xref: /openbmc/linux/arch/mips/kvm/vz.c (revision 4b7de028)
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_PRID &&
734 				    (sel == 0 ||	/* PRid */
735 				     sel == 2 ||	/* CDMMBase */
736 				     sel == 3)) ||	/* CMGCRBase */
737 				   (rd == MIPS_CP0_STATUS &&
738 				    (sel == 2 ||	/* SRSCtl */
739 				     sel == 3)) ||	/* SRSMap */
740 				   (rd == MIPS_CP0_CONFIG &&
741 				    (sel == 7)) ||	/* Config7 */
742 				   (rd == MIPS_CP0_ERRCTL &&
743 				    (sel == 0))) {	/* ErrCtl */
744 				val = cop0->reg[rd][sel];
745 			} else {
746 				val = 0;
747 				er = EMULATE_FAIL;
748 			}
749 
750 			if (er != EMULATE_FAIL) {
751 				/* Sign extend */
752 				if (inst.c0r_format.rs == mfc_op)
753 					val = (int)val;
754 				vcpu->arch.gprs[rt] = val;
755 			}
756 
757 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mfc_op) ?
758 					KVM_TRACE_MFC0 : KVM_TRACE_DMFC0,
759 				      KVM_TRACE_COP0(rd, sel), val);
760 			break;
761 
762 		case dmtc_op:
763 		case mtc_op:
764 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
765 			cop0->stat[rd][sel]++;
766 #endif
767 			val = vcpu->arch.gprs[rt];
768 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mtc_op) ?
769 					KVM_TRACE_MTC0 : KVM_TRACE_DMTC0,
770 				      KVM_TRACE_COP0(rd, sel), val);
771 
772 			if (rd == MIPS_CP0_COUNT &&
773 			    sel == 0) {			/* Count */
774 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
775 			} else if (rd == MIPS_CP0_COMPARE &&
776 				   sel == 0) {		/* Compare */
777 				kvm_mips_write_compare(vcpu,
778 						       vcpu->arch.gprs[rt],
779 						       true);
780 			} else if (rd == MIPS_CP0_ERRCTL &&
781 				   (sel == 0)) {	/* ErrCtl */
782 				/* ignore the written value */
783 			} else {
784 				er = EMULATE_FAIL;
785 			}
786 			break;
787 
788 		default:
789 			er = EMULATE_FAIL;
790 			break;
791 		}
792 	}
793 	/* Rollback PC only if emulation was unsuccessful */
794 	if (er == EMULATE_FAIL) {
795 		kvm_err("[%#lx]%s: unsupported cop0 instruction 0x%08x\n",
796 			curr_pc, __func__, inst.word);
797 
798 		vcpu->arch.pc = curr_pc;
799 	}
800 
801 	return er;
802 }
803 
804 static enum emulation_result kvm_vz_gpsi_cache(union mips_instruction inst,
805 					       u32 *opc, u32 cause,
806 					       struct kvm_run *run,
807 					       struct kvm_vcpu *vcpu)
808 {
809 	enum emulation_result er = EMULATE_DONE;
810 	u32 cache, op_inst, op, base;
811 	s16 offset;
812 	struct kvm_vcpu_arch *arch = &vcpu->arch;
813 	unsigned long va, curr_pc;
814 
815 	/*
816 	 * Update PC and hold onto current PC in case there is
817 	 * an error and we want to rollback the PC
818 	 */
819 	curr_pc = vcpu->arch.pc;
820 	er = update_pc(vcpu, cause);
821 	if (er == EMULATE_FAIL)
822 		return er;
823 
824 	base = inst.i_format.rs;
825 	op_inst = inst.i_format.rt;
826 	if (cpu_has_mips_r6)
827 		offset = inst.spec3_format.simmediate;
828 	else
829 		offset = inst.i_format.simmediate;
830 	cache = op_inst & CacheOp_Cache;
831 	op = op_inst & CacheOp_Op;
832 
833 	va = arch->gprs[base] + offset;
834 
835 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
836 		  cache, op, base, arch->gprs[base], offset);
837 
838 	/* Secondary or tirtiary cache ops ignored */
839 	if (cache != Cache_I && cache != Cache_D)
840 		return EMULATE_DONE;
841 
842 	switch (op_inst) {
843 	case Index_Invalidate_I:
844 		flush_icache_line_indexed(va);
845 		return EMULATE_DONE;
846 	case Index_Writeback_Inv_D:
847 		flush_dcache_line_indexed(va);
848 		return EMULATE_DONE;
849 	default:
850 		break;
851 	};
852 
853 	kvm_err("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
854 		curr_pc, vcpu->arch.gprs[31], cache, op, base, arch->gprs[base],
855 		offset);
856 	/* Rollback PC */
857 	vcpu->arch.pc = curr_pc;
858 
859 	return EMULATE_FAIL;
860 }
861 
862 static enum emulation_result kvm_trap_vz_handle_gpsi(u32 cause, u32 *opc,
863 						     struct kvm_vcpu *vcpu)
864 {
865 	enum emulation_result er = EMULATE_DONE;
866 	struct kvm_vcpu_arch *arch = &vcpu->arch;
867 	struct kvm_run *run = vcpu->run;
868 	union mips_instruction inst;
869 	int rd, rt, sel;
870 	int err;
871 
872 	/*
873 	 *  Fetch the instruction.
874 	 */
875 	if (cause & CAUSEF_BD)
876 		opc += 1;
877 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
878 	if (err)
879 		return EMULATE_FAIL;
880 
881 	switch (inst.r_format.opcode) {
882 	case cop0_op:
883 		er = kvm_vz_gpsi_cop0(inst, opc, cause, run, vcpu);
884 		break;
885 #ifndef CONFIG_CPU_MIPSR6
886 	case cache_op:
887 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
888 		er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
889 		break;
890 #endif
891 	case spec3_op:
892 		switch (inst.spec3_format.func) {
893 #ifdef CONFIG_CPU_MIPSR6
894 		case cache6_op:
895 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
896 			er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
897 			break;
898 #endif
899 		case rdhwr_op:
900 			if (inst.r_format.rs || (inst.r_format.re >> 3))
901 				goto unknown;
902 
903 			rd = inst.r_format.rd;
904 			rt = inst.r_format.rt;
905 			sel = inst.r_format.re & 0x7;
906 
907 			switch (rd) {
908 			case MIPS_HWR_CC:	/* Read count register */
909 				arch->gprs[rt] =
910 					(long)(int)kvm_mips_read_count(vcpu);
911 				break;
912 			default:
913 				trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
914 					      KVM_TRACE_HWR(rd, sel), 0);
915 				goto unknown;
916 			};
917 
918 			trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
919 				      KVM_TRACE_HWR(rd, sel), arch->gprs[rt]);
920 
921 			er = update_pc(vcpu, cause);
922 			break;
923 		default:
924 			goto unknown;
925 		};
926 		break;
927 unknown:
928 
929 	default:
930 		kvm_err("GPSI exception not supported (%p/%#x)\n",
931 				opc, inst.word);
932 		kvm_arch_vcpu_dump_regs(vcpu);
933 		er = EMULATE_FAIL;
934 		break;
935 	}
936 
937 	return er;
938 }
939 
940 static enum emulation_result kvm_trap_vz_handle_gsfc(u32 cause, u32 *opc,
941 						     struct kvm_vcpu *vcpu)
942 {
943 	enum emulation_result er = EMULATE_DONE;
944 	struct kvm_vcpu_arch *arch = &vcpu->arch;
945 	union mips_instruction inst;
946 	int err;
947 
948 	/*
949 	 *  Fetch the instruction.
