvm_vz_vcpu_put,
3202 	.vcpu_run = kvm_vz_vcpu_run,
3203 	.vcpu_reenter = kvm_vz_vcpu_reenter,
3204 };
3205 
3206 int kvm_mips_emulation_init(struct kvm_mips_callbacks **install_callbacks)
3207 {
3208 	if (!cpu_has_vz)
3209 		return -ENODEV;
3210 
3211 	/*
3212 	 * VZ requires at least 2 KScratch registers, so it should have been
3213 	 * possible to allocate pgd_reg.
3214 	 */
3215 	if (WARN(pgd_reg == -1,
3216 		 "pgd_reg not allocated even though cpu_has_vz\n"))
3217 		return -ENODEV;
3218 
3219 	pr_info("Starting KVM with MIPS VZ extensions\n");
3220 
3221 	*install_callbacks = &kvm_vz_callbacks;
3222 	return 0;
3223 }
3224