xref: /openbmc/linux/arch/mips/kvm/vz.c (revision 273819a6)
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);
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_restore_stimer() - Restore soft timer state.
358  * @vcpu:	Virtual CPU.
359  * @compare:	CP0_Compare register value, restored by caller.
360  * @cause:	CP0_Cause register to restore.
361  *
362  * Restore VZ state relating to the soft timer.
363  */
364 static void _kvm_vz_restore_stimer(struct kvm_vcpu *vcpu, u32 compare,
365 				   u32 cause)
366 {
367 	/*
368 	 * Avoid spurious counter interrupts by setting Guest CP0_Count to just
369 	 * after Guest CP0_Compare.
370 	 */
371 	write_c0_gtoffset(compare - read_c0_count());
372 
373 	back_to_back_c0_hazard();
374 	write_gc0_cause(cause);
375 }
376 
377 /**
378  * kvm_vz_restore_timer() - Restore guest timer state.
379  * @vcpu:	Virtual CPU.
380  *
381  * Restore soft timer state from saved context.
382  */
383 static void kvm_vz_restore_timer(struct kvm_vcpu *vcpu)
384 {
385 	struct mips_coproc *cop0 = vcpu->arch.cop0;
386 	u32 cause, compare;
387 
388 	compare = kvm_read_sw_gc0_compare(cop0);
389 	cause = kvm_read_sw_gc0_cause(cop0);
390 
391 	write_gc0_compare(compare);
392 	_kvm_vz_restore_stimer(vcpu, compare, cause);
393 }
394 
395 /**
396  * kvm_vz_save_timer() - Save guest timer state.
397  * @vcpu:	Virtual CPU.
398  *
399  * Save VZ guest timer state.
400  */
401 static void kvm_vz_save_timer(struct kvm_vcpu *vcpu)
402 {
403 	struct mips_coproc *cop0 = vcpu->arch.cop0;
404 	u32 compare, cause;
405 
406 	compare = read_gc0_compare();
407 	cause = read_gc0_cause();
408 
409 	/* save timer-related state to VCPU context */
410 	kvm_write_sw_gc0_cause(cop0, cause);
411 	kvm_write_sw_gc0_compare(cop0, compare);
412 }
413 
414 /**
415  * is_eva_access() - Find whether an instruction is an EVA memory accessor.
416  * @inst:	32-bit instruction encoding.
417  *
418  * Finds whether @inst encodes an EVA memory access instruction, which would
419  * indicate that emulation of it should access the user mode address space
420  * instead of the kernel mode address space. This matters for MUSUK segments
421  * which are TLB mapped for user mode but unmapped for kernel mode.
422  *
423  * Returns:	Whether @inst encodes an EVA accessor instruction.
424  */
425 static bool is_eva_access(union mips_instruction inst)
426 {
427 	if (inst.spec3_format.opcode != spec3_op)
428 		return false;
429 
430 	switch (inst.spec3_format.func) {
431 	case lwle_op:
432 	case lwre_op:
433 	case cachee_op:
434 	case sbe_op:
435 	case she_op:
436 	case sce_op:
437 	case swe_op:
438 	case swle_op:
439 	case swre_op:
440 	case prefe_op:
441 	case lbue_op:
442 	case lhue_op:
443 	case lbe_op:
444 	case lhe_op:
445 	case lle_op:
446 	case lwe_op:
447 		return true;
448 	default:
449 		return false;
450 	}
451 }
452 
453 /**
454  * is_eva_am_mapped() - Find whether an access mode is mapped.
455  * @vcpu:	KVM VCPU state.
456  * @am:		3-bit encoded access mode.
457  * @eu:		Segment becomes unmapped and uncached when Status.ERL=1.
458  *
459  * Decode @am to find whether it encodes a mapped segment for the current VCPU
460  * state. Where necessary @eu and the actual instruction causing the fault are
461  * taken into account to make the decision.
462  *
463  * Returns:	Whether the VCPU faulted on a TLB mapped address.
464  */
465 static bool is_eva_am_mapped(struct kvm_vcpu *vcpu, unsigned int am, bool eu)
466 {
467 	u32 am_lookup;
468 	int err;
469 
470 	/*
471 	 * Interpret access control mode. We assume address errors will already
472 	 * have been caught by the guest, leaving us with:
473 	 *      AM      UM  SM  KM  31..24 23..16
474 	 * UK    0 000          Unm   0      0
475 	 * MK    1 001          TLB   1
476 	 * MSK   2 010      TLB TLB   1
477 	 * MUSK  3 011  TLB TLB TLB   1
478 	 * MUSUK 4 100  TLB TLB Unm   0      1
479 	 * USK   5 101      Unm Unm   0      0
480 	 * -     6 110                0      0
481 	 * UUSK  7 111  Unm Unm Unm   0      0
482 	 *
483 	 * We shift a magic value by AM across the sign bit to find if always
484 	 * TLB mapped, and if not shift by 8 again to find if it depends on KM.
485 	 */
486 	am_lookup = 0x70080000 << am;
487 	if ((s32)am_lookup < 0) {
488 		/*
489 		 * MK, MSK, MUSK
490 		 * Always TLB mapped, unless SegCtl.EU && ERL
491 		 */
492 		if (!eu || !(read_gc0_status() & ST0_ERL))
493 			return true;
494 	} else {
495 		am_lookup <<= 8;
496 		if ((s32)am_lookup < 0) {
497 			union mips_instruction inst;
498 			unsigned int status;
499 			u32 *opc;
500 
501 			/*
502 			 * MUSUK
503 			 * TLB mapped if not in kernel mode
504 			 */
505 			status = read_gc0_status();
506 			if (!(status & (ST0_EXL | ST0_ERL)) &&
507 			    (status & ST0_KSU))
508 				return true;
509 			/*
510 			 * EVA access instructions in kernel
511 			 * mode access user address space.
512 			 */
513 			opc = (u32 *)vcpu->arch.pc;
514 			if (vcpu->arch.host_cp0_cause & CAUSEF_BD)
515 				opc += 1;
516 			err = kvm_get_badinstr(opc, vcpu, &inst.word);
517 			if (!err && is_eva_access(inst))
518 				return true;
519 		}
520 	}
521 
522 	return false;
523 }
524 
525 /**
526  * kvm_vz_gva_to_gpa() - Convert valid GVA to GPA.
527  * @vcpu:	KVM VCPU state.
528  * @gva:	Guest virtual address to convert.
529  * @gpa:	Output guest physical address.
530  *
531  * Convert a guest virtual address (GVA) which is valid according to the guest
532  * context, to a guest physical address (GPA).
533  *
534  * Returns:	0 on success.
535  *		-errno on failure.
536  */
537 static int kvm_vz_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
538 			     unsigned long *gpa)
539 {
540 	u32 gva32 = gva;
541 	unsigned long segctl;
542 
543 	if ((long)gva == (s32)gva32) {
544 		/* Handle canonical 32-bit virtual address */
545 		if (cpu_guest_has_segments) {
546 			unsigned long mask, pa;
547 
548 			switch (gva32 >> 29) {
549 			case 0:
550 			case 1: /* CFG5 (1GB) */
551 				segctl = read_gc0_segctl2() >> 16;
552 				mask = (unsigned long)0xfc0000000ull;
553 				break;
554 			case 2:
555 			case 3: /* CFG4 (1GB) */
556 				segctl = read_gc0_segctl2();
557 				mask = (unsigned long)0xfc0000000ull;
558 				break;
559 			case 4: /* CFG3 (512MB) */
560 				segctl = read_gc0_segctl1() >> 16;
561 				mask = (unsigned long)0xfe0000000ull;
562 				break;
563 			case 5: /* CFG2 (512MB) */
564 				segctl = read_gc0_segctl1();
565 				mask = (unsigned long)0xfe0000000ull;
566 				break;
567 			case 6: /* CFG1 (512MB) */
568 				segctl = read_gc0_segctl0() >> 16;
569 				mask = (unsigned long)0xfe0000000ull;
570 				break;
571 			case 7: /* CFG0 (512MB) */
572 				segctl = read_gc0_segctl0();
573 				mask = (unsigned long)0xfe0000000ull;
574 				break;
575 			default:
576 				/*
577 				 * GCC 4.9 isn't smart enough to figure out that
578 				 * segctl and mask are always initialised.
579 				 */
580 				unreachable();
581 			}
582 
583 			if (is_eva_am_mapped(vcpu, (segctl >> 4) & 0x7,
584 					     segctl & 0x0008))
585 				goto tlb_mapped;
586 
587 			/* Unmapped, find guest physical address */
588 			pa = (segctl << 20) & mask;
589 			pa |= gva32 & ~mask;
590 			*gpa = pa;
591 			return 0;
592 		} else if ((s32)gva32 < (s32)0xc0000000) {
593 			/* legacy unmapped KSeg0 or KSeg1 */
594 			*gpa = gva32 & 0x1fffffff;
595 			return 0;
596 		}
597 #ifdef CONFIG_64BIT
598 	} else if ((gva & 0xc000000000000000) == 0x8000000000000000) {
599 		/* XKPHYS */
600 		if (cpu_guest_has_segments) {
601 			/*
602 			 * Each of the 8 regions can be overridden by SegCtl2.XR
603 			 * to use SegCtl1.XAM.
604 			 */
605 			segctl = read_gc0_segctl2();
606 			if (segctl & (1ull << (56 + ((gva >> 59) & 0x7)))) {
607 				segctl = read_gc0_segctl1();
608 				if (is_eva_am_mapped(vcpu, (segctl >> 59) & 0x7,
609 						     0))
610 					goto tlb_mapped;
611 			}
612 
613 		}
614 		/*
615 		 * Traditionally fully unmapped.
616 		 * Bits 61:59 specify the CCA, which we can just mask off here.
617 		 * Bits 58:PABITS should be zero, but we shouldn't have got here
618 		 * if it wasn't.
619 		 */
620 		*gpa = gva & 0x07ffffffffffffff;
621 		return 0;
622 #endif
623 	}
624 
625 tlb_mapped:
626 	return kvm_vz_guest_tlb_lookup(vcpu, gva, gpa);
627 }
628 
629 /**
630  * kvm_vz_badvaddr_to_gpa() - Convert GVA BadVAddr from root exception to GPA.
631  * @vcpu:	KVM VCPU state.
632  * @badvaddr:	Root BadVAddr.
633  * @gpa:	Output guest physical address.
634  *
635  * VZ implementations are permitted to report guest virtual addresses (GVA) in
636  * BadVAddr on a root exception during guest execution, instead of the more
637  * convenient guest physical addresses (GPA). When we get a GVA, this function
638  * converts it to a GPA, taking into account guest segmentation and guest TLB
639  * state.
640  *
641  * Returns:	0 on success.
642  *		-errno on failure.
643  */
644 static int kvm_vz_badvaddr_to_gpa(struct kvm_vcpu *vcpu, unsigned long badvaddr,
645 				  unsigned long *gpa)
646 {
647 	unsigned int gexccode = (vcpu->arch.host_cp0_guestctl0 &
648 				 MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
649 
650 	/* If BadVAddr is GPA, then all is well in the world */
651 	if (likely(gexccode == MIPS_GCTL0_GEXC_GPA)) {
652 		*gpa = badvaddr;
653 		return 0;
654 	}
655 
656 	/* Otherwise we'd expect it to be GVA ... */
657 	if (WARN(gexccode != MIPS_GCTL0_GEXC_GVA,
658 		 "Unexpected gexccode %#x\n", gexccode))
659 		return -EINVAL;
660 
661 	/* ... and we need to perform the GVA->GPA translation in software */
662 	return kvm_vz_gva_to_gpa(vcpu, badvaddr, gpa);
663 }
664 
665 static int kvm_trap_vz_no_handler(struct kvm_vcpu *vcpu)
666 {
667 	u32 *opc = (u32 *) vcpu->arch.pc;
668 	u32 cause = vcpu->arch.host_cp0_cause;
669 	u32 exccode = (cause & CAUSEF_EXCCODE) >> CAUSEB_EXCCODE;
670 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
671 	u32 inst = 0;
672 
673 	/*
674 	 *  Fetch the instruction.