950 	 */
951 	if (cause & CAUSEF_BD)
952 		opc += 1;
953 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
954 	if (err)
955 		return EMULATE_FAIL;
956 
957 	/* complete MTC0 on behalf of guest and advance EPC */
958 	if (inst.c0r_format.opcode == cop0_op &&
959 	    inst.c0r_format.rs == mtc_op &&
960 	    inst.c0r_format.z == 0) {
961 		int rt = inst.c0r_format.rt;
962 		int rd = inst.c0r_format.rd;
963 		int sel = inst.c0r_format.sel;
964 		unsigned int val = arch->gprs[rt];
965 		unsigned int old_val, change;
966 
967 		trace_kvm_hwr(vcpu, KVM_TRACE_MTC0, KVM_TRACE_COP0(rd, sel),
968 			      val);
969 
970 		if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
971 			/* FR bit should read as zero if no FPU */
972 			if (!kvm_mips_guest_has_fpu(&vcpu->arch))
973 				val &= ~(ST0_CU1 | ST0_FR);
974 
975 			/*
976 			 * Also don't allow FR to be set if host doesn't support
977 			 * it.
978 			 */
979 			if (!(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
980 				val &= ~ST0_FR;
981 
982 			old_val = read_gc0_status();
983 			change = val ^ old_val;
984 
985 			if (change & ST0_FR) {
986 				/*
987 				 * FPU and Vector register state is made
988 				 * UNPREDICTABLE by a change of FR, so don't
989 				 * even bother saving it.
990 				 */
991 				kvm_drop_fpu(vcpu);
992 			}
993 
994 			/*
995 			 * If MSA state is already live, it is undefined how it
996 			 * interacts with FR=0 FPU state, and we don't want to
997 			 * hit reserved instruction exceptions trying to save
998 			 * the MSA state later when CU=1 && FR=1, so play it
999 			 * safe and save it first.
1000 			 */
1001 			if (change & ST0_CU1 && !(val & ST0_FR) &&
1002 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1003 				kvm_lose_fpu(vcpu);
1004 
1005 			write_gc0_status(val);
1006 		} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1007 			u32 old_cause = read_gc0_cause();
1008 			u32 change = old_cause ^ val;
1009 
1010 			/* DC bit enabling/disabling timer? */
1011 			if (change & CAUSEF_DC) {
1012 				if (val & CAUSEF_DC)
1013 					kvm_mips_count_disable_cause(vcpu);
1014 				else
1015 					kvm_mips_count_enable_cause(vcpu);
1016 			}
1017 
1018 			/* Only certain bits are RW to the guest */
1019 			change &= (CAUSEF_DC | CAUSEF_IV | CAUSEF_WP |
1020 				   CAUSEF_IP0 | CAUSEF_IP1);
1021 
1022 			/* WP can only be cleared */
1023 			change &= ~CAUSEF_WP | old_cause;
1024 
1025 			write_gc0_cause(old_cause ^ change);
1026 		} else if ((rd == MIPS_CP0_STATUS) && (sel == 1)) { /* IntCtl */
1027 			write_gc0_intctl(val);
1028 		} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1029 			old_val = read_gc0_config5();
1030 			change = val ^ old_val;
1031 			/* Handle changes in FPU/MSA modes */
1032 			preempt_disable();
1033 
1034 			/*
1035 			 * Propagate FRE changes immediately if the FPU
1036 			 * context is already loaded.
1037 			 */
1038 			if (change & MIPS_CONF5_FRE &&
1039 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1040 				change_c0_config5(MIPS_CONF5_FRE, val);
1041 
1042 			preempt_enable();
1043 
1044 			val = old_val ^
1045 				(change & kvm_vz_config5_guest_wrmask(vcpu));
1046 			write_gc0_config5(val);
1047 		} else {
1048 			kvm_err("Handle GSFC, unsupported field change @ %p: %#x\n",
1049 			    opc, inst.word);
1050 			er = EMULATE_FAIL;
1051 		}
1052 
1053 		if (er != EMULATE_FAIL)
1054 			er = update_pc(vcpu, cause);
1055 	} else {
1056 		kvm_err("Handle GSFC, unrecognized instruction @ %p: %#x\n",
1057 			opc, inst.word);
1058 		er = EMULATE_FAIL;
1059 	}
1060 
1061 	return er;
1062 }
1063 
1064 static enum emulation_result kvm_trap_vz_handle_hc(u32 cause, u32 *opc,
1065 						   struct kvm_vcpu *vcpu)
1066 {
1067 	enum emulation_result er;
1068 	union mips_instruction inst;
1069 	unsigned long curr_pc;
1070 	int err;
1071 
1072 	if (cause & CAUSEF_BD)
1073 		opc += 1;
1074 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1075 	if (err)
1076 		return EMULATE_FAIL;
1077 
1078 	/*
1079 	 * Update PC and hold onto current PC in case there is
1080 	 * an error and we want to rollback the PC
1081 	 */
1082 	curr_pc = vcpu->arch.pc;
1083 	er = update_pc(vcpu, cause);
1084 	if (er == EMULATE_FAIL)
1085 		return er;
1086 
1087 	er = kvm_mips_emul_hypcall(vcpu, inst);
1088 	if (er == EMULATE_FAIL)
1089 		vcpu->arch.pc = curr_pc;
1090 
1091 	return er;
1092 }
1093 
1094 static enum emulation_result kvm_trap_vz_no_handler_guest_exit(u32 gexccode,
1095 							u32 cause,
1096 							u32 *opc,
1097 							struct kvm_vcpu *vcpu)
1098 {
1099 	u32 inst;
1100 
1101 	/*
1102 	 *  Fetch the instruction.
1103 	 */
1104 	if (cause & CAUSEF_BD)
1105 		opc += 1;
1106 	kvm_get_badinstr(opc, vcpu, &inst);
1107 
1108 	kvm_err("Guest Exception Code: %d not yet handled @ PC: %p, inst: 0x%08x  Status: %#x\n",
1109 		gexccode, opc, inst, read_gc0_status());
1110 
1111 	return EMULATE_FAIL;
1112 }
1113 
1114 static int kvm_trap_vz_handle_guest_exit(struct kvm_vcpu *vcpu)
1115 {
1116 	u32 *opc = (u32 *) vcpu->arch.pc;
1117 	u32 cause = vcpu->arch.host_cp0_cause;
1118 	enum emulation_result er = EMULATE_DONE;
1119 	u32 gexccode = (vcpu->arch.host_cp0_guestctl0 &
1120 			MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
1121 	int ret = RESUME_GUEST;
1122 
1123 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_GEXCCODE_BASE + gexccode);
1124 	switch (gexccode) {
1125 	case MIPS_GCTL0_GEXC_GPSI:
1126 		++vcpu->stat.vz_gpsi_exits;
1127 		er = kvm_trap_vz_handle_gpsi(cause, opc, vcpu);
1128 		break;
1129 	case MIPS_GCTL0_GEXC_GSFC:
1130 		++vcpu->stat.vz_gsfc_exits;
1131 		er = kvm_trap_vz_handle_gsfc(cause, opc, vcpu);
1132 		break;
1133 	case MIPS_GCTL0_GEXC_HC:
1134 		++vcpu->stat.vz_hc_exits;
1135 		er = kvm_trap_vz_handle_hc(cause, opc, vcpu);
1136 		break;
1137 	case MIPS_GCTL0_GEXC_GRR:
1138 		++vcpu->stat.vz_grr_exits;
1139 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1140 						       vcpu);
1141 		break;
1142 	case MIPS_GCTL0_GEXC_GVA:
1143 		++vcpu->stat.vz_gva_exits;
1144 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1145 						       vcpu);
1146 		break;
1147 	case MIPS_GCTL0_GEXC_GHFC:
1148 		++vcpu->stat.vz_ghfc_exits;
1149 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1150 						       vcpu);
1151 		break;
1152 	case MIPS_GCTL0_GEXC_GPA:
1153 		++vcpu->stat.vz_gpa_exits;
1154 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1155 						       vcpu);
1156 		break;
1157 	default:
1158 		++vcpu->stat.vz_resvd_exits;
1159 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1160 						       vcpu);
1161 		break;
1162 
1163 	}
1164 
1165 	if (er == EMULATE_DONE) {
1166 		ret = RESUME_GUEST;
1167 	} else if (er == EMULATE_HYPERCALL) {
1168 		ret = kvm_mips_handle_hypcall(vcpu);
1169 	} else {
1170 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1171 		ret = RESUME_HOST;
1172 	}
1173 	return ret;
1174 }
1175 
1176 /**
1177  * kvm_trap_vz_handle_cop_unusuable() - Guest used unusable coprocessor.