675 	 */
676 	if (cause & CAUSEF_BD)
677 		opc += 1;
678 	kvm_get_badinstr(opc, vcpu, &inst);
679 
680 	kvm_err("Exception Code: %d not handled @ PC: %p, inst: 0x%08x BadVaddr: %#lx Status: %#x\n",
681 		exccode, opc, inst, badvaddr,
682 		read_gc0_status());
683 	kvm_arch_vcpu_dump_regs(vcpu);
684 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
685 	return RESUME_HOST;
686 }
687 
688 static enum emulation_result kvm_vz_gpsi_cop0(union mips_instruction inst,
689 					      u32 *opc, u32 cause,
690 					      struct kvm_run *run,
691 					      struct kvm_vcpu *vcpu)
692 {
693 	struct mips_coproc *cop0 = vcpu->arch.cop0;
694 	enum emulation_result er = EMULATE_DONE;
695 	u32 rt, rd, sel;
696 	unsigned long curr_pc;
697 	unsigned long val;
698 
699 	/*
700 	 * Update PC and hold onto current PC in case there is
701 	 * an error and we want to rollback the PC
702 	 */
703 	curr_pc = vcpu->arch.pc;
704 	er = update_pc(vcpu, cause);
705 	if (er == EMULATE_FAIL)
706 		return er;
707 
708 	if (inst.co_format.co) {
709 		switch (inst.co_format.func) {
710 		case wait_op:
711 			er = kvm_mips_emul_wait(vcpu);
712 			break;
713 		default:
714 			er = EMULATE_FAIL;
715 		}
716 	} else {
717 		rt = inst.c0r_format.rt;
718 		rd = inst.c0r_format.rd;
719 		sel = inst.c0r_format.sel;
720 
721 		switch (inst.c0r_format.rs) {
722 		case dmfc_op:
723 		case mfc_op:
724 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
725 			cop0->stat[rd][sel]++;
726 #endif
727 			if (rd == MIPS_CP0_COUNT &&
728 			    sel == 0) {			/* Count */
729 				val = kvm_mips_read_count(vcpu);
730 			} else if (rd == MIPS_CP0_COMPARE &&
731 				   sel == 0) {		/* Compare */
732 				val = read_gc0_compare();
733 			} else if (rd == MIPS_CP0_LLADDR &&
734 				   sel == 0) {		/* LLAddr */
735 				if (cpu_guest_has_rw_llb)
736 					val = read_gc0_lladdr() &
737 						MIPS_LLADDR_LLB;
738 				else
739 					val = 0;
740 			} else if ((rd == MIPS_CP0_PRID &&
741 				    (sel == 0 ||	/* PRid */
742 				     sel == 2 ||	/* CDMMBase */
743 				     sel == 3)) ||	/* CMGCRBase */
744 				   (rd == MIPS_CP0_STATUS &&
745 				    (sel == 2 ||	/* SRSCtl */
746 				     sel == 3)) ||	/* SRSMap */
747 				   (rd == MIPS_CP0_CONFIG &&
748 				    (sel == 7)) ||	/* Config7 */
749 				   (rd == MIPS_CP0_ERRCTL &&
750 				    (sel == 0))) {	/* ErrCtl */
751 				val = cop0->reg[rd][sel];
752 			} else {
753 				val = 0;
754 				er = EMULATE_FAIL;
755 			}
756 
757 			if (er != EMULATE_FAIL) {
758 				/* Sign extend */
759 				if (inst.c0r_format.rs == mfc_op)
760 					val = (int)val;
761 				vcpu->arch.gprs[rt] = val;
762 			}
763 
764 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mfc_op) ?
765 					KVM_TRACE_MFC0 : KVM_TRACE_DMFC0,
766 				      KVM_TRACE_COP0(rd, sel), val);
767 			break;
768 
769 		case dmtc_op:
770 		case mtc_op:
771 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
772 			cop0->stat[rd][sel]++;
773 #endif
774 			val = vcpu->arch.gprs[rt];
775 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mtc_op) ?
776 					KVM_TRACE_MTC0 : KVM_TRACE_DMTC0,
777 				      KVM_TRACE_COP0(rd, sel), val);
778 
779 			if (rd == MIPS_CP0_COUNT &&
780 			    sel == 0) {			/* Count */
781 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
782 			} else if (rd == MIPS_CP0_COMPARE &&
783 				   sel == 0) {		/* Compare */
784 				kvm_mips_write_compare(vcpu,
785 						       vcpu->arch.gprs[rt],
786 						       true);
787 			} else if (rd == MIPS_CP0_LLADDR &&
788 				   sel == 0) {		/* LLAddr */
789 				/*
790 				 * P5600 generates GPSI on guest MTC0 LLAddr.
791 				 * Only allow the guest to clear LLB.
792 				 */
793 				if (cpu_guest_has_rw_llb &&
794 				    !(val & MIPS_LLADDR_LLB))
795 					write_gc0_lladdr(0);
796 			} else if (rd == MIPS_CP0_ERRCTL &&
797 				   (sel == 0)) {	/* ErrCtl */
798 				/* ignore the written value */
799 			} else {
800 				er = EMULATE_FAIL;
801 			}
802 			break;
803 
804 		default:
805 			er = EMULATE_FAIL;
806 			break;
807 		}
808 	}
809 	/* Rollback PC only if emulation was unsuccessful */
810 	if (er == EMULATE_FAIL) {
811 		kvm_err("[%#lx]%s: unsupported cop0 instruction 0x%08x\n",
812 			curr_pc, __func__, inst.word);
813 
814 		vcpu->arch.pc = curr_pc;
815 	}
816 
817 	return er;
818 }
819 
820 static enum emulation_result kvm_vz_gpsi_cache(union mips_instruction inst,
821 					       u32 *opc, u32 cause,
822 					       struct kvm_run *run,
823 					       struct kvm_vcpu *vcpu)
824 {
825 	enum emulation_result er = EMULATE_DONE;
826 	u32 cache, op_inst, op, base;
827 	s16 offset;
828 	struct kvm_vcpu_arch *arch = &vcpu->arch;
829 	unsigned long va, curr_pc;
830 
831 	/*
832 	 * Update PC and hold onto current PC in case there is
833 	 * an error and we want to rollback the PC
834 	 */
835 	curr_pc = vcpu->arch.pc;
836 	er = update_pc(vcpu, cause);
837 	if (er == EMULATE_FAIL)
838 		return er;
839 
840 	base = inst.i_format.rs;
841 	op_inst = inst.i_format.rt;
842 	if (cpu_has_mips_r6)
843 		offset = inst.spec3_format.simmediate;
844 	else
845 		offset = inst.i_format.simmediate;
846 	cache = op_inst & CacheOp_Cache;
847 	op = op_inst & CacheOp_Op;
848 
849 	va = arch->gprs[base] + offset;
850 
851 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
852 		  cache, op, base, arch->gprs[base], offset);
853 
854 	/* Secondary or tirtiary cache ops ignored */
855 	if (cache != Cache_I && cache != Cache_D)
856 		return EMULATE_DONE;
857 
858 	switch (op_inst) {
859 	case Index_Invalidate_I:
860 		flush_icache_line_indexed(va);
861 		return EMULATE_DONE;
862 	case Index_Writeback_Inv_D:
863 		flush_dcache_line_indexed(va);
864 		return EMULATE_DONE;
865 	default:
866 		break;
867 	};
868 
869 	kvm_err("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
870 		curr_pc, vcpu->arch.gprs[31], cache, op, base, arch->gprs[base],
871 		offset);
872 	/* Rollback PC */
873 	vcpu->arch.pc = curr_pc;
874 
875 	return EMULATE_FAIL;
876 }
877 
878 static enum emulation_result kvm_trap_vz_handle_gpsi(u32 cause, u32 *opc,
879 						     struct kvm_vcpu *vcpu)
880 {
881 	enum emulation_result er = EMULATE_DONE;
882 	struct kvm_vcpu_arch *arch = &vcpu->arch;
883 	struct kvm_run *run = vcpu->run;
884 	union mips_instruction inst;
885 	int rd, rt, sel;
886 	int err;
887 
888 	/*
889 	 *  Fetch the instruction.
890 	 */
891 	if (cause & CAUSEF_BD)
892 		opc += 1;
893 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
894 	if (err)
895 		return EMULATE_FAIL;
896 
897 	switch (inst.r_format.opcode) {
898 	case cop0_op:
899 		er = kvm_vz_gpsi_cop0(inst, opc, cause, run, vcpu);
900 		break;
901 #ifndef CONFIG_CPU_MIPSR6
902 	case cache_op:
903 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
904 		er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
905 		break;
906 #endif
907 	case spec3_op:
908 		switch (inst.spec3_format.func) {
909 #ifdef CONFIG_CPU_MIPSR6
910 		case cache6_op:
911 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
912 			er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
913 			break;
914 #endif
915 		case rdhwr_op:
916 			if (inst.r_format.rs || (inst.r_format.re >> 3))
917 				goto unknown;
918 
919 			rd = inst.r_format.rd;
920 			rt = inst.r_format.rt;
921 			sel = inst.r_format.re & 0x7;
922 
923 			switch (rd) {
924 			case MIPS_HWR_CC:	/* Read count register */
925 				arch->gprs[rt] =
926 					(long)(int)kvm_mips_read_count(vcpu);
927 				break;
928 			default:
929 				trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
930 					      KVM_TRACE_HWR(rd, sel), 0);
931 				goto unknown;
932 			};
933 
934 			trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
935 				      KVM_TRACE_HWR(rd, sel), arch->gprs[rt]);
936 
937 			er = update_pc(vcpu, cause);
938 			break;
939 		default:
940 			goto unknown;
941 		};
942 		break;
943 unknown:
944 
945 	default:
946 		kvm_err("GPSI exception not supported (%p/%#x)\n",
947 				opc, inst.word);
948 		kvm_arch_vcpu_dump_regs(vcpu);
949 		er = EMULATE_FAIL;
950 		break;
951 	}
952 
953 	return er;
954 }
955 
956 static enum emulation_result kvm_trap_vz_handle_gsfc(u32 cause, u32 *opc,
957 						     struct kvm_vcpu *vcpu)
958 {
959 	enum emulation_result er = EMULATE_DONE;
960 	struct kvm_vcpu_arch *arch = &vcpu->arch;
961 	union mips_instruction inst;
962 	int err;
963 
964 	/*
965 	 *  Fetch the instruction.
966 	 */
967 	if (cause & CAUSEF_BD)
968 		opc += 1;
969 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
970 	if (err)
971 		return EMULATE_FAIL;
972 
973 	/* complete MTC0 on behalf of guest and advance EPC */
974 	if (inst.c0r_format.opcode == cop0_op &&
975 	    inst.c0r_format.rs == mtc_op &&
976 	    inst.c0r_format.z == 0) {
977 		int rt = inst.c0r_format.rt;
978 		int rd = inst.c0r_format.rd;
979 		int sel = inst.c0r_format.sel;
980 		unsigned int val = arch->gprs[rt];
981 		unsigned int old_val, change;
982 
983 		trace_kvm_hwr(vcpu, KVM_TRACE_MTC0, KVM_TRACE_COP0(rd, sel),
984 			      val);
985 
986 		if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
987 			/* FR bit should read as zero if no FPU */
988 			if (!kvm_mips_guest_has_fpu(&vcpu->arch))
989 				val &= ~(ST0_CU1 | ST0_FR);
990 
991 			/*
992 			 * Also don't allow FR to be set if host doesn't support
993 			 * it.
994 			 */
995 			if (!(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
996 				val &= ~ST0_FR;
997 
998 			old_val = read_gc0_status();
999 			change = val ^ old_val;
1000 
1001 			if (change & ST0_FR) {
1002 				/*
1003 				 * FPU and Vector register state is made
1004 				 * UNPREDICTABLE by a change of FR, so don't
1005 				 * even bother saving it.
1006 				 */
1007 				kvm_drop_fpu(vcpu);
1008 			}
1009 
1010 			/*
1011 			 * If MSA state is already live, it is undefined how it
1012 			 * interacts with FR=0 FPU state, and we don't want to
1013 			 * hit reserved instruction exceptions trying to save
1014 			 * the MSA state later when CU=1 && FR=1, so play it
1015 			 * safe and save it first.
1016 			 */
1017 			if (change & ST0_CU1 && !(val & ST0_FR) &&
1018 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1019 				kvm_lose_fpu(vcpu);
1020 
1021 			write_gc0_status(val);
1022 		} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1023 			u32 old_cause = read_gc0_cause();
1024 			u32 change = old_cause ^ val;
1025 
1026 			/* DC bit enabling/disabling timer? */
1027 			if (change & CAUSEF_DC) {
1028 				if (val & CAUSEF_DC)
1029 					kvm_mips_count_disable_cause(vcpu);
1030 				else
1031 					kvm_mips_count_enable_cause(vcpu);
1032 			}
1033 
1034 			/* Only certain bits are RW to the guest */
1035 			change &= (CAUSEF_DC | CAUSEF_IV | CAUSEF_WP |
1036 				   CAUSEF_IP0 | CAUSEF_IP1);
1037 
1038 			/* WP can only be cleared */
1039 			change &= ~CAUSEF_WP | old_cause;
1040 
1041 			write_gc0_cause(old_cause ^ change);
1042 		} else if ((rd == MIPS_CP0_STATUS) && (sel == 1)) { /* IntCtl */
1043 			write_gc0_intctl(val);
1044 		} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1045 			old_val = read_gc0_config5();
1046 			change = val ^ old_val;
1047 			/* Handle changes in FPU/MSA modes */
1048 			preempt_disable();
1049 
1050 			/*
1051 			 * Propagate FRE changes immediately if the FPU
1052 			 * context is already loaded.