1178  * @vcpu:	Virtual CPU context.
1179  *
1180  * Handle when the guest attempts to use a coprocessor which hasn't been allowed
1181  * by the root context.
1182  */
1183 static int kvm_trap_vz_handle_cop_unusable(struct kvm_vcpu *vcpu)
1184 {
1185 	struct kvm_run *run = vcpu->run;
1186 	u32 cause = vcpu->arch.host_cp0_cause;
1187 	enum emulation_result er = EMULATE_FAIL;
1188 	int ret = RESUME_GUEST;
1189 
1190 	if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 1) {
1191 		/*
1192 		 * If guest FPU not present, the FPU operation should have been
1193 		 * treated as a reserved instruction!
1194 		 * If FPU already in use, we shouldn't get this at all.
1195 		 */
1196 		if (WARN_ON(!kvm_mips_guest_has_fpu(&vcpu->arch) ||
1197 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1198 			preempt_enable();
1199 			return EMULATE_FAIL;
1200 		}
1201 
1202 		kvm_own_fpu(vcpu);
1203 		er = EMULATE_DONE;
1204 	}
1205 	/* other coprocessors not handled */
1206 
1207 	switch (er) {
1208 	case EMULATE_DONE:
1209 		ret = RESUME_GUEST;
1210 		break;
1211 
1212 	case EMULATE_FAIL:
1213 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1214 		ret = RESUME_HOST;
1215 		break;
1216 
1217 	default:
1218 		BUG();
1219 	}
1220 	return ret;
1221 }
1222 
1223 /**
1224  * kvm_trap_vz_handle_msa_disabled() - Guest used MSA while disabled in root.
1225  * @vcpu:	Virtual CPU context.
1226  *
1227  * Handle when the guest attempts to use MSA when it is disabled in the root
1228  * context.
1229  */
1230 static int kvm_trap_vz_handle_msa_disabled(struct kvm_vcpu *vcpu)
1231 {
1232 	struct kvm_run *run = vcpu->run;
1233 
1234 	/*
1235 	 * If MSA not present or not exposed to guest or FR=0, the MSA operation
1236 	 * should have been treated as a reserved instruction!
1237 	 * Same if CU1=1, FR=0.
1238 	 * If MSA already in use, we shouldn't get this at all.
1239 	 */
1240 	if (!kvm_mips_guest_has_msa(&vcpu->arch) ||
1241 	    (read_gc0_status() & (ST0_CU1 | ST0_FR)) == ST0_CU1 ||
1242 	    !(read_gc0_config5() & MIPS_CONF5_MSAEN) ||
1243 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1244 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1245 		return RESUME_HOST;
1246 	}
1247 
1248 	kvm_own_msa(vcpu);
1249 
1250 	return RESUME_GUEST;
1251 }
1252 
1253 static int kvm_trap_vz_handle_tlb_ld_miss(struct kvm_vcpu *vcpu)
1254 {
1255 	struct kvm_run *run = vcpu->run;
1256 	u32 *opc = (u32 *) vcpu->arch.pc;
1257 	u32 cause = vcpu->arch.host_cp0_cause;
1258 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1259 	union mips_instruction inst;
1260 	enum emulation_result er = EMULATE_DONE;
1261 	int err, ret = RESUME_GUEST;
1262 
1263 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, false)) {
1264 		/* A code fetch fault doesn't count as an MMIO */
1265 		if (kvm_is_ifetch_fault(&vcpu->arch)) {
1266 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1267 			return RESUME_HOST;
1268 		}
1269 
1270 		/* Fetch the instruction */
1271 		if (cause & CAUSEF_BD)
1272 			opc += 1;
1273 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1274 		if (err) {
1275 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1276 			return RESUME_HOST;
1277 		}
1278 
1279 		/* Treat as MMIO */
1280 		er = kvm_mips_emulate_load(inst, cause, run, vcpu);
1281 		if (er == EMULATE_FAIL) {
1282 			kvm_err("Guest Emulate Load from MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1283 				opc, badvaddr);
1284 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1285 		}
1286 	}
1287 
1288 	if (er == EMULATE_DONE) {
1289 		ret = RESUME_GUEST;
1290 	} else if (er == EMULATE_DO_MMIO) {
1291 		run->exit_reason = KVM_EXIT_MMIO;
1292 		ret = RESUME_HOST;
1293 	} else {
1294 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1295 		ret = RESUME_HOST;
1296 	}
1297 	return ret;
1298 }
1299 
1300 static int kvm_trap_vz_handle_tlb_st_miss(struct kvm_vcpu *vcpu)
1301 {
1302 	struct kvm_run *run = vcpu->run;
1303 	u32 *opc = (u32 *) vcpu->arch.pc;
1304 	u32 cause = vcpu->arch.host_cp0_cause;
1305 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1306 	union mips_instruction inst;
1307 	enum emulation_result er = EMULATE_DONE;
1308 	int err;
1309 	int ret = RESUME_GUEST;
1310 
1311 	/* Just try the access again if we couldn't do the translation */
1312 	if (kvm_vz_badvaddr_to_gpa(vcpu, badvaddr, &badvaddr))
1313 		return RESUME_GUEST;
1314 	vcpu->arch.host_cp0_badvaddr = badvaddr;
1315 
1316 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, true)) {
1317 		/* Fetch the instruction */
1318 		if (cause & CAUSEF_BD)
1319 			opc += 1;
1320 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1321 		if (err) {
1322 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1323 			return RESUME_HOST;
1324 		}
1325 
1326 		/* Treat as MMIO */
1327 		er = kvm_mips_emulate_store(inst, cause, run, vcpu);
1328 		if (er == EMULATE_FAIL) {
1329 			kvm_err("Guest Emulate Store to MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1330 				opc, badvaddr);
1331 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1332 		}
1333 	}
1334 
1335 	if (er == EMULATE_DONE) {
1336 		ret = RESUME_GUEST;
1337 	} else if (er == EMULATE_DO_MMIO) {
1338 		run->exit_reason = KVM_EXIT_MMIO;
1339 		ret = RESUME_HOST;
1340 	} else {
1341 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1342 		ret = RESUME_HOST;
1343 	}
1344 	return ret;
1345 }
1346 
1347 static u64 kvm_vz_get_one_regs[] = {
1348 	KVM_REG_MIPS_CP0_INDEX,
1349 	KVM_REG_MIPS_CP0_ENTRYLO0,
1350 	KVM_REG_MIPS_CP0_ENTRYLO1,
1351 	KVM_REG_MIPS_CP0_CONTEXT,
1352 	KVM_REG_MIPS_CP0_PAGEMASK,
1353 	KVM_REG_MIPS_CP0_PAGEGRAIN,
1354 	KVM_REG_MIPS_CP0_WIRED,
1355 	KVM_REG_MIPS_CP0_HWRENA,
1356 	KVM_REG_MIPS_CP0_BADVADDR,
1357 	KVM_REG_MIPS_CP0_COUNT,
1358 	KVM_REG_MIPS_CP0_ENTRYHI,
1359 	KVM_REG_MIPS_CP0_COMPARE,
1360 	KVM_REG_MIPS_CP0_STATUS,
1361 	KVM_REG_MIPS_CP0_INTCTL,
1362 	KVM_REG_MIPS_CP0_CAUSE,
1363 	KVM_REG_MIPS_CP0_EPC,
1364 	KVM_REG_MIPS_CP0_PRID,
1365 	KVM_REG_MIPS_CP0_EBASE,
1366 	KVM_REG_MIPS_CP0_CONFIG,
1367 	KVM_REG_MIPS_CP0_CONFIG1,
1368 	KVM_REG_MIPS_CP0_CONFIG2,
1369 	KVM_REG_MIPS_CP0_CONFIG3,
1370 	KVM_REG_MIPS_CP0_CONFIG4,
1371 	KVM_REG_MIPS_CP0_CONFIG5,
1372 #ifdef CONFIG_64BIT
1373 	KVM_REG_MIPS_CP0_XCONTEXT,
1374 #endif
1375 	KVM_REG_MIPS_CP0_ERROREPC,
1376 
1377 	KVM_REG_MIPS_COUNT_CTL,
1378 	KVM_REG_MIPS_COUNT_RESUME,
1379 	KVM_REG_MIPS_COUNT_HZ,
1380 };
1381 
1382 static u64 kvm_vz_get_one_regs_contextconfig[] = {
1383 	KVM_REG_MIPS_CP0_CONTEXTCONFIG,
1384 #ifdef CONFIG_64BIT
1385 	KVM_REG_MIPS_CP0_XCONTEXTCONFIG,
1386 #endif
1387 };
1388 
1389 static u64 kvm_vz_get_one_regs_segments[] = {
1390 	KVM_REG_MIPS_CP0_SEGCTL0,
1391 	KVM_REG_MIPS_CP0_SEGCTL1,
1392 	KVM_REG_MIPS_CP0_SEGCTL2,
1393 };
1394 
1395 static u64 kvm_vz_get_one_regs_kscratch[] = {
1396 	KVM_REG_MIPS_CP0_KSCRATCH1,
1397 	KVM_REG_MIPS_CP0_KSCRATCH2,
1398 	KVM_REG_MIPS_CP0_KSCRATCH3,
1399 	KVM_REG_MIPS_CP0_KSCRATCH4,
1400 	KVM_REG_MIPS_CP0_KSCRATCH5,
1401 	KVM_REG_MIPS_CP0_KSCRATCH6,
1402 };
1403 
1404 static unsigned long kvm_vz_num_regs(struct kvm_vcpu *vcpu)
1405 {
1406 	unsigned long ret;
1407 
1408 	ret = ARRAY_SIZE(kvm_vz_get_one_regs);
1409 	if (cpu_guest_has_userlocal)
1410 		++ret;
1411 	if (cpu_guest_has_badinstr)
1412 		++ret;
1413 	if (cpu_guest_has_badinstrp)
1414 		++ret;
1415 	if (cpu_guest_has_contextconfig)
1416 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1417 	if (cpu_guest_has_segments)
1418 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1419 	ret += __arch_hweight8(cpu_data[0].guest.kscratch_mask);
1420 
1421 	return ret;
1422 }
1423 
1424 static int kvm_vz_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
1425 {
1426 	u64 index;
1427 	unsigned int i;
1428 
1429 	if (copy_to_user(indices, kvm_vz_get_one_regs,
1430 			 sizeof(kvm_vz_get_one_regs)))
1431 		return -EFAULT;
1432 	indices += ARRAY_SIZE(kvm_vz_get_one_regs);
1433 
1434 	if (cpu_guest_has_userlocal) {
1435 		index = KVM_REG_MIPS_CP0_USERLOCAL;
1436 		if (copy_to_user(indices, &index, sizeof(index)))
1437 			return -EFAULT;
1438 		++indices;
1439 	}
1440 	if (cpu_guest_has_badinstr) {
1441 		index = KVM_REG_MIPS_CP0_BADINSTR;
1442 		if (copy_to_user(indices, &index, sizeof(index)))
1443 			return -EFAULT;
1444 		++indices;
1445 	}
1446 	if (cpu_guest_has_badinstrp) {
1447 		index = KVM_REG_MIPS_CP0_BADINSTRP;
1448 		if (copy_to_user(indices, &index, sizeof(index)))
1449 			return -EFAULT;
1450 		++indices;
1451 	}
1452 	if (cpu_guest_has_contextconfig) {
1453 		if (copy_to_user(indices, kvm_vz_get_one_regs_contextconfig,
1454 				 sizeof(kvm_vz_get_one_regs_contextconfig)))
1455 			return -EFAULT;
1456 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1457 	}
1458 	if (cpu_guest_has_segments) {
1459 		if (copy_to_user(indices, kvm_vz_get_one_regs_segments,
1460 				 sizeof(kvm_vz_get_one_regs_segments)))
1461 			return -EFAULT;
1462 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1463 	}
1464 	for (i = 0; i < 6; ++i) {
1465 		if (!cpu_guest_has_kscr(i + 2))
1466 			continue;
1467 
1468 		if (copy_to_user(indices, &kvm_vz_get_one_regs_kscratch[i],
1469 				 sizeof(kvm_vz_get_one_regs_kscratch[i])))
1470 			return -EFAULT;
1471 		++indices;
1472 	}
1473 
1474 	return 0;
1475 }
1476 
1477 static inline s64 entrylo_kvm_to_user(unsigned long v)
1478 {
1479 	s64 mask, ret = v;
1480 
1481 	if (BITS_PER_LONG == 32) {
1482 		/*
1483 		 * KVM API exposes 64-bit version of the register, so move the
1484 		 * RI/XI bits up into place.
1485 		 */
1486 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1487 		ret &= ~mask;
1488 		ret |= ((s64)v & mask) << 32;
1489 	}
1490 	return ret;
1491 }
1492 
1493 static inline unsigned long entrylo_user_to_kvm(s64 v)
1494 {
1495 	unsigned long mask, ret = v;
1496 
1497 	if (BITS_PER_LONG == 32) {
1498 		/*
1499 		 * KVM API exposes 64-bit versiono of the register, so move the
1500 		 * RI/XI bits down into place.