1053 			 */
1054 			if (change & MIPS_CONF5_FRE &&
1055 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1056 				change_c0_config5(MIPS_CONF5_FRE, val);
1057 
1058 			preempt_enable();
1059 
1060 			val = old_val ^
1061 				(change & kvm_vz_config5_guest_wrmask(vcpu));
1062 			write_gc0_config5(val);
1063 		} else {
1064 			kvm_err("Handle GSFC, unsupported field change @ %p: %#x\n",
1065 			    opc, inst.word);
1066 			er = EMULATE_FAIL;
1067 		}
1068 
1069 		if (er != EMULATE_FAIL)
1070 			er = update_pc(vcpu, cause);
1071 	} else {
1072 		kvm_err("Handle GSFC, unrecognized instruction @ %p: %#x\n",
1073 			opc, inst.word);
1074 		er = EMULATE_FAIL;
1075 	}
1076 
1077 	return er;
1078 }
1079 
1080 static enum emulation_result kvm_trap_vz_handle_hc(u32 cause, u32 *opc,
1081 						   struct kvm_vcpu *vcpu)
1082 {
1083 	enum emulation_result er;
1084 	union mips_instruction inst;
1085 	unsigned long curr_pc;
1086 	int err;
1087 
1088 	if (cause & CAUSEF_BD)
1089 		opc += 1;
1090 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1091 	if (err)
1092 		return EMULATE_FAIL;
1093 
1094 	/*
1095 	 * Update PC and hold onto current PC in case there is
1096 	 * an error and we want to rollback the PC
1097 	 */
1098 	curr_pc = vcpu->arch.pc;
1099 	er = update_pc(vcpu, cause);
1100 	if (er == EMULATE_FAIL)
1101 		return er;
1102 
1103 	er = kvm_mips_emul_hypcall(vcpu, inst);
1104 	if (er == EMULATE_FAIL)
1105 		vcpu->arch.pc = curr_pc;
1106 
1107 	return er;
1108 }
1109 
1110 static enum emulation_result kvm_trap_vz_no_handler_guest_exit(u32 gexccode,
1111 							u32 cause,
1112 							u32 *opc,
1113 							struct kvm_vcpu *vcpu)
1114 {
1115 	u32 inst;
1116 
1117 	/*
1118 	 *  Fetch the instruction.
1119 	 */
1120 	if (cause & CAUSEF_BD)
1121 		opc += 1;
1122 	kvm_get_badinstr(opc, vcpu, &inst);
1123 
1124 	kvm_err("Guest Exception Code: %d not yet handled @ PC: %p, inst: 0x%08x  Status: %#x\n",
1125 		gexccode, opc, inst, read_gc0_status());
1126 
1127 	return EMULATE_FAIL;
1128 }
1129 
1130 static int kvm_trap_vz_handle_guest_exit(struct kvm_vcpu *vcpu)
1131 {
1132 	u32 *opc = (u32 *) vcpu->arch.pc;
1133 	u32 cause = vcpu->arch.host_cp0_cause;
1134 	enum emulation_result er = EMULATE_DONE;
1135 	u32 gexccode = (vcpu->arch.host_cp0_guestctl0 &
1136 			MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
1137 	int ret = RESUME_GUEST;
1138 
1139 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_GEXCCODE_BASE + gexccode);
1140 	switch (gexccode) {
1141 	case MIPS_GCTL0_GEXC_GPSI:
1142 		++vcpu->stat.vz_gpsi_exits;
1143 		er = kvm_trap_vz_handle_gpsi(cause, opc, vcpu);
1144 		break;
1145 	case MIPS_GCTL0_GEXC_GSFC:
1146 		++vcpu->stat.vz_gsfc_exits;
1147 		er = kvm_trap_vz_handle_gsfc(cause, opc, vcpu);
1148 		break;
1149 	case MIPS_GCTL0_GEXC_HC:
1150 		++vcpu->stat.vz_hc_exits;
1151 		er = kvm_trap_vz_handle_hc(cause, opc, vcpu);
1152 		break;
1153 	case MIPS_GCTL0_GEXC_GRR:
1154 		++vcpu->stat.vz_grr_exits;
1155 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1156 						       vcpu);
1157 		break;
1158 	case MIPS_GCTL0_GEXC_GVA:
1159 		++vcpu->stat.vz_gva_exits;
1160 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1161 						       vcpu);
1162 		break;
1163 	case MIPS_GCTL0_GEXC_GHFC:
1164 		++vcpu->stat.vz_ghfc_exits;
1165 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1166 						       vcpu);
1167 		break;
1168 	case MIPS_GCTL0_GEXC_GPA:
1169 		++vcpu->stat.vz_gpa_exits;
1170 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1171 						       vcpu);
1172 		break;
1173 	default:
1174 		++vcpu->stat.vz_resvd_exits;
1175 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1176 						       vcpu);
1177 		break;
1178 
1179 	}
1180 
1181 	if (er == EMULATE_DONE) {
1182 		ret = RESUME_GUEST;
1183 	} else if (er == EMULATE_HYPERCALL) {
1184 		ret = kvm_mips_handle_hypcall(vcpu);
1185 	} else {
1186 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1187 		ret = RESUME_HOST;
1188 	}
1189 	return ret;
1190 }
1191 
1192 /**
1193  * kvm_trap_vz_handle_cop_unusuable() - Guest used unusable coprocessor.
1194  * @vcpu:	Virtual CPU context.
1195  *
1196  * Handle when the guest attempts to use a coprocessor which hasn't been allowed
1197  * by the root context.
1198  */
1199 static int kvm_trap_vz_handle_cop_unusable(struct kvm_vcpu *vcpu)
1200 {
1201 	struct kvm_run *run = vcpu->run;
1202 	u32 cause = vcpu->arch.host_cp0_cause;
1203 	enum emulation_result er = EMULATE_FAIL;
1204 	int ret = RESUME_GUEST;
1205 
1206 	if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 1) {
1207 		/*
1208 		 * If guest FPU not present, the FPU operation should have been
1209 		 * treated as a reserved instruction!
1210 		 * If FPU already in use, we shouldn't get this at all.
1211 		 */
1212 		if (WARN_ON(!kvm_mips_guest_has_fpu(&vcpu->arch) ||
1213 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1214 			preempt_enable();
1215 			return EMULATE_FAIL;
1216 		}
1217 
1218 		kvm_own_fpu(vcpu);
1219 		er = EMULATE_DONE;
1220 	}
1221 	/* other coprocessors not handled */
1222 
1223 	switch (er) {
1224 	case EMULATE_DONE:
1225 		ret = RESUME_GUEST;
1226 		break;
1227 
1228 	case EMULATE_FAIL:
1229 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1230 		ret = RESUME_HOST;
1231 		break;
1232 
1233 	default:
1234 		BUG();
1235 	}
1236 	return ret;
1237 }
1238 
1239 /**
1240  * kvm_trap_vz_handle_msa_disabled() - Guest used MSA while disabled in root.
1241  * @vcpu:	Virtual CPU context.
1242  *
1243  * Handle when the guest attempts to use MSA when it is disabled in the root
1244  * context.
1245  */
1246 static int kvm_trap_vz_handle_msa_disabled(struct kvm_vcpu *vcpu)
1247 {
1248 	struct kvm_run *run = vcpu->run;
1249 
1250 	/*
1251 	 * If MSA not present or not exposed to guest or FR=0, the MSA operation
1252 	 * should have been treated as a reserved instruction!
1253 	 * Same if CU1=1, FR=0.
1254 	 * If MSA already in use, we shouldn't get this at all.
1255 	 */
1256 	if (!kvm_mips_guest_has_msa(&vcpu->arch) ||
1257 	    (read_gc0_status() & (ST0_CU1 | ST0_FR)) == ST0_CU1 ||
1258 	    !(read_gc0_config5() & MIPS_CONF5_MSAEN) ||
1259 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1260 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1261 		return RESUME_HOST;
1262 	}
1263 
1264 	kvm_own_msa(vcpu);
1265 
1266 	return RESUME_GUEST;
1267 }
1268 
1269 static int kvm_trap_vz_handle_tlb_ld_miss(struct kvm_vcpu *vcpu)
1270 {
1271 	struct kvm_run *run = vcpu->run;
1272 	u32 *opc = (u32 *) vcpu->arch.pc;
1273 	u32 cause = vcpu->arch.host_cp0_cause;
1274 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1275 	union mips_instruction inst;
1276 	enum emulation_result er = EMULATE_DONE;
1277 	int err, ret = RESUME_GUEST;
1278 
1279 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, false)) {
1280 		/* A code fetch fault doesn't count as an MMIO */
1281 		if (kvm_is_ifetch_fault(&vcpu->arch)) {
1282 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1283 			return RESUME_HOST;
1284 		}
1285 
1286 		/* Fetch the instruction */
1287 		if (cause & CAUSEF_BD)
1288 			opc += 1;
1289 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1290 		if (err) {
1291 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1292 			return RESUME_HOST;
1293 		}
1294 
1295 		/* Treat as MMIO */
1296 		er = kvm_mips_emulate_load(inst, cause, run, vcpu);
1297 		if (er == EMULATE_FAIL) {
1298 			kvm_err("Guest Emulate Load from MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1299 				opc, badvaddr);
1300 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1301 		}
1302 	}
1303 
1304 	if (er == EMULATE_DONE) {
1305 		ret = RESUME_GUEST;
1306 	} else if (er == EMULATE_DO_MMIO) {
1307 		run->exit_reason = KVM_EXIT_MMIO;
1308 		ret = RESUME_HOST;
1309 	} else {
1310 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1311 		ret = RESUME_HOST;
1312 	}
1313 	return ret;
1314 }
1315 
1316 static int kvm_trap_vz_handle_tlb_st_miss(struct kvm_vcpu *vcpu)
1317 {
1318 	struct kvm_run *run = vcpu->run;
1319 	u32 *opc = (u32 *) vcpu->arch.pc;
1320 	u32 cause = vcpu->arch.host_cp0_cause;
1321 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1322 	union mips_instruction inst;
1323 	enum emulation_result er = EMULATE_DONE;
1324 	int err;
1325 	int ret = RESUME_GUEST;
1326 
1327 	/* Just try the access again if we couldn't do the translation */
1328 	if (kvm_vz_badvaddr_to_gpa(vcpu, badvaddr, &badvaddr))
1329 		return RESUME_GUEST;
1330 	vcpu->arch.host_cp0_badvaddr = badvaddr;
1331 
1332 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, true)) {
1333 		/* Fetch the instruction */
1334 		if (cause & CAUSEF_BD)
1335 			opc += 1;
1336 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1337 		if (err) {
1338 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1339 			return RESUME_HOST;
1340 		}
1341 
1342 		/* Treat as MMIO */
1343 		er = kvm_mips_emulate_store(inst, cause, run, vcpu);
1344 		if (er == EMULATE_FAIL) {
1345 			kvm_err("Guest Emulate Store to MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1346 				opc, badvaddr);
1347 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1348 		}
1349 	}
1350 
1351 	if (er == EMULATE_DONE) {
1352 		ret = RESUME_GUEST;
1353 	} else if (er == EMULATE_DO_MMIO) {
1354 		run->exit_reason = KVM_EXIT_MMIO;
1355 		ret = RESUME_HOST;
1356 	} else {
1357 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1358 		ret = RESUME_HOST;
1359 	}
1360 	return ret;
1361 }
1362 
1363 static u64 kvm_vz_get_one_regs[] = {
1364 	KVM_REG_MIPS_CP0_INDEX,
1365 	KVM_REG_MIPS_CP0_ENTRYLO0,