1501 		 */
1502 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1503 		ret &= ~mask;
1504 		ret |= (v >> 32) & mask;
1505 	}
1506 	return ret;
1507 }
1508 
1509 static int kvm_vz_get_one_reg(struct kvm_vcpu *vcpu,
1510 			      const struct kvm_one_reg *reg,
1511 			      s64 *v)
1512 {
1513 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1514 	unsigned int idx;
1515 
1516 	switch (reg->id) {
1517 	case KVM_REG_MIPS_CP0_INDEX:
1518 		*v = (long)read_gc0_index();
1519 		break;
1520 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1521 		*v = entrylo_kvm_to_user(read_gc0_entrylo0());
1522 		break;
1523 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1524 		*v = entrylo_kvm_to_user(read_gc0_entrylo1());
1525 		break;
1526 	case KVM_REG_MIPS_CP0_CONTEXT:
1527 		*v = (long)read_gc0_context();
1528 		break;
1529 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1530 		if (!cpu_guest_has_contextconfig)
1531 			return -EINVAL;
1532 		*v = read_gc0_contextconfig();
1533 		break;
1534 	case KVM_REG_MIPS_CP0_USERLOCAL:
1535 		if (!cpu_guest_has_userlocal)
1536 			return -EINVAL;
1537 		*v = read_gc0_userlocal();
1538 		break;
1539 #ifdef CONFIG_64BIT
1540 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1541 		if (!cpu_guest_has_contextconfig)
1542 			return -EINVAL;
1543 		*v = read_gc0_xcontextconfig();
1544 		break;
1545 #endif
1546 	case KVM_REG_MIPS_CP0_PAGEMASK:
1547 		*v = (long)read_gc0_pagemask();
1548 		break;
1549 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1550 		*v = (long)read_gc0_pagegrain();
1551 		break;
1552 	case KVM_REG_MIPS_CP0_SEGCTL0:
1553 		if (!cpu_guest_has_segments)
1554 			return -EINVAL;
1555 		*v = read_gc0_segctl0();
1556 		break;
1557 	case KVM_REG_MIPS_CP0_SEGCTL1:
1558 		if (!cpu_guest_has_segments)
1559 			return -EINVAL;
1560 		*v = read_gc0_segctl1();
1561 		break;
1562 	case KVM_REG_MIPS_CP0_SEGCTL2:
1563 		if (!cpu_guest_has_segments)
1564 			return -EINVAL;
1565 		*v = read_gc0_segctl2();
1566 		break;
1567 	case KVM_REG_MIPS_CP0_WIRED:
1568 		*v = (long)read_gc0_wired();
1569 		break;
1570 	case KVM_REG_MIPS_CP0_HWRENA:
1571 		*v = (long)read_gc0_hwrena();
1572 		break;
1573 	case KVM_REG_MIPS_CP0_BADVADDR:
1574 		*v = (long)read_gc0_badvaddr();
1575 		break;
1576 	case KVM_REG_MIPS_CP0_BADINSTR:
1577 		if (!cpu_guest_has_badinstr)
1578 			return -EINVAL;
1579 		*v = read_gc0_badinstr();
1580 		break;
1581 	case KVM_REG_MIPS_CP0_BADINSTRP:
1582 		if (!cpu_guest_has_badinstrp)
1583 			return -EINVAL;
1584 		*v = read_gc0_badinstrp();
1585 		break;
1586 	case KVM_REG_MIPS_CP0_COUNT:
1587 		*v = kvm_mips_read_count(vcpu);
1588 		break;
1589 	case KVM_REG_MIPS_CP0_ENTRYHI:
1590 		*v = (long)read_gc0_entryhi();
1591 		break;
1592 	case KVM_REG_MIPS_CP0_COMPARE:
1593 		*v = (long)read_gc0_compare();
1594 		break;
1595 	case KVM_REG_MIPS_CP0_STATUS:
1596 		*v = (long)read_gc0_status();
1597 		break;
1598 	case KVM_REG_MIPS_CP0_INTCTL:
1599 		*v = read_gc0_intctl();
1600 		break;
1601 	case KVM_REG_MIPS_CP0_CAUSE:
1602 		*v = (long)read_gc0_cause();
1603 		break;
1604 	case KVM_REG_MIPS_CP0_EPC:
1605 		*v = (long)read_gc0_epc();
1606 		break;
1607 	case KVM_REG_MIPS_CP0_PRID:
1608 		*v = (long)kvm_read_c0_guest_prid(cop0);
1609 		break;
1610 	case KVM_REG_MIPS_CP0_EBASE:
1611 		*v = kvm_vz_read_gc0_ebase();
1612 		break;
1613 	case KVM_REG_MIPS_CP0_CONFIG:
1614 		*v = read_gc0_config();
1615 		break;
1616 	case KVM_REG_MIPS_CP0_CONFIG1:
1617 		if (!cpu_guest_has_conf1)
1618 			return -EINVAL;
1619 		*v = read_gc0_config1();
1620 		break;
1621 	case KVM_REG_MIPS_CP0_CONFIG2:
1622 		if (!cpu_guest_has_conf2)
1623 			return -EINVAL;
1624 		*v = read_gc0_config2();
1625 		break;
1626 	case KVM_REG_MIPS_CP0_CONFIG3:
1627 		if (!cpu_guest_has_conf3)
1628 			return -EINVAL;
1629 		*v = read_gc0_config3();
1630 		break;
1631 	case KVM_REG_MIPS_CP0_CONFIG4:
1632 		if (!cpu_guest_has_conf4)
1633 			return -EINVAL;
1634 		*v = read_gc0_config4();
1635 		break;
1636 	case KVM_REG_MIPS_CP0_CONFIG5:
1637 		if (!cpu_guest_has_conf5)
1638 			return -EINVAL;
1639 		*v = read_gc0_config5();
1640 		break;
1641 #ifdef CONFIG_64BIT
1642 	case KVM_REG_MIPS_CP0_XCONTEXT:
1643 		*v = read_gc0_xcontext();
1644 		break;
1645 #endif
1646 	case KVM_REG_MIPS_CP0_ERROREPC:
1647 		*v = (long)read_gc0_errorepc();
1648 		break;
1649 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
1650 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
1651 		if (!cpu_guest_has_kscr(idx))
1652 			return -EINVAL;
1653 		switch (idx) {
1654 		case 2:
1655 			*v = (long)read_gc0_kscratch1();
1656 			break;
1657 		case 3:
1658 			*v = (long)read_gc0_kscratch2();
1659 			break;
1660 		case 4:
1661 			*v = (long)read_gc0_kscratch3();
1662 			break;
1663 		case 5:
1664 			*v = (long)read_gc0_kscratch4();
1665 			break;
1666 		case 6:
1667 			*v = (long)read_gc0_kscratch5();
1668 			break;
1669 		case 7:
1670 			*v = (long)read_gc0_kscratch6();
1671 			break;
1672 		}
1673 		break;
1674 	case KVM_REG_MIPS_COUNT_CTL:
1675 		*v = vcpu->arch.count_ctl;
1676 		break;
1677 	case KVM_REG_MIPS_COUNT_RESUME:
1678 		*v = ktime_to_ns(vcpu->arch.count_resume);
1679 		break;
1680 	case KVM_REG_MIPS_COUNT_HZ:
1681 		*v = vcpu->arch.count_hz;
1682 		break;