1366 	KVM_REG_MIPS_CP0_ENTRYLO1,
1367 	KVM_REG_MIPS_CP0_CONTEXT,
1368 	KVM_REG_MIPS_CP0_PAGEMASK,
1369 	KVM_REG_MIPS_CP0_PAGEGRAIN,
1370 	KVM_REG_MIPS_CP0_WIRED,
1371 	KVM_REG_MIPS_CP0_HWRENA,
1372 	KVM_REG_MIPS_CP0_BADVADDR,
1373 	KVM_REG_MIPS_CP0_COUNT,
1374 	KVM_REG_MIPS_CP0_ENTRYHI,
1375 	KVM_REG_MIPS_CP0_COMPARE,
1376 	KVM_REG_MIPS_CP0_STATUS,
1377 	KVM_REG_MIPS_CP0_INTCTL,
1378 	KVM_REG_MIPS_CP0_CAUSE,
1379 	KVM_REG_MIPS_CP0_EPC,
1380 	KVM_REG_MIPS_CP0_PRID,
1381 	KVM_REG_MIPS_CP0_EBASE,
1382 	KVM_REG_MIPS_CP0_CONFIG,
1383 	KVM_REG_MIPS_CP0_CONFIG1,
1384 	KVM_REG_MIPS_CP0_CONFIG2,
1385 	KVM_REG_MIPS_CP0_CONFIG3,
1386 	KVM_REG_MIPS_CP0_CONFIG4,
1387 	KVM_REG_MIPS_CP0_CONFIG5,
1388 #ifdef CONFIG_64BIT
1389 	KVM_REG_MIPS_CP0_XCONTEXT,
1390 #endif
1391 	KVM_REG_MIPS_CP0_ERROREPC,
1392 
1393 	KVM_REG_MIPS_COUNT_CTL,
1394 	KVM_REG_MIPS_COUNT_RESUME,
1395 	KVM_REG_MIPS_COUNT_HZ,
1396 };
1397 
1398 static u64 kvm_vz_get_one_regs_contextconfig[] = {
1399 	KVM_REG_MIPS_CP0_CONTEXTCONFIG,
1400 #ifdef CONFIG_64BIT
1401 	KVM_REG_MIPS_CP0_XCONTEXTCONFIG,
1402 #endif
1403 };
1404 
1405 static u64 kvm_vz_get_one_regs_segments[] = {
1406 	KVM_REG_MIPS_CP0_SEGCTL0,
1407 	KVM_REG_MIPS_CP0_SEGCTL1,
1408 	KVM_REG_MIPS_CP0_SEGCTL2,
1409 };
1410 
1411 static u64 kvm_vz_get_one_regs_htw[] = {
1412 	KVM_REG_MIPS_CP0_PWBASE,
1413 	KVM_REG_MIPS_CP0_PWFIELD,
1414 	KVM_REG_MIPS_CP0_PWSIZE,
1415 	KVM_REG_MIPS_CP0_PWCTL,
1416 };
1417 
1418 static u64 kvm_vz_get_one_regs_kscratch[] = {
1419 	KVM_REG_MIPS_CP0_KSCRATCH1,
1420 	KVM_REG_MIPS_CP0_KSCRATCH2,
1421 	KVM_REG_MIPS_CP0_KSCRATCH3,
1422 	KVM_REG_MIPS_CP0_KSCRATCH4,
1423 	KVM_REG_MIPS_CP0_KSCRATCH5,
1424 	KVM_REG_MIPS_CP0_KSCRATCH6,
1425 };
1426 
1427 static unsigned long kvm_vz_num_regs(struct kvm_vcpu *vcpu)
1428 {
1429 	unsigned long ret;
1430 
1431 	ret = ARRAY_SIZE(kvm_vz_get_one_regs);
1432 	if (cpu_guest_has_userlocal)
1433 		++ret;
1434 	if (cpu_guest_has_badinstr)
1435 		++ret;
1436 	if (cpu_guest_has_badinstrp)
1437 		++ret;
1438 	if (cpu_guest_has_contextconfig)
1439 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1440 	if (cpu_guest_has_segments)
1441 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1442 	if (cpu_guest_has_htw)
1443 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1444 	ret += __arch_hweight8(cpu_data[0].guest.kscratch_mask);
1445 
1446 	return ret;
1447 }
1448 
1449 static int kvm_vz_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
1450 {
1451 	u64 index;
1452 	unsigned int i;
1453 
1454 	if (copy_to_user(indices, kvm_vz_get_one_regs,
1455 			 sizeof(kvm_vz_get_one_regs)))
1456 		return -EFAULT;
1457 	indices += ARRAY_SIZE(kvm_vz_get_one_regs);
1458 
1459 	if (cpu_guest_has_userlocal) {
1460 		index = KVM_REG_MIPS_CP0_USERLOCAL;
1461 		if (copy_to_user(indices, &index, sizeof(index)))
1462 			return -EFAULT;
1463 		++indices;
1464 	}
1465 	if (cpu_guest_has_badinstr) {
1466 		index = KVM_REG_MIPS_CP0_BADINSTR;
1467 		if (copy_to_user(indices, &index, sizeof(index)))
1468 			return -EFAULT;
1469 		++indices;
1470 	}
1471 	if (cpu_guest_has_badinstrp) {
1472 		index = KVM_REG_MIPS_CP0_BADINSTRP;
1473 		if (copy_to_user(indices, &index, sizeof(index)))
1474 			return -EFAULT;
1475 		++indices;
1476 	}
1477 	if (cpu_guest_has_contextconfig) {
1478 		if (copy_to_user(indices, kvm_vz_get_one_regs_contextconfig,
1479 				 sizeof(kvm_vz_get_one_regs_contextconfig)))
1480 			return -EFAULT;
1481 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1482 	}
1483 	if (cpu_guest_has_segments) {
1484 		if (copy_to_user(indices, kvm_vz_get_one_regs_segments,
1485 				 sizeof(kvm_vz_get_one_regs_segments)))
1486 			return -EFAULT;
1487 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1488 	}
1489 	if (cpu_guest_has_htw) {
1490 		if (copy_to_user(indices, kvm_vz_get_one_regs_htw,
1491 				 sizeof(kvm_vz_get_one_regs_htw)))
1492 			return -EFAULT;
1493 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1494 	}
1495 	for (i = 0; i < 6; ++i) {
1496 		if (!cpu_guest_has_kscr(i + 2))
1497 			continue;
1498 
1499 		if (copy_to_user(indices, &kvm_vz_get_one_regs_kscratch[i],
1500 				 sizeof(kvm_vz_get_one_regs_kscratch[i])))
1501 			return -EFAULT;
1502 		++indices;
1503 	}
1504 
1505 	return 0;
1506 }
1507 
1508 static inline s64 entrylo_kvm_to_user(unsigned long v)
1509 {
1510 	s64 mask, ret = v;
1511 
1512 	if (BITS_PER_LONG == 32) {
1513 		/*
1514 		 * KVM API exposes 64-bit version of the register, so move the
1515 		 * RI/XI bits up into place.
1516 		 */
1517 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1518 		ret &= ~mask;
1519 		ret |= ((s64)v & mask) << 32;
1520 	}
1521 	return ret;
1522 }
1523 
1524 static inline unsigned long entrylo_user_to_kvm(s64 v)
1525 {
1526 	unsigned long mask, ret = v;
1527 
1528 	if (BITS_PER_LONG == 32) {
1529 		/*
1530 		 * KVM API exposes 64-bit versiono of the register, so move the
1531 		 * RI/XI bits down into place.
1532 		 */
1533 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1534 		ret &= ~mask;
1535 		ret |= (v >> 32) & mask;
1536 	}
1537 	return ret;
1538 }
1539 
1540 static int kvm_vz_get_one_reg(struct kvm_vcpu *vcpu,
1541 			      const struct kvm_one_reg *reg,
1542 			      s64 *v)
1543 {
1544 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1545 	unsigned int idx;
1546 
1547 	switch (reg->id) {
1548 	case KVM_REG_MIPS_CP0_INDEX:
1549 		*v = (long)read_gc0_index();
1550 		break;
1551 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1552 		*v = entrylo_kvm_to_user(read_gc0_entrylo0());
1553 		break;
1554 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1555 		*v = entrylo_kvm_to_user(read_gc0_entrylo1());
1556 		break;
1557 	case KVM_REG_MIPS_CP0_CONTEXT:
1558 		*v = (long)read_gc0_context();
1559 		break;
1560 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1561 		if (!cpu_guest_has_contextconfig)
1562 			return -EINVAL;
1563 		*v = read_gc0_contextconfig();
1564 		break;
1565 	case KVM_REG_MIPS_CP0_USERLOCAL:
1566 		if (!cpu_guest_has_userlocal)
1567 			return -EINVAL;
1568 		*v = read_gc0_userlocal();
1569 		break;
1570 #ifdef CONFIG_64BIT
1571 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1572 		if (!cpu_guest_has_contextconfig)
1573 			return -EINVAL;
1574 		*v = read_gc0_xcontextconfig();
1575 		break;
1576 #endif
1577 	case KVM_REG_MIPS_CP0_PAGEMASK:
1578 		*v = (long)read_gc0_pagemask();
1579 		break;
1580 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1581 		*v = (long)read_gc0_pagegrain();
1582 		break;
1583 	case KVM_REG_MIPS_CP0_SEGCTL0:
1584 		if (!cpu_guest_has_segments)
1585 			return -EINVAL;
1586 		*v = read_gc0_segctl0();
1587 		break;
1588 	case KVM_REG_MIPS_CP0_SEGCTL1:
1589 		if (!cpu_guest_has_segments)
1590 			return -EINVAL;
1591 		*v = read_gc0_segctl1();
1592 		break;
1593 	case KVM_REG_MIPS_CP0_SEGCTL2:
1594 		if (!cpu_guest_has_segments)
1595 			return -EINVAL;
1596 		*v = read_gc0_segctl2();
1597 		break;
1598 	case KVM_REG_MIPS_CP0_PWBASE:
1599 		if (!cpu_guest_has_htw)
1600 			return -EINVAL;
1601 		*v = read_gc0_pwbase();
1602 		break;
1603 	case KVM_REG_MIPS_CP0_PWFIELD:
1604 		if (!cpu_guest_has_htw)
1605 			return -EINVAL;
1606 		*v = read_gc0_pwfield();
1607 		break;
1608 	case KVM_REG_MIPS_CP0_PWSIZE:
1609 		if (!cpu_guest_has_htw)
1610 			return -EINVAL;
1611 		*v = read_gc0_pwsize();
1612 		break;
1613 	case KVM_REG_MIPS_CP0_WIRED:
1614 		*v = (long)read_gc0_wired();
1615 		break;
1616 	case KVM_REG_MIPS_CP0_PWCTL:
1617 		if (!cpu_guest_has_htw)
1618 			return -EINVAL;
1619 		*v = read_gc0_pwctl();
1620 		break;
1621 	case KVM_REG_MIPS_CP0_HWRENA:
1622 		*v = (long)read_gc0_hwrena();
1623 		break;
1624 	case KVM_REG_MIPS_CP0_BADVADDR:
1625 		*v = (long)read_gc0_badvaddr();
1626 		break;
1627 	case KVM_REG_MIPS_CP0_BADINSTR:
1628 		if (!cpu_guest_has_badinstr)
1629 			return -EINVAL;
1630 		*v = read_gc0_badinstr();
1631 		break;
1632 	case KVM_REG_MIPS_CP0_BADINSTRP:
1633 		if (!cpu_guest_has_badinstrp)
1634 			return -EINVAL;
1635 		*v = read_gc0_badinstrp();
1636 		break;
1637 	case KVM_REG_MIPS_CP0_COUNT:
1638 		*v = kvm_mips_read_count(vcpu);
1639 		break;
1640 	case KVM_REG_MIPS_CP0_ENTRYHI:
1641 		*v = (long)read_gc0_entryhi();
1642 		break;
1643 	case KVM_REG_MIPS_CP0_COMPARE:
1644 		*v = (long)read_gc0_compare();
1645 		break;
1646 	case KVM_REG_MIPS_CP0_STATUS:
1647 		*v = (long)read_gc0_status();
1648 		break;
1649 	case KVM_REG_MIPS_CP0_INTCTL:
1650 		*v = read_gc0_intctl();
1651 		break;
1652 	case KVM_REG_MIPS_CP0_CAUSE:
1653 		*v = (long)read_gc0_cause();
1654 		break;
1655 	case KVM_REG_MIPS_CP0_EPC:
1656 		*v = (long)read_gc0_epc();
1657 		break;
1658 	case KVM_REG_MIPS_CP0_PRID:
1659 		*v = (long)kvm_read_c0_guest_prid(cop0);
1660 		break;
1661 	case KVM_REG_MIPS_CP0_EBASE:
1662 		*v = kvm_vz_read_gc0_ebase();
1663 		break;
1664 	case KVM_REG_MIPS_CP0_CONFIG:
1665 		*v = read_gc0_config();
1666 		break;
1667 	case KVM_REG_MIPS_CP0_CONFIG1:
1668 		if (!cpu_guest_has_conf1)
1669 			return -EINVAL;
1670 		*v = read_gc0_config1();
1671 		break;
1672 	case KVM_REG_MIPS_CP0_CONFIG2:
1673 		if (!cpu_guest_has_conf2)
1674 			return -EINVAL;
1675 		*v = read_gc0_config2();
1676 		break;
1677 	case KVM_REG_MIPS_CP0_CONFIG3:
1678 		if (!cpu_guest_has_conf3)
1679 			return -EINVAL;
1680 		*v = read_gc0_config3();
1681 		break;
1682 	case KVM_REG_MIPS_CP0_CONFIG4:
1683 		if (!cpu_guest_has_conf4)
1684 			return -EINVAL;
1685 		*v = read_gc0_config4();
1686 		break;
1687 	case KVM_REG_MIPS_CP0_CONFIG5:
1688 		if (!cpu_guest_has_conf5)
1689 			return -EINVAL;
1690 		*v = read_gc0_config5();
1691 		break;
1692 #ifdef CONFIG_64BIT
1693 	case KVM_REG_MIPS_CP0_XCONTEXT:
1694 		*v = read_gc0_xcontext();
1695 		break;
1696 #endif
1697 	case KVM_REG_MIPS_CP0_ERROREPC:
1698 		*v = (long)read_gc0_errorepc();
1699 		break;
1700 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
1701 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
1702 		if (!cpu_guest_has_kscr(idx))
1703 			return -EINVAL;
1704 		switch (idx) {
1705 		case 2:
1706 			*v = (long)read_gc0_kscratch1();
1707 			break;
1708 		case 3:
1709 			*v = (long)read_gc0_kscratch2();
1710 			break;
1711 		case 4:
1712 			*v = (long)read_gc0_kscratch3();
1713 			break;
1714 		case 5:
1715 			*v = (long)read_gc0_kscratch4();
1716 			break;
1717 		case 6:
1718 			*v = (long)read_gc0_kscratch5();
1719 			break;
1720 		case 7:
1721 			*v = (long)read_gc0_kscratch6();
1722 			break;
1723 		}
1724 		break;
1725 	case KVM_REG_MIPS_COUNT_CTL:
1726 		*v = vcpu->arch.count_ctl;
1727 		break;
1728 	case KVM_REG_MIPS_COUNT_RESUME:
1729 		*v = ktime_to_ns(vcpu->arch.count_resume);
1730 		break;
1731 	case KVM_REG_MIPS_COUNT_HZ:
1732 		*v = vcpu->arch.count_hz;
1733 		break;
1734 	default:
1735 		return -EINVAL;
1736 	}
1737 	return 0;
1738 }
1739 
1740 static int kvm_vz_set_one_reg(struct kvm_vcpu *vcpu,
1741 			      const struct kvm_one_reg *reg,
1742 			      s64 v)
1743 {
1744 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1745 	unsigned int idx;
1746 	int ret = 0;
1747 	unsigned int cur, change;
1748 
1749 	switch (reg->id) {
1750 	case KVM_REG_MIPS_CP0_INDEX:
1751 		write_gc0_index(v);
1752 		break;
1753 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1754 		write_gc0_entrylo0(entrylo_user_to_kvm(v));
1755 		break;
1756 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1757 		write_gc0_entrylo1(entrylo_user_to_kvm(v));
1758 		break;
1759 	case KVM_REG_MIPS_CP0_CONTEXT:
1760 		write_gc0_context(v);
1761 		break;
1762 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1763 		if (!cpu_guest_has_contextconfig)
1764 			return -EINVAL;
1765 		write_gc0_contextconfig(v);
1766 		break;
1767 	case KVM_REG_MIPS_CP0_USERLOCAL:
1768 		if (!cpu_guest_has_userlocal)
1769 			return -EINVAL;
1770 		write_gc0_userlocal(v);
1771 		break;
1772 #ifdef CONFIG_64BIT
1773 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1774 		if (!cpu_guest_has_contextconfig)
1775 			return -EINVAL;
1776 		write_gc0_xcontextconfig(v);
1777 		break;
1778 #endif
1779 	case KVM_REG_MIPS_CP0_PAGEMASK:
1780 		write_gc0_pagemask(v);
1781 		break;
1782 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1783 		write_gc0_pagegrain(v);
1784 		break;
1785 	case KVM_REG_MIPS_CP0_SEGCTL0:
1786 		if (!cpu_guest_has_segments)
1787 			return -EINVAL;
1788 		write_gc0_segctl0(v);
1789 		break;
1790 	case KVM_REG_MIPS_CP0_SEGCTL1:
1791 		if (!cpu_guest_has_segments)
1792 			return -EINVAL;
1793 		write_gc0_segctl1(v);
1794 		break;
1795 	case KVM_REG_MIPS_CP0_SEGCTL2:
1796 		if (!cpu_guest_has_segments)
1797 			return -EINVAL;
1798 		write_gc0_segctl2(v);
1799 		break;
1800 	case KVM_REG_MIPS_CP0_PWBASE:
1801 		if (!cpu_guest_has_htw)
1802 			return -EINVAL;
1803 		write_gc0_pwbase(v);
1804 		break;
1805 	case KVM_REG_MIPS_CP0_PWFIELD:
1806 		if (!cpu_guest_has_htw)
1807 			return -EINVAL;
1808 		write_gc0_pwfield(v);
1809 		break;
1810 	case KVM_REG_MIPS_CP0_PWSIZE:
1811 		if (!cpu_guest_has_htw)
1812 			return -EINVAL;
1813 		write_gc0_pwsize(v);
1814 		break;
1815 	case KVM_REG_MIPS_CP0_WIRED:
1816 		change_gc0_wired(MIPSR6_WIRED_WIRED, v);
1817 		break;
1818 	case KVM_REG_MIPS_CP0_PWCTL:
1819 		if (!cpu_guest_has_htw)
1820 			return -EINVAL;
1821 		write_gc0_pwctl(v);
1822 		break;
1823 	case KVM_REG_MIPS_CP0_HWRENA:
1824 		write_gc0_hwrena(v);
1825 		break;
1826 	case KVM_REG_MIPS_CP0_BADVADDR:
1827 		write_gc0_badvaddr(v);
1828 		break;
1829 	case KVM_REG_MIPS_CP0_BADINSTR:
1830 		if (!cpu_guest_has_badinstr)
1831 			return -EINVAL;
1832 		write_gc0_badinstr(v);
1833 		break;
1834 	case KVM_REG_MIPS_CP0_BADINSTRP:
1835 		if (!cpu_guest_has_badinstrp)
1836 			return -EINVAL;
1837 		write_gc0_badinstrp(v);
1838 		break;
1839 	case KVM_REG_MIPS_CP0_COUNT:
1840 		kvm_mips_write_count(vcpu, v);
1841 		break;
1842 	case KVM_REG_MIPS_CP0_ENTRYHI:
1843 		write_gc0_entryhi(v);
1844 		break;
1845 	case KVM_REG_MIPS_CP0_COMPARE:
1846 		kvm_mips_write_compare(vcpu, v, false);
1847 		break;
1848 	case KVM_REG_MIPS_CP0_STATUS:
1849 		write_gc0_status(v);
1850 		break;
1851 	case KVM_REG_MIPS_CP0_INTCTL:
1852 		write_gc0_intctl(v);
1853 		break;
1854 	case KVM_REG_MIPS_CP0_CAUSE:
1855 		/*
1856 		 * If the timer is stopped or started (DC bit) it must look
1857 		 * atomic with changes to the timer interrupt pending bit (TI).
1858 		 * A timer interrupt should not happen in between.
1859 		 */
1860 		if ((read_gc0_cause() ^ v) & CAUSEF_DC) {
1861 			if (v & CAUSEF_DC) {
1862 				/* disable timer first */
1863 				kvm_mips_count_disable_cause(vcpu);
1864 				change_gc0_cause((u32)~CAUSEF_DC, v);
1865 			} else {
1866 				/* enable timer last */
1867 				change_gc0_cause((u32)~CAUSEF_DC, v);
1868 				kvm_mips_count_enable_cause(vcpu);
1869 			}
1870 		} else {
1871 			write_gc0_cause(v);
1872 		}
1873 		break;
1874 	case KVM_REG_MIPS_CP0_EPC:
1875 		write_gc0_epc(v);
1876 		break;
1877 	case KVM_REG_MIPS_CP0_PRID:
1878 		kvm_write_c0_guest_prid(cop0, v);
1879 		break;
1880 	case KVM_REG_MIPS_CP0_EBASE:
1881 		kvm_vz_write_gc0_ebase(v);
1882 		break;
1883 	case KVM_REG_MIPS_CP0_CONFIG:
1884 		cur = read_gc0_config();
1885 		change = (cur ^ v) & kvm_vz_config_user_wrmask(vcpu);
1886 		if (change) {
1887 			v = cur ^ change;
1888 			write_gc0_config(v);
1889 		}
1890 		break;
1891 	case KVM_REG_MIPS_CP0_CONFIG1:
1892 		if (!cpu_guest_has_conf1)
1893 			break;
1894 		cur = read_gc0_config1();
1895 		change = (cur ^ v) & kvm_vz_config1_user_wrmask(vcpu);
1896 		if (change) {
1897 			v = cur ^ change;
1898 			write_gc0_config1(v);
1899 		}
1900 		break;
1901 	case KVM_REG_MIPS_CP0_CONFIG2:
1902 		if (!cpu_guest_has_conf2)
1903 			break;
1904 		cur = read_gc0_config2();
1905 		change = (cur ^ v) & kvm_vz_config2_user_wrmask(vcpu);
1906 		if (change) {
1907 			v = cur ^ change;
1908 			write_gc0_config2(v);
1909 		}
1910 		break;
1911 	case KVM_REG_MIPS_CP0_CONFIG3:
1912 		if (!cpu_guest_has_conf3)
1913 			break;
1914 		cur = read_gc0_config3();
1915 		change = (cur ^ v) & kvm_vz_config3_user_wrmask(vcpu);
1916 		if (change) {
1917 			v = cur ^ change;
1918 			write_gc0_config3(v);
1919 		}
1920 		break;
1921 	case KVM_REG_MIPS_CP0_CONFIG4:
1922 		if (!cpu_guest_has_conf4)
1923 			break;
1924 		cur = read_gc0_config4();
1925 		change = (cur ^ v) & kvm_vz_config4_user_wrmask(vcpu);
1926 		if (change) {
1927 			v = cur ^ change;
1928 			write_gc0_config4(v);
1929 		}
1930 		break;
1931 	case KVM_REG_MIPS_CP0_CONFIG5:
1932 		if (!cpu_guest_has_conf5)
1933 			break;
1934 		cur = read_gc0_config5();
1935 		change = (cur ^ v) & kvm_vz_config5_user_wrmask(vcpu);
1936 		if (change) {
1937 			v = cur ^ change;
1938 			write_gc0_config5(v);
1939 		}
1940 		break;
1941 #ifdef CONFIG_64BIT
1942 	case KVM_REG_MIPS_CP0_XCONTEXT:
1943 		write_gc0_xcontext(v);
1944 		break;
1945 #endif
1946 	case KVM_REG_MIPS_CP0_ERROREPC:
1947 		write_gc0_errorepc(v);
1948 		break;
1949 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
1950 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
1951 		if (!cpu_guest_has_kscr(idx))
1952 			return -EINVAL;
1953 		switch (idx) {
1954 		case 2:
1955 			write_gc0_kscratch1(v);
1956 			break;
1957 		case 3:
1958 			write_gc0_kscratch2(v);
1959 			break;
1960 		case 4:
1961 			write_gc0_kscratch3(v);
1962 			break;
1963 		case 5:
1964 			write_gc0_kscratch4(v);
1965 			break;
1966 		case 6:
1967 			write_gc0_kscratch5(v);
1968 			break;
1969 		case 7:
1970 			write_gc0_kscratch6(v);
1971 			break;
1972 		}
1973 		break;
1974 	case KVM_REG_MIPS_COUNT_CTL:
1975 		ret = kvm_mips_set_count_ctl(vcpu, v);
1976 		break;
1977 	case KVM_REG_MIPS_COUNT_RESUME:
1978 		ret = kvm_mips_set_count_resume(vcpu, v);
1979 		break;
1980 	case KVM_REG_MIPS_COUNT_HZ:
1981 		ret = kvm_mips_set_count_hz(vcpu, v);
1982 		break;
1983 	default:
1984 		return -EINVAL;
1985 	}
1986 	return ret;
1987 }
1988 
1989 #define guestid_cache(cpu)	(cpu_data[cpu].guestid_cache)
1990 static void kvm_vz_get_new_guestid(unsigned long cpu, struct kvm_vcpu *vcpu)
1991 {
1992 	unsigned long guestid = guestid_cache(cpu);
1993 
1994 	if (!(++guestid & GUESTID_MASK)) {
1995 		if (cpu_has_vtag_icache)
1996 			flush_icache_all();
1997 
1998 		if (!guestid)		/* fix version if needed */
1999 			guestid = GUESTID_FIRST_VERSION;
2000 
2001 		++guestid;		/* guestid 0 reserved for root */
2002 
2003 		/* start new guestid cycle */
2004 		kvm_vz_local_flush_roottlb_all_guests();
2005 		kvm_vz_local_flush_guesttlb_all();
2006 	}
2007 
2008 	guestid_cache(cpu) = guestid;
2009 }
2010 
2011 /* Returns 1 if the guest TLB may be clobbered */
2012 static int kvm_vz_check_requests(struct kvm_vcpu *vcpu, int cpu)
2013 {
2014 	int ret = 0;
2015 	int i;
2016 
2017 	if (!vcpu->requests)
2018 		return 0;
2019 
2020 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2021 		if (cpu_has_guestid) {
2022 			/* Drop all GuestIDs for this VCPU */
2023 			for_each_possible_cpu(i)
2024 				vcpu->arch.vzguestid[i] = 0;
2025 			/* This will clobber guest TLB contents too */
2026 			ret = 1;
2027 		}
2028 		/*
2029 		 * For Root ASID Dealias (RAD) we don't do anything here, but we
2030 		 * still need the request to ensure we recheck asid_flush_mask.