1683 	default:
1684 		return -EINVAL;
1685 	}
1686 	return 0;
1687 }
1688 
1689 static int kvm_vz_set_one_reg(struct kvm_vcpu *vcpu,
1690 			      const struct kvm_one_reg *reg,
1691 			      s64 v)
1692 {
1693 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1694 	unsigned int idx;
1695 	int ret = 0;
1696 	unsigned int cur, change;
1697 
1698 	switch (reg->id) {
1699 	case KVM_REG_MIPS_CP0_INDEX:
1700 		write_gc0_index(v);
1701 		break;
1702 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1703 		write_gc0_entrylo0(entrylo_user_to_kvm(v));
1704 		break;
1705 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1706 		write_gc0_entrylo1(entrylo_user_to_kvm(v));
1707 		break;
1708 	case KVM_REG_MIPS_CP0_CONTEXT:
1709 		write_gc0_context(v);
1710 		break;
1711 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1712 		if (!cpu_guest_has_contextconfig)
1713 			return -EINVAL;
1714 		write_gc0_contextconfig(v);
1715 		break;
1716 	case KVM_REG_MIPS_CP0_USERLOCAL:
1717 		if (!cpu_guest_has_userlocal)
1718 			return -EINVAL;
1719 		write_gc0_userlocal(v);
1720 		break;
1721 #ifdef CONFIG_64BIT
1722 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1723 		if (!cpu_guest_has_contextconfig)
1724 			return -EINVAL;
1725 		write_gc0_xcontextconfig(v);
1726 		break;
1727 #endif
1728 	case KVM_REG_MIPS_CP0_PAGEMASK:
1729 		write_gc0_pagemask(v);
1730 		break;
1731 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1732 		write_gc0_pagegrain(v);
1733 		break;
1734 	case KVM_REG_MIPS_CP0_SEGCTL0:
1735 		if (!cpu_guest_has_segments)
1736 			return -EINVAL;
1737 		write_gc0_segctl0(v);
1738 		break;
1739 	case KVM_REG_MIPS_CP0_SEGCTL1:
1740 		if (!cpu_guest_has_segments)
1741 			return -EINVAL;
1742 		write_gc0_segctl1(v);
1743 		break;
1744 	case KVM_REG_MIPS_CP0_SEGCTL2:
1745 		if (!cpu_guest_has_segments)
1746 			return -EINVAL;
1747 		write_gc0_segctl2(v);
1748 		break;
1749 	case KVM_REG_MIPS_CP0_WIRED:
1750 		change_gc0_wired(MIPSR6_WIRED_WIRED, v);
1751 		break;
1752 	case KVM_REG_MIPS_CP0_HWRENA:
1753 		write_gc0_hwrena(v);
1754 		break;
1755 	case KVM_REG_MIPS_CP0_BADVADDR:
1756 		write_gc0_badvaddr(v);
1757 		break;
1758 	case KVM_REG_MIPS_CP0_BADINSTR:
1759 		if (!cpu_guest_has_badinstr)
1760 			return -EINVAL;
1761 		write_gc0_badinstr(v);
1762 		break;
1763 	case KVM_REG_MIPS_CP0_BADINSTRP:
1764 		if (!cpu_guest_has_badinstrp)
1765 			return -EINVAL;
1766 		write_gc0_badinstrp(v);
1767 		break;
1768 	case KVM_REG_MIPS_CP0_COUNT:
1769 		kvm_mips_write_count(vcpu, v);
1770 		break;
1771 	case KVM_REG_MIPS_CP0_ENTRYHI:
1772 		write_gc0_entryhi(v);
1773 		break;
1774 	case KVM_REG_MIPS_CP0_COMPARE:
1775 		kvm_mips_write_compare(vcpu, v, false);
1776 		break;
1777 	case KVM_REG_MIPS_CP0_STATUS:
1778 		write_gc0_status(v);
1779 		break;
1780 	case KVM_REG_MIPS_CP0_INTCTL:
1781 		write_gc0_intctl(v);
1782 		break;
1783 	case KVM_REG_MIPS_CP0_CAUSE:
1784 		/*
1785 		 * If the timer is stopped or started (DC bit) it must look
1786 		 * atomic with changes to the timer interrupt pending bit (TI).
1787 		 * A timer interrupt should not happen in between.
1788 		 */
1789 		if ((read_gc0_cause() ^ v) & CAUSEF_DC) {
1790 			if (v & CAUSEF_DC) {
1791 				/* disable timer first */
1792 				kvm_mips_count_disable_cause(vcpu);
1793 				change_gc0_cause((u32)~CAUSEF_DC, v);
1794 			} else {
1795 				/* enable timer last */
1796 				change_gc0_cause((u32)~CAUSEF_DC, v);
1797 				kvm_mips_count_enable_cause(vcpu);
1798 			}
1799 		} else {
1800 			write_gc0_cause(v);
1801 		}
1802 		break;
1803 	case KVM_REG_MIPS_CP0_EPC:
1804 		write_gc0_epc(v);
1805 		break;
1806 	case KVM_REG_MIPS_CP0_PRID:
1807 		kvm_write_c0_guest_prid(cop0, v);
1808 		break;
1809 	case KVM_REG_MIPS_CP0_EBASE:
1810 		kvm_vz_write_gc0_ebase(v);
1811 		break;
1812 	case KVM_REG_MIPS_CP0_CONFIG:
1813 		cur = read_gc0_config();
1814 		change = (cur ^ v) & kvm_vz_config_user_wrmask(vcpu);
1815 		if (change) {
1816 			v = cur ^ change;
1817 			write_gc0_config(v);
1818 		}
1819 		break;
1820 	case KVM_REG_MIPS_CP0_CONFIG1:
1821 		if (!cpu_guest_has_conf1)
1822 			break;
1823 		cur = read_gc0_config1();
1824 		change = (cur ^ v) & kvm_vz_config1_user_wrmask(vcpu);
1825 		if (change) {
1826 			v = cur ^ change;
1827 			write_gc0_config1(v);
1828 		}
1829 		break;
1830 	case KVM_REG_MIPS_CP0_CONFIG2:
1831 		if (!cpu_guest_has_conf2)
1832 			break;
1833 		cur = read_gc0_config2();
1834 		change = (cur ^ v) & kvm_vz_config2_user_wrmask(vcpu);
1835 		if (change) {
1836 			v = cur ^ change;
1837 			write_gc0_config2(v);
1838 		}
1839 		break;
1840 	case KVM_REG_MIPS_CP0_CONFIG3:
1841 		if (!cpu_guest_has_conf3)
1842 			break;
1843 		cur = read_gc0_config3();
1844 		change = (cur ^ v) & kvm_vz_config3_user_wrmask(vcpu);
1845 		if (change) {
1846 			v = cur ^ change;
1847 			write_gc0_config3(v);
1848 		}
1849 		break;
1850 	case KVM_REG_MIPS_CP0_CONFIG4:
1851 		if (!cpu_guest_has_conf4)
1852 			break;
1853 		cur = read_gc0_config4();
1854 		change = (cur ^ v) & kvm_vz_config4_user_wrmask(vcpu);
1855 		if (change) {
1856 			v = cur ^ change;
1857 			write_gc0_config4(v);
1858 		}
1859 		break;
1860 	case KVM_REG_MIPS_CP0_CONFIG5:
1861 		if (!cpu_guest_has_conf5)
1862 			break;
1863 		cur = read_gc0_config5();