2031 		 * We can still return 0 as only the root TLB will be affected
2032 		 * by a root ASID flush.
2033 		 */
2034 	}
2035 
2036 	return ret;
2037 }
2038 
2039 static void kvm_vz_vcpu_save_wired(struct kvm_vcpu *vcpu)
2040 {
2041 	unsigned int wired = read_gc0_wired();
2042 	struct kvm_mips_tlb *tlbs;
2043 	int i;
2044 
2045 	/* Expand the wired TLB array if necessary */
2046 	wired &= MIPSR6_WIRED_WIRED;
2047 	if (wired > vcpu->arch.wired_tlb_limit) {
2048 		tlbs = krealloc(vcpu->arch.wired_tlb, wired *
2049 				sizeof(*vcpu->arch.wired_tlb), GFP_ATOMIC);
2050 		if (WARN_ON(!tlbs)) {
2051 			/* Save whatever we can */
2052 			wired = vcpu->arch.wired_tlb_limit;
2053 		} else {
2054 			vcpu->arch.wired_tlb = tlbs;
2055 			vcpu->arch.wired_tlb_limit = wired;
2056 		}
2057 	}
2058 
2059 	if (wired)
2060 		/* Save wired entries from the guest TLB */
2061 		kvm_vz_save_guesttlb(vcpu->arch.wired_tlb, 0, wired);
2062 	/* Invalidate any dropped entries since last time */
2063 	for (i = wired; i < vcpu->arch.wired_tlb_used; ++i) {
2064 		vcpu->arch.wired_tlb[i].tlb_hi = UNIQUE_GUEST_ENTRYHI(i);
2065 		vcpu->arch.wired_tlb[i].tlb_lo[0] = 0;
2066 		vcpu->arch.wired_tlb[i].tlb_lo[1] = 0;
2067 		vcpu->arch.wired_tlb[i].tlb_mask = 0;
2068 	}
2069 	vcpu->arch.wired_tlb_used = wired;
2070 }
2071 
2072 static void kvm_vz_vcpu_load_wired(struct kvm_vcpu *vcpu)
2073 {
2074 	/* Load wired entries into the guest TLB */
2075 	if (vcpu->arch.wired_tlb)
2076 		kvm_vz_load_guesttlb(vcpu->arch.wired_tlb, 0,
2077 				     vcpu->arch.wired_tlb_used);
2078 }
2079 
2080 static void kvm_vz_vcpu_load_tlb(struct kvm_vcpu *vcpu, int cpu)
2081 {
2082 	struct kvm *kvm = vcpu->kvm;
2083 	struct mm_struct *gpa_mm = &kvm->arch.gpa_mm;
2084 	bool migrated;
2085 
2086 	/*
2087 	 * Are we entering guest context on a different CPU to last time?
2088 	 * If so, the VCPU's guest TLB state on this CPU may be stale.
2089 	 */
2090 	migrated = (vcpu->arch.last_exec_cpu != cpu);
2091 	vcpu->arch.last_exec_cpu = cpu;
2092 
2093 	/*
2094 	 * A vcpu's GuestID is set in GuestCtl1.ID when the vcpu is loaded and
2095 	 * remains set until another vcpu is loaded in.  As a rule GuestRID
2096 	 * remains zeroed when in root context unless the kernel is busy
2097 	 * manipulating guest tlb entries.
2098 	 */
2099 	if (cpu_has_guestid) {
2100 		/*
2101 		 * Check if our GuestID is of an older version and thus invalid.
2102 		 *
2103 		 * We also discard the stored GuestID if we've executed on
2104 		 * another CPU, as the guest mappings may have changed without
2105 		 * hypervisor knowledge.
2106 		 */
2107 		if (migrated ||
2108 		    (vcpu->arch.vzguestid[cpu] ^ guestid_cache(cpu)) &
2109 					GUESTID_VERSION_MASK) {
2110 			kvm_vz_get_new_guestid(cpu, vcpu);
2111 			vcpu->arch.vzguestid[cpu] = guestid_cache(cpu);
2112 			trace_kvm_guestid_change(vcpu,
2113 						 vcpu->arch.vzguestid[cpu]);
2114 		}
2115 
2116 		/* Restore GuestID */
2117 		change_c0_guestctl1(GUESTID_MASK, vcpu->arch.vzguestid[cpu]);
2118 	} else {
2119 		/*
2120 		 * The Guest TLB only stores a single guest's TLB state, so
2121 		 * flush it if another VCPU has executed on this CPU.
2122 		 *
2123 		 * We also flush if we've executed on another CPU, as the guest
2124 		 * mappings may have changed without hypervisor knowledge.
2125 		 */
2126 		if (migrated || last_exec_vcpu[cpu] != vcpu)
2127 			kvm_vz_local_flush_guesttlb_all();
2128 		last_exec_vcpu[cpu] = vcpu;
2129 
2130 		/*
2131 		 * Root ASID dealiases guest GPA mappings in the root TLB.
2132 		 * Allocate new root ASID if needed.
2133 		 */
2134 		if (cpumask_test_and_clear_cpu(cpu, &kvm->arch.asid_flush_mask)
2135 		    || (cpu_context(cpu, gpa_mm) ^ asid_cache(cpu)) &
2136 						asid_version_mask(cpu))
2137 			get_new_mmu_context(gpa_mm, cpu);
2138 	}
2139 }
2140 
2141 static int kvm_vz_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2142 {
2143 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2144 	bool migrated, all;
2145 
2146 	/*
2147 	 * Have we migrated to a different CPU?
2148 	 * If so, any old guest TLB state may be stale.
2149 	 */
2150 	migrated = (vcpu->arch.last_sched_cpu != cpu);
2151 
2152 	/*
2153 	 * Was this the last VCPU to run on this CPU?
2154 	 * If not, any old guest state from this VCPU will have been clobbered.
2155 	 */
2156 	all = migrated || (last_vcpu[cpu] != vcpu);
2157 	last_vcpu[cpu] = vcpu;
2158 
2159 	/*
2160 	 * Restore CP0_Wired unconditionally as we clear it after use, and
2161 	 * restore wired guest TLB entries (while in guest context).
2162 	 */
2163 	kvm_restore_gc0_wired(cop0);
2164 	if (current->flags & PF_VCPU) {
2165 		tlbw_use_hazard();
2166 		kvm_vz_vcpu_load_tlb(vcpu, cpu);
2167 		kvm_vz_vcpu_load_wired(vcpu);
2168 	}
2169 
2170 	/*
2171 	 * Restore timer state regardless, as e.g. Cause.TI can change over time
2172 	 * if left unmaintained.
2173 	 */
2174 	kvm_vz_restore_timer(vcpu);
2175 
2176 	/* Don't bother restoring registers multiple times unless necessary */
2177 	if (!all)
2178 		return 0;
2179 
2180 	/*
2181 	 * Restore config registers first, as some implementations restrict
2182 	 * writes to other registers when the corresponding feature bits aren't
2183 	 * set. For example Status.CU1 cannot be set unless Config1.FP is set.
2184 	 */
2185 	kvm_restore_gc0_config(cop0);
2186 	if (cpu_guest_has_conf1)
2187 		kvm_restore_gc0_config1(cop0);
2188 	if (cpu_guest_has_conf2)
2189 		kvm_restore_gc0_config2(cop0);
2190 	if (cpu_guest_has_conf3)
2191 		kvm_restore_gc0_config3(cop0);
2192 	if (cpu_guest_has_conf4)
2193 		kvm_restore_gc0_config4(cop0);
2194 	if (cpu_guest_has_conf5)
2195 		kvm_restore_gc0_config5(cop0);
2196 	if (cpu_guest_has_conf6)
2197 		kvm_restore_gc0_config6(cop0);
2198 	if (cpu_guest_has_conf7)
2199 		kvm_restore_gc0_config7(cop0);
2200 
2201 	kvm_restore_gc0_index(cop0);
2202 	kvm_restore_gc0_entrylo0(cop0);
2203 	kvm_restore_gc0_entrylo1(cop0);
2204 	kvm_restore_gc0_context(cop0);
2205 	if (cpu_guest_has_contextconfig)
2206 		kvm_restore_gc0_contextconfig(cop0);
2207 #ifdef CONFIG_64BIT
2208 	kvm_restore_gc0_xcontext(cop0);
2209 	if (cpu_guest_has_contextconfig)
2210 		kvm_restore_gc0_xcontextconfig(cop0);
2211 #endif
2212 	kvm_restore_gc0_pagemask(cop0);
2213 	kvm_restore_gc0_pagegrain(cop0);
2214 	kvm_restore_gc0_hwrena(cop0);
2215 	kvm_restore_gc0_badvaddr(cop0);
2216 	kvm_restore_gc0_entryhi(cop0);
2217 	kvm_restore_gc0_status(cop0);
2218 	kvm_restore_gc0_intctl(cop0);
2219 	kvm_restore_gc0_epc(cop0);
2220 	kvm_vz_write_gc0_ebase(kvm_read_sw_gc0_ebase(cop0));
2221 	if (cpu_guest_has_userlocal)
2222 		kvm_restore_gc0_userlocal(cop0);
2223 
2224 	kvm_restore_gc0_errorepc(cop0);
2225 
2226 	/* restore KScratch registers if enabled in guest */
2227 	if (cpu_guest_has_conf4) {
2228 		if (cpu_guest_has_kscr(2))
2229 			kvm_restore_gc0_kscratch1(cop0);
2230 		if (cpu_guest_has_kscr(3))
2231 			kvm_restore_gc0_kscratch2(cop0);
2232 		if (cpu_guest_has_kscr(4))
2233 			kvm_restore_gc0_kscratch3(cop0);
2234 		if (cpu_guest_has_kscr(5))
2235 			kvm_restore_gc0_kscratch4(cop0);
2236 		if (cpu_guest_has_kscr(6))
2237 			kvm_restore_gc0_kscratch5(cop0);
2238 		if (cpu_guest_has_kscr(7))
2239 			kvm_restore_gc0_kscratch6(cop0);
2240 	}
2241 
2242 	if (cpu_guest_has_badinstr)
2243 		kvm_restore_gc0_badinstr(cop0);
2244 	if (cpu_guest_has_badinstrp)
2245 		kvm_restore_gc0_badinstrp(cop0);
2246 
2247 	if (cpu_guest_has_segments) {
2248 		kvm_restore_gc0_segctl0(cop0);
2249 		kvm_restore_gc0_segctl1(cop0);
2250 		kvm_restore_gc0_segctl2(cop0);
2251 	}
2252 
2253 	/* restore HTW registers */
2254 	if (cpu_guest_has_htw) {
2255 		kvm_restore_gc0_pwbase(cop0);
2256 		kvm_restore_gc0_pwfield(cop0);
2257 		kvm_restore_gc0_pwsize(cop0);
2258 		kvm_restore_gc0_pwctl(cop0);
2259 	}
2260 
2261 	/* restore Root.GuestCtl2 from unused Guest guestctl2 register */
2262 	if (cpu_has_guestctl2)
2263 		write_c0_guestctl2(
2264 			cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL]);
2265 
2266 	/*
2267 	 * We should clear linked load bit to break interrupted atomics. This
2268 	 * prevents a SC on the next VCPU from succeeding by matching a LL on
2269 	 * the previous VCPU.
2270 	 */
2271 	if (cpu_guest_has_rw_llb)
2272 		write_gc0_lladdr(0);
2273 
2274 	return 0;
2275 }
2276 
2277 static int kvm_vz_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
2278 {
2279 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2280 
2281 	if (current->flags & PF_VCPU)
2282 		kvm_vz_vcpu_save_wired(vcpu);
2283 
2284 	kvm_lose_fpu(vcpu);
2285 
2286 	kvm_save_gc0_index(cop0);
2287 	kvm_save_gc0_entrylo0(cop0);
2288 	kvm_save_gc0_entrylo1(cop0);
2289 	kvm_save_gc0_context(cop0);
2290 	if (cpu_guest_has_contextconfig)
2291 		kvm_save_gc0_contextconfig(cop0);
2292 #ifdef CONFIG_64BIT
2293 	kvm_save_gc0_xcontext(cop0);
2294 	if (cpu_guest_has_contextconfig)
2295 		kvm_save_gc0_xcontextconfig(cop0);
2296 #endif
2297 	kvm_save_gc0_pagemask(cop0);
2298 	kvm_save_gc0_pagegrain(cop0);
2299 	kvm_save_gc0_wired(cop0);
2300 	/* allow wired TLB entries to be overwritten */
2301 	clear_gc0_wired(MIPSR6_WIRED_WIRED);
2302 	kvm_save_gc0_hwrena(cop0);
2303 	kvm_save_gc0_badvaddr(cop0);
2304 	kvm_save_gc0_entryhi(cop0);
2305 	kvm_save_gc0_status(cop0);
2306 	kvm_save_gc0_intctl(cop0);
2307 	kvm_save_gc0_epc(cop0);
2308 	kvm_write_sw_gc0_ebase(cop0, kvm_vz_read_gc0_ebase());
2309 	if (cpu_guest_has_userlocal)
2310 		kvm_save_gc0_userlocal(cop0);
2311 
2312 	/* only save implemented config registers */
2313 	kvm_save_gc0_config(cop0);
2314 	if (cpu_guest_has_conf1)
2315 		kvm_save_gc0_config1(cop0);
2316 	if (cpu_guest_has_conf2)
2317 		kvm_save_gc0_config2(cop0);
2318 	if (cpu_guest_has_conf3)
2319 		kvm_save_gc0_config3(cop0);
2320 	if (cpu_guest_has_conf4)
2321 		kvm_save_gc0_config4(cop0);
2322 	if (cpu_guest_has_conf5)
2323 		kvm_save_gc0_config5(cop0);
2324 	if (cpu_guest_has_conf6)
2325 		kvm_save_gc0_config6(cop0);
2326 	if (cpu_guest_has_conf7)
2327 		kvm_save_gc0_config7(cop0);
2328 
2329 	kvm_save_gc0_errorepc(cop0);
2330 
2331 	/* save KScratch registers if enabled in guest */
2332 	if (cpu_guest_has_conf4) {
2333 		if (cpu_guest_has_kscr(2))
2334 			kvm_save_gc0_kscratch1(cop0);
2335 		if (cpu_guest_has_kscr(3))
2336 			kvm_save_gc0_kscratch2(cop0);
2337 		if (cpu_guest_has_kscr(4))
2338 			kvm_save_gc0_kscratch3(cop0);
2339 		if (cpu_guest_has_kscr(5))
2340 			kvm_save_gc0_kscratch4(cop0);
2341 		if (cpu_guest_has_kscr(6))
2342 			kvm_save_gc0_kscratch5(cop0);
2343 		if (cpu_guest_has_kscr(7))
2344 			kvm_save_gc0_kscratch6(cop0);
2345 	}
2346 
2347 	if (cpu_guest_has_badinstr)
2348 		kvm_save_gc0_badinstr(cop0);
2349 	if (cpu_guest_has_badinstrp)
2350 		kvm_save_gc0_badinstrp(cop0);
2351 
2352 	if (cpu_guest_has_segments) {
2353 		kvm_save_gc0_segctl0(cop0);
2354 		kvm_save_gc0_segctl1(cop0);
2355 		kvm_save_gc0_segctl2(cop0);
2356 	}
2357 
2358 	/* save HTW registers if enabled in guest */
2359 	if (cpu_guest_has_htw &&
2360 	    kvm_read_sw_gc0_config3(cop0) & MIPS_CONF3_PW) {
2361 		kvm_save_gc0_pwbase(cop0);
2362 		kvm_save_gc0_pwfield(cop0);
2363 		kvm_save_gc0_pwsize(cop0);
2364 		kvm_save_gc0_pwctl(cop0);
2365 	}
2366 
2367 	kvm_vz_save_timer(vcpu);
2368 
2369 	/* save Root.GuestCtl2 in unused Guest guestctl2 register */
2370 	if (cpu_has_guestctl2)
2371 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] =
2372 			read_c0_guestctl2();
2373 
2374 	return 0;
2375 }
2376 
2377 /**
2378  * kvm_vz_resize_guest_vtlb() - Attempt to resize guest VTLB.
2379  * @size:	Number of guest VTLB entries (0 < @size <= root VTLB entries).
2380  *
2381  * Attempt to resize the guest VTLB by writing guest Config registers. This is
2382  * necessary for cores with a shared root/guest TLB to avoid overlap with wired
2383  * entries in the root VTLB.
2384  *
2385  * Returns:	The resulting guest VTLB size.
2386  */
2387 static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
2388 {
2389 	unsigned int config4 = 0, ret = 0, limit;
2390 
2391 	/* Write MMUSize - 1 into guest Config registers */
2392 	if (cpu_guest_has_conf1)
2393 		change_gc0_config1(MIPS_CONF1_TLBS,
2394 				   (size - 1) << MIPS_CONF1_TLBS_SHIFT);
2395 	if (cpu_guest_has_conf4) {
2396 		config4 = read_gc0_config4();
2397 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2398 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT) {
2399 			config4 &= ~MIPS_CONF4_VTLBSIZEEXT;
2400 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2401 				MIPS_CONF4_VTLBSIZEEXT_SHIFT;
2402 		} else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2403 			   MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT) {
2404 			config4 &= ~MIPS_CONF4_MMUSIZEEXT;
2405 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2406 				MIPS_CONF4_MMUSIZEEXT_SHIFT;
2407 		}
2408 		write_gc0_config4(config4);
2409 	}
2410 
2411 	/*
2412 	 * Set Guest.Wired.Limit = 0 (no limit up to Guest.MMUSize-1), unless it
2413 	 * would exceed Root.Wired.Limit (clearing Guest.Wired.Wired so write
2414 	 * not dropped)
2415 	 */
2416 	if (cpu_has_mips_r6) {
2417 		limit = (read_c0_wired() & MIPSR6_WIRED_LIMIT) >>
2418 						MIPSR6_WIRED_LIMIT_SHIFT;
2419 		if (size - 1 <= limit)
2420 			limit = 0;
2421 		write_gc0_wired(limit << MIPSR6_WIRED_LIMIT_SHIFT);
2422 	}
2423 
2424 	/* Read back MMUSize - 1 */
2425 	back_to_back_c0_hazard();
2426 	if (cpu_guest_has_conf1)
2427 		ret = (read_gc0_config1() & MIPS_CONF1_TLBS) >>
2428 						MIPS_CONF1_TLBS_SHIFT;
2429 	if (config4) {
2430 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2431 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT)
2432 			ret |= ((config4 & MIPS_CONF4_VTLBSIZEEXT) >>
2433 				MIPS_CONF4_VTLBSIZEEXT_SHIFT) <<
2434 				MIPS_CONF1_TLBS_SIZE;
2435 		else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2436 			 MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT)
2437 			ret |= ((config4 & MIPS_CONF4_MMUSIZEEXT) >>
2438 				MIPS_CONF4_MMUSIZEEXT_SHIFT) <<
2439 				MIPS_CONF1_TLBS_SIZE;
2440 	}
2441 	return ret + 1;
2442 }
2443 
2444 static int kvm_vz_hardware_enable(void)
2445 {
2446 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
2447 
2448 	/*
2449 	 * ImgTec cores tend to use a shared root/guest TLB. To avoid overlap of
2450 	 * root wired and guest entries, the guest TLB may need resizing.
2451 	 */
2452 	mmu_size = current_cpu_data.tlbsizevtlb;
2453 	ftlb_size = current_cpu_data.tlbsize - mmu_size;
2454 
2455 	/* Try switching to maximum guest VTLB size for flush */
2456 	guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
2457 	current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2458 	kvm_vz_local_flush_guesttlb_all();
2459 
2460 	/*
2461 	 * Reduce to make space for root wired entries and at least 2 root
2462 	 * non-wired entries. This does assume that long-term wired entries
2463 	 * won't be added later.
2464 	 */
2465 	guest_mmu_size = mmu_size - num_wired_entries() - 2;
2466 	guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
2467 	current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2468 
2469 	/*
2470 	 * Write the VTLB size, but if another CPU has already written, check it
2471 	 * matches or we won't provide a consistent view to the guest. If this
2472 	 * ever happens it suggests an asymmetric number of wired entries.
2473 	 */
2474 	if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
2475 	    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
2476 		 "Available guest VTLB size mismatch"))
2477 		return -EINVAL;
2478 
2479 	/*
2480 	 * Enable virtualization features granting guest direct control of
2481 	 * certain features:
2482 	 * CP0=1:	Guest coprocessor 0 context.
2483 	 * AT=Guest:	Guest MMU.
2484 	 * CG=1:	Hit (virtual address) CACHE operations (optional).
2485 	 * CF=1:	Guest Config registers.
2486 	 * CGI=1:	Indexed flush CACHE operations (optional).
2487 	 */
2488 	write_c0_guestctl0(MIPS_GCTL0_CP0 |
2489 			   (MIPS_GCTL0_AT_GUEST << MIPS_GCTL0_AT_SHIFT) |
2490 			   MIPS_GCTL0_CG | MIPS_GCTL0_CF);
2491 	if (cpu_has_guestctl0ext)
2492 		set_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2493 
2494 	if (cpu_has_guestid) {
2495 		write_c0_guestctl1(0);
2496 		kvm_vz_local_flush_roottlb_all_guests();
2497 
2498 		GUESTID_MASK = current_cpu_data.guestid_mask;
2499 		GUESTID_FIRST_VERSION = GUESTID_MASK + 1;
2500 		GUESTID_VERSION_MASK = ~GUESTID_MASK;
2501 
2502 		current_cpu_data.guestid_cache = GUESTID_FIRST_VERSION;
2503 	}
2504 
2505 	/* clear any pending injected virtual guest interrupts */
2506 	if (cpu_has_guestctl2)
2507 		clear_c0_guestctl2(0x3f << 10);
2508 
2509 	return 0;
2510 }
2511 
2512 static void kvm_vz_hardware_disable(void)
2513 {
2514 	kvm_vz_local_flush_guesttlb_all();
2515 
2516 	if (cpu_has_guestid) {
2517 		write_c0_guestctl1(0);
2518 		kvm_vz_local_flush_roottlb_all_guests();
2519 	}
2520 }
2521 
2522 static int kvm_vz_check_extension(struct kvm *kvm, long ext)
2523 {
2524 	int r;
2525 
2526 	switch (ext) {
2527 	case KVM_CAP_MIPS_VZ:
2528 		/* we wouldn't be here unless cpu_has_vz */
2529 		r = 1;
2530 		break;
2531 #ifdef CONFIG_64BIT
2532 	case KVM_CAP_MIPS_64BIT:
2533 		/* We support 64-bit registers/operations and addresses */
2534 		r = 2;
2535 		break;
2536 #endif
2537 	default:
2538 		r = 0;
2539 		break;
2540 	}
2541 
2542 	return r;
2543 }
2544 
2545 static int kvm_vz_vcpu_init(struct kvm_vcpu *vcpu)
2546 {
2547 	int i;
2548 
2549 	for_each_possible_cpu(i)
2550 		vcpu->arch.vzguestid[i] = 0;
2551 
2552 	return 0;
2553 }
2554 
2555 static void kvm_vz_vcpu_uninit(struct kvm_vcpu *vcpu)
2556 {
2557 	int cpu;
2558 
2559 	/*
2560 	 * If the VCPU is freed and reused as another VCPU, we don't want the
2561 	 * matching pointer wrongly hanging around in last_vcpu[] or
2562 	 * last_exec_vcpu[].
2563 	 */
2564 	for_each_possible_cpu(cpu) {
2565 		if (last_vcpu[cpu] == vcpu)
2566 			last_vcpu[cpu] = NULL;
2567 		if (last_exec_vcpu[cpu] == vcpu)
2568 			last_exec_vcpu[cpu] = NULL;
2569 	}
2570 }
2571 
2572 static int kvm_vz_vcpu_setup(struct kvm_vcpu *vcpu)
2573 {
2574 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2575 	unsigned long count_hz = 100*1000*1000; /* default to 100 MHz */
2576 
2577 	/*
2578 	 * Start off the timer at the same frequency as the host timer, but the
2579 	 * soft timer doesn't handle frequencies greater than 1GHz yet.