1864 		change = (cur ^ v) & kvm_vz_config5_user_wrmask(vcpu);
1865 		if (change) {
1866 			v = cur ^ change;
1867 			write_gc0_config5(v);
1868 		}
1869 		break;
1870 #ifdef CONFIG_64BIT
1871 	case KVM_REG_MIPS_CP0_XCONTEXT:
1872 		write_gc0_xcontext(v);
1873 		break;
1874 #endif
1875 	case KVM_REG_MIPS_CP0_ERROREPC:
1876 		write_gc0_errorepc(v);
1877 		break;
1878 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
1879 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
1880 		if (!cpu_guest_has_kscr(idx))
1881 			return -EINVAL;
1882 		switch (idx) {
1883 		case 2:
1884 			write_gc0_kscratch1(v);
1885 			break;
1886 		case 3:
1887 			write_gc0_kscratch2(v);
1888 			break;
1889 		case 4:
1890 			write_gc0_kscratch3(v);
1891 			break;
1892 		case 5:
1893 			write_gc0_kscratch4(v);
1894 			break;
1895 		case 6:
1896 			write_gc0_kscratch5(v);
1897 			break;
1898 		case 7:
1899 			write_gc0_kscratch6(v);
1900 			break;
1901 		}
1902 		break;
1903 	case KVM_REG_MIPS_COUNT_CTL:
1904 		ret = kvm_mips_set_count_ctl(vcpu, v);
1905 		break;
1906 	case KVM_REG_MIPS_COUNT_RESUME:
1907 		ret = kvm_mips_set_count_resume(vcpu, v);
1908 		break;
1909 	case KVM_REG_MIPS_COUNT_HZ:
1910 		ret = kvm_mips_set_count_hz(vcpu, v);
1911 		break;
1912 	default:
1913 		return -EINVAL;
1914 	}
1915 	return ret;
1916 }
1917 
1918 #define guestid_cache(cpu)	(cpu_data[cpu].guestid_cache)
1919 static void kvm_vz_get_new_guestid(unsigned long cpu, struct kvm_vcpu *vcpu)
1920 {
1921 	unsigned long guestid = guestid_cache(cpu);
1922 
1923 	if (!(++guestid & GUESTID_MASK)) {
1924 		if (cpu_has_vtag_icache)
1925 			flush_icache_all();
1926 
1927 		if (!guestid)		/* fix version if needed */
1928 			guestid = GUESTID_FIRST_VERSION;
1929 
1930 		++guestid;		/* guestid 0 reserved for root */
1931 
1932 		/* start new guestid cycle */
1933 		kvm_vz_local_flush_roottlb_all_guests();
1934 		kvm_vz_local_flush_guesttlb_all();
1935 	}
1936 
1937 	guestid_cache(cpu) = guestid;
1938 }
1939 
1940 /* Returns 1 if the guest TLB may be clobbered */
1941 static int kvm_vz_check_requests(struct kvm_vcpu *vcpu, int cpu)
1942 {
1943 	int ret = 0;
1944 	int i;
1945 
1946 	if (!vcpu->requests)
1947 		return 0;
1948 
1949 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
1950 		if (cpu_has_guestid) {
1951 			/* Drop all GuestIDs for this VCPU */
1952 			for_each_possible_cpu(i)
1953 				vcpu->arch.vzguestid[i] = 0;
1954 			/* This will clobber guest TLB contents too */
1955 			ret = 1;
1956 		}
1957 		/*
1958 		 * For Root ASID Dealias (RAD) we don't do anything here, but we
1959 		 * still need the request to ensure we recheck asid_flush_mask.
1960 		 * We can still return 0 as only the root TLB will be affected
1961 		 * by a root ASID flush.
1962 		 */
1963 	}
1964 
1965 	return ret;
1966 }
1967 
1968 static void kvm_vz_vcpu_save_wired(struct kvm_vcpu *vcpu)
1969 {
1970 	unsigned int wired = read_gc0_wired();
1971 	struct kvm_mips_tlb *tlbs;
1972 	int i;
1973 
1974 	/* Expand the wired TLB array if necessary */
1975 	wired &= MIPSR6_WIRED_WIRED;
1976 	if (wired > vcpu->arch.wired_tlb_limit) {
1977 		tlbs = krealloc(vcpu->arch.wired_tlb, wired *
1978 				sizeof(*vcpu->arch.wired_tlb), GFP_ATOMIC);
1979 		if (WARN_ON(!tlbs)) {
1980 			/* Save whatever we can */
1981 			wired = vcpu->arch.wired_tlb_limit;
1982 		} else {
1983 			vcpu->arch.wired_tlb = tlbs;
1984 			vcpu->arch.wired_tlb_limit = wired;
1985 		}
1986 	}
1987 
1988 	if (wired)
1989 		/* Save wired entries from the guest TLB */
1990 		kvm_vz_save_guesttlb(vcpu->arch.wired_tlb, 0, wired);
1991 	/* Invalidate any dropped entries since last time */
1992 	for (i = wired; i < vcpu->arch.wired_tlb_used; ++i) {
1993 		vcpu->arch.wired_tlb[i].tlb_hi = UNIQUE_GUEST_ENTRYHI(i);
1994 		vcpu->arch.wired_tlb[i].tlb_lo[0] = 0;
1995 		vcpu->arch.wired_tlb[i].tlb_lo[1] = 0;
1996 		vcpu->arch.wired_tlb[i].tlb_mask = 0;
1997 	}
1998 	vcpu->arch.wired_tlb_used = wired;
1999 }
2000 
2001 static void kvm_vz_vcpu_load_wired(struct kvm_vcpu *vcpu)
2002 {
2003 	/* Load wired entries into the guest TLB */
2004 	if (vcpu->arch.wired_tlb)
2005 		kvm_vz_load_guesttlb(vcpu->arch.wired_tlb, 0,
2006 				     vcpu->arch.wired_tlb_used);
2007 }
2008 
2009 static void kvm_vz_vcpu_load_tlb(struct kvm_vcpu *vcpu, int cpu)
2010 {
2011 	struct kvm *kvm = vcpu->kvm;
2012 	struct mm_struct *gpa_mm = &kvm->arch.gpa_mm;
2013 	bool migrated;
2014 
2015 	/*
2016 	 * Are we entering guest context on a different CPU to last time?
2017 	 * If so, the VCPU's guest TLB state on this CPU may be stale.
2018 	 */
2019 	migrated = (vcpu->arch.last_exec_cpu != cpu);
2020 	vcpu->arch.last_exec_cpu = cpu;
2021 
2022 	/*
2023 	 * A vcpu's GuestID is set in GuestCtl1.ID when the vcpu is loaded and
2024 	 * remains set until another vcpu is loaded in.  As a rule GuestRID
2025 	 * remains zeroed when in root context unless the kernel is busy
2026 	 * manipulating guest tlb entries.
2027 	 */
2028 	if (cpu_has_guestid) {
2029 		/*
2030 		 * Check if our GuestID is of an older version and thus invalid.
2031 		 *
2032 		 * We also discard the stored GuestID if we've executed on
2033 		 * another CPU, as the guest mappings may have changed without
2034 		 * hypervisor knowledge.