2580 	 */
2581 	if (mips_hpt_frequency && mips_hpt_frequency <= NSEC_PER_SEC)
2582 		count_hz = mips_hpt_frequency;
2583 	kvm_mips_init_count(vcpu, count_hz);
2584 
2585 	/*
2586 	 * Initialize guest register state to valid architectural reset state.
2587 	 */
2588 
2589 	/* PageGrain */
2590 	if (cpu_has_mips_r6)
2591 		kvm_write_sw_gc0_pagegrain(cop0, PG_RIE | PG_XIE | PG_IEC);
2592 	/* Wired */
2593 	if (cpu_has_mips_r6)
2594 		kvm_write_sw_gc0_wired(cop0,
2595 				       read_gc0_wired() & MIPSR6_WIRED_LIMIT);
2596 	/* Status */
2597 	kvm_write_sw_gc0_status(cop0, ST0_BEV | ST0_ERL);
2598 	if (cpu_has_mips_r6)
2599 		kvm_change_sw_gc0_status(cop0, ST0_FR, read_gc0_status());
2600 	/* IntCtl */
2601 	kvm_write_sw_gc0_intctl(cop0, read_gc0_intctl() &
2602 				(INTCTLF_IPFDC | INTCTLF_IPPCI | INTCTLF_IPTI));
2603 	/* PRId */
2604 	kvm_write_sw_gc0_prid(cop0, boot_cpu_data.processor_id);
2605 	/* EBase */
2606 	kvm_write_sw_gc0_ebase(cop0, (s32)0x80000000 | vcpu->vcpu_id);
2607 	/* Config */
2608 	kvm_save_gc0_config(cop0);
2609 	/* architecturally writable (e.g. from guest) */
2610 	kvm_change_sw_gc0_config(cop0, CONF_CM_CMASK,
2611 				 _page_cachable_default >> _CACHE_SHIFT);
2612 	/* architecturally read only, but maybe writable from root */
2613 	kvm_change_sw_gc0_config(cop0, MIPS_CONF_MT, read_c0_config());
2614 	if (cpu_guest_has_conf1) {
2615 		kvm_set_sw_gc0_config(cop0, MIPS_CONF_M);
2616 		/* Config1 */
2617 		kvm_save_gc0_config1(cop0);
2618 		/* architecturally read only, but maybe writable from root */
2619 		kvm_clear_sw_gc0_config1(cop0, MIPS_CONF1_C2	|
2620 					       MIPS_CONF1_MD	|
2621 					       MIPS_CONF1_PC	|
2622 					       MIPS_CONF1_WR	|
2623 					       MIPS_CONF1_CA	|
2624 					       MIPS_CONF1_FP);
2625 	}
2626 	if (cpu_guest_has_conf2) {
2627 		kvm_set_sw_gc0_config1(cop0, MIPS_CONF_M);
2628 		/* Config2 */
2629 		kvm_save_gc0_config2(cop0);
2630 	}
2631 	if (cpu_guest_has_conf3) {
2632 		kvm_set_sw_gc0_config2(cop0, MIPS_CONF_M);
2633 		/* Config3 */
2634 		kvm_save_gc0_config3(cop0);
2635 		/* architecturally writable (e.g. from guest) */
2636 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_ISA_OE);
2637 		/* architecturally read only, but maybe writable from root */
2638 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_MSA	|
2639 					       MIPS_CONF3_BPG	|
2640 					       MIPS_CONF3_ULRI	|
2641 					       MIPS_CONF3_DSP	|
2642 					       MIPS_CONF3_CTXTC	|
2643 					       MIPS_CONF3_ITL	|
2644 					       MIPS_CONF3_LPA	|
2645 					       MIPS_CONF3_VEIC	|
2646 					       MIPS_CONF3_VINT	|
2647 					       MIPS_CONF3_SP	|
2648 					       MIPS_CONF3_CDMM	|
2649 					       MIPS_CONF3_MT	|
2650 					       MIPS_CONF3_SM	|
2651 					       MIPS_CONF3_TL);
2652 	}
2653 	if (cpu_guest_has_conf4) {
2654 		kvm_set_sw_gc0_config3(cop0, MIPS_CONF_M);
2655 		/* Config4 */
2656 		kvm_save_gc0_config4(cop0);
2657 	}
2658 	if (cpu_guest_has_conf5) {
2659 		kvm_set_sw_gc0_config4(cop0, MIPS_CONF_M);
2660 		/* Config5 */
2661 		kvm_save_gc0_config5(cop0);
2662 		/* architecturally writable (e.g. from guest) */
2663 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_K	|
2664 					       MIPS_CONF5_CV	|
2665 					       MIPS_CONF5_MSAEN	|
2666 					       MIPS_CONF5_UFE	|
2667 					       MIPS_CONF5_FRE	|
2668 					       MIPS_CONF5_SBRI	|
2669 					       MIPS_CONF5_UFR);
2670 		/* architecturally read only, but maybe writable from root */
2671 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_MRP);
2672 	}
2673 
2674 	if (cpu_guest_has_contextconfig) {
2675 		/* ContextConfig */
2676 		kvm_write_sw_gc0_contextconfig(cop0, 0x007ffff0);
2677 #ifdef CONFIG_64BIT
2678 		/* XContextConfig */
2679 		/* bits SEGBITS-13+3:4 set */
2680 		kvm_write_sw_gc0_xcontextconfig(cop0,
2681 					((1ull << (cpu_vmbits - 13)) - 1) << 4);
2682 #endif
2683 	}
2684 
2685 	/* Implementation dependent, use the legacy layout */
2686 	if (cpu_guest_has_segments) {
2687 		/* SegCtl0, SegCtl1, SegCtl2 */
2688 		kvm_write_sw_gc0_segctl0(cop0, 0x00200010);
2689 		kvm_write_sw_gc0_segctl1(cop0, 0x00000002 |
2690 				(_page_cachable_default >> _CACHE_SHIFT) <<
2691 						(16 + MIPS_SEGCFG_C_SHIFT));
2692 		kvm_write_sw_gc0_segctl2(cop0, 0x00380438);
2693 	}
2694 
2695 	/* reset HTW registers */
2696 	if (cpu_guest_has_htw && cpu_has_mips_r6) {
2697 		/* PWField */
2698 		kvm_write_sw_gc0_pwfield(cop0, 0x0c30c302);
2699 		/* PWSize */
2700 		kvm_write_sw_gc0_pwsize(cop0, 1 << MIPS_PWSIZE_PTW_SHIFT);
2701 	}
2702 
2703 	/* start with no pending virtual guest interrupts */
2704 	if (cpu_has_guestctl2)
2705 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] = 0;
2706 
2707 	/* Put PC at reset vector */
2708 	vcpu->arch.pc = CKSEG1ADDR(0x1fc00000);
2709 
2710 	return 0;
2711 }
2712 
2713 static void kvm_vz_flush_shadow_all(struct kvm *kvm)
2714 {
2715 	if (cpu_has_guestid) {
2716 		/* Flush GuestID for each VCPU individually */
2717 		kvm_flush_remote_tlbs(kvm);
2718 	} else {
2719 		/*
2720 		 * For each CPU there is a single GPA ASID used by all VCPUs in
2721 		 * the VM, so it doesn't make sense for the VCPUs to handle
2722 		 * invalidation of these ASIDs individually.
2723 		 *
2724 		 * Instead mark all CPUs as needing ASID invalidation in
2725 		 * asid_flush_mask, and just use kvm_flush_remote_tlbs(kvm) to
2726 		 * kick any running VCPUs so they check asid_flush_mask.
2727 		 */
2728 		cpumask_setall(&kvm->arch.asid_flush_mask);
2729 		kvm_flush_remote_tlbs(kvm);
2730 	}
2731 }
2732 
2733 static void kvm_vz_flush_shadow_memslot(struct kvm *kvm,
2734 					const struct kvm_memory_slot *slot)
2735 {
2736 	kvm_vz_flush_shadow_all(kvm);
2737 }
2738 
2739 static void kvm_vz_vcpu_reenter(struct kvm_run *run, struct kvm_vcpu *vcpu)
2740 {
2741 	int cpu = smp_processor_id();
2742 	int preserve_guest_tlb;
2743 
2744 	preserve_guest_tlb = kvm_vz_check_requests(vcpu, cpu);
2745 
2746 	if (preserve_guest_tlb)
2747 		kvm_vz_vcpu_save_wired(vcpu);
2748 
2749 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
2750 
2751 	if (preserve_guest_tlb)
2752 		kvm_vz_vcpu_load_wired(vcpu);
2753 }
2754 
2755 static int kvm_vz_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
2756 {
2757 	int cpu = smp_processor_id();
2758 	int r;
2759 
2760 	/* Check if we have any exceptions/interrupts pending */
2761 	kvm_mips_deliver_interrupts(vcpu, read_gc0_cause());
2762 
2763 	kvm_vz_check_requests(vcpu, cpu);
2764 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
2765 	kvm_vz_vcpu_load_wired(vcpu);
2766 
2767 	r = vcpu->arch.vcpu_run(run, vcpu);
2768 
2769 	kvm_vz_vcpu_save_wired(vcpu);
2770 
2771 	return r;
2772 }
2773 
2774 static struct kvm_mips_callbacks kvm_vz_callbacks = {
2775 	.handle_cop_unusable = kvm_trap_vz_handle_cop_unusable,
2776 	.handle_tlb_mod = kvm_trap_vz_handle_tlb_st_miss,
2777 	.handle_tlb_ld_miss = kvm_trap_vz_handle_tlb_ld_miss,
2778 	.handle_tlb_st_miss = kvm_trap_vz_handle_tlb_st_miss,
2779 	.handle_addr_err_st = kvm_trap_vz_no_handler,
2780 	.handle_addr_err_ld = kvm_trap_vz_no_handler,
2781 	.handle_syscall = kvm_trap_vz_no_handler,
2782 	.handle_res_inst = kvm_trap_vz_no_handler,
2783 	.handle_break = kvm_trap_vz_no_handler,
2784 	.handle_msa_disabled = kvm_trap_vz_handle_msa_disabled,
2785 	.handle_guest_exit = kvm_trap_vz_handle_guest_exit,
2786 
2787 	.hardware_enable = kvm_vz_hardware_enable,
2788 	.hardware_disable = kvm_vz_hardware_disable,
2789 	.check_extension = kvm_vz_check_extension,
2790 	.vcpu_init = kvm_vz_vcpu_init,
2791 	.vcpu_uninit = kvm_vz_vcpu_uninit,
2792 	.vcpu_setup = kvm_vz_vcpu_setup,
2793 	.flush_shadow_all = kvm_vz_flush_shadow_all,
2794 	.flush_shadow_memslot = kvm_vz_flush_shadow_memslot,
2795 	.gva_to_gpa = kvm_vz_gva_to_gpa_cb,
2796 	.queue_timer_int = kvm_vz_queue_timer_int_cb,
2797 	.dequeue_timer_int = kvm_vz_dequeue_timer_int_cb,
2798 	.queue_io_int = kvm_vz_queue_io_int_cb,
2799 	.dequeue_io_int = kvm_vz_dequeue_io_int_cb,
2800 	.irq_deliver = kvm_vz_irq_deliver_cb,
2801 	.irq_clear = kvm_vz_irq_clear_cb,
2802 	.num_regs = kvm_vz_num_regs,
2803 	.copy_reg_indices = kvm_vz_copy_reg_indices,
2804 	.get_one_reg = kvm_vz_get_one_reg,
2805 	.set_one_reg = kvm_vz_set_one_reg,
2806 	.vcpu_load = kvm_vz_vcpu_load,
2807 	.vcpu_put = kvm_vz_vcpu_put,
2808 	.vcpu_run = kvm_vz_vcpu_run,
2809 	.vcpu_reenter = kvm_vz_vcpu_reenter,
2810 };
2811 
2812 int kvm_mips_emulation_init(struct kvm_mips_callbacks **install_callbacks)
2813 {
2814 	if (!cpu_has_vz)
2815 		return -ENODEV;
2816 
2817 	/*
2818 	 * VZ requires at least 2 KScratch registers, so it should have been
2819 	 * possible to allocate pgd_reg.
2820 	 */
2821 	if (WARN(pgd_reg == -1,
2822 		 "pgd_reg not allocated even though cpu_has_vz\n"))
2823 		return -ENODEV;
2824 
2825 	pr_info("Starting KVM with MIPS VZ extensions\n");
2826 
2827 	*install_callbacks = &kvm_vz_callbacks;
2828 	return 0;
2829 }
2830