2035 		 */
2036 		if (migrated ||
2037 		    (vcpu->arch.vzguestid[cpu] ^ guestid_cache(cpu)) &
2038 					GUESTID_VERSION_MASK) {
2039 			kvm_vz_get_new_guestid(cpu, vcpu);
2040 			vcpu->arch.vzguestid[cpu] = guestid_cache(cpu);
2041 			trace_kvm_guestid_change(vcpu,
2042 						 vcpu->arch.vzguestid[cpu]);
2043 		}
2044 
2045 		/* Restore GuestID */
2046 		change_c0_guestctl1(GUESTID_MASK, vcpu->arch.vzguestid[cpu]);
2047 	} else {
2048 		/*
2049 		 * The Guest TLB only stores a single guest's TLB state, so
2050 		 * flush it if another VCPU has executed on this CPU.
2051 		 *
2052 		 * We also flush if we've executed on another CPU, as the guest
2053 		 * mappings may have changed without hypervisor knowledge.
2054 		 */
2055 		if (migrated || last_exec_vcpu[cpu] != vcpu)
2056 			kvm_vz_local_flush_guesttlb_all();
2057 		last_exec_vcpu[cpu] = vcpu;
2058 
2059 		/*
2060 		 * Root ASID dealiases guest GPA mappings in the root TLB.
2061 		 * Allocate new root ASID if needed.
2062 		 */
2063 		if (cpumask_test_and_clear_cpu(cpu, &kvm->arch.asid_flush_mask)
2064 		    || (cpu_context(cpu, gpa_mm) ^ asid_cache(cpu)) &
2065 						asid_version_mask(cpu))
2066 			get_new_mmu_context(gpa_mm, cpu);
2067 	}
2068 }
2069 
2070 static int kvm_vz_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2071 {
2072 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2073 	bool migrated, all;
2074 
2075 	/*
2076 	 * Have we migrated to a different CPU?
2077 	 * If so, any old guest TLB state may be stale.
2078 	 */
2079 	migrated = (vcpu->arch.last_sched_cpu != cpu);
2080 
2081 	/*
2082 	 * Was this the last VCPU to run on this CPU?
2083 	 * If not, any old guest state from this VCPU will have been clobbered.
2084 	 */
2085 	all = migrated || (last_vcpu[cpu] != vcpu);
2086 	last_vcpu[cpu] = vcpu;
2087 
2088 	/*
2089 	 * Restore CP0_Wired unconditionally as we clear it after use, and
2090 	 * restore wired guest TLB entries (while in guest context).
2091 	 */
2092 	kvm_restore_gc0_wired(cop0);
2093 	if (current->flags & PF_VCPU) {
2094 		tlbw_use_hazard();
2095 		kvm_vz_vcpu_load_tlb(vcpu, cpu);
2096 		kvm_vz_vcpu_load_wired(vcpu);
2097 	}
2098 
2099 	/*
2100 	 * Restore timer state regardless, as e.g. Cause.TI can change over time
2101 	 * if left unmaintained.
2102 	 */
2103 	kvm_vz_restore_timer(vcpu);
2104 
2105 	/* Don't bother restoring registers multiple times unless necessary */
2106 	if (!all)
2107 		return 0;
2108 
2109 	/*
2110 	 * Restore config registers first, as some implementations restrict
2111 	 * writes to other registers when the corresponding feature bits aren't
2112 	 * set. For example Status.CU1 cannot be set unless Config1.FP is set.
2113 	 */
2114 	kvm_restore_gc0_config(cop0);
2115 	if (cpu_guest_has_conf1)
2116 		kvm_restore_gc0_config1(cop0);
2117 	if (cpu_guest_has_conf2)
2118 		kvm_restore_gc0_config2(cop0);
2119 	if (cpu_guest_has_conf3)
2120 		kvm_restore_gc0_config3(cop0);
2121 	if (cpu_guest_has_conf4)
2122 		kvm_restore_gc0_config4(cop0);
2123 	if (cpu_guest_has_conf5)
2124 		kvm_restore_gc0_config5(cop0);
2125 	if (cpu_guest_has_conf6)
2126 		kvm_restore_gc0_config6(cop0);
2127 	if (cpu_guest_has_conf7)
2128 		kvm_restore_gc0_config7(cop0);
2129 
2130 	kvm_restore_gc0_index(cop0);
2131 	kvm_restore_gc0_entrylo0(cop0);
2132 	kvm_restore_gc0_entrylo1(cop0);
2133 	kvm_restore_gc0_context(cop0);
2134 	if (cpu_guest_has_contextconfig)
2135 		kvm_restore_gc0_contextconfig(cop0);
2136 #ifdef CONFIG_64BIT
2137 	kvm_restore_gc0_xcontext(cop0);
2138 	if (cpu_guest_has_contextconfig)
2139 		kvm_restore_gc0_xcontextconfig(cop0);
2140 #endif
2141 	kvm_restore_gc0_pagemask(cop0);
2142 	kvm_restore_gc0_pagegrain(cop0);
2143 	kvm_restore_gc0_hwrena(cop0);
2144 	kvm_restore_gc0_badvaddr(cop0);
2145 	kvm_restore_gc0_entryhi(cop0);
2146 	kvm_restore_gc0_status(cop0);
2147 	kvm_restore_gc0_intctl(cop0);
2148 	kvm_restore_gc0_epc(cop0);
2149 	kvm_vz_write_gc0_ebase(kvm_read_sw_gc0_ebase(cop0));
2150 	if (cpu_guest_has_userlocal)
2151 		kvm_restore_gc0_userlocal(cop0);
2152 
2153 	kvm_restore_gc0_errorepc(cop0);
2154 
2155 	/* restore KScratch registers if enabled in guest */
2156 	if (cpu_guest_has_conf4) {
2157 		if (cpu_guest_has_kscr(2))
2158 			kvm_restore_gc0_kscratch1(cop0);
2159 		if (cpu_guest_has_kscr(3))
2160 			kvm_restore_gc0_kscratch2(cop0);
2161 		if (cpu_guest_has_kscr(4))
2162 			kvm_restore_gc0_kscratch3(cop0);
2163 		if (cpu_guest_has_kscr(5))
2164 			kvm_restore_gc0_kscratch4(cop0);
2165 		if (cpu_guest_has_kscr(6))
2166 			kvm_restore_gc0_kscratch5(cop0);
2167 		if (cpu_guest_has_kscr(7))
2168 			kvm_restore_gc0_kscratch6(cop0);
2169 	}
2170 
2171 	if (cpu_guest_has_badinstr)
2172 		kvm_restore_gc0_badinstr(cop0);
2173 	if (cpu_guest_has_badinstrp)
2174 		kvm_restore_gc0_badinstrp(cop0);
2175 
2176 	if (cpu_guest_has_segments) {
2177 		kvm_restore_gc0_segctl0(cop0);
2178 		kvm_restore_gc0_segctl1(cop0);
2179 		kvm_restore_gc0_segctl2(cop0);
2180 	}
2181 
2182 	/* restore Root.GuestCtl2 from unused Guest guestctl2 register */
2183 	if (cpu_has_guestctl2)
2184 		write_c0_guestctl2(
2185 			cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL]);
2186 
2187 	return 0;
2188 }
2189 
2190 static int kvm_vz_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
2191 {
2192 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2193 
2194 	if (current->flags & PF_VCPU)
2195 		kvm_vz_vcpu_save_wired(vcpu);
2196 
2197 	kvm_lose_fpu(vcpu);
2198 
2199 	kvm_save_gc0_index(cop0);
2200 	kvm_save_gc0_entrylo0(cop0);
2201 	kvm_save_gc0_entrylo1(cop0);
2202 	kvm_save_gc0_context(cop0);
2203 	if (cpu_guest_has_contextconfig)
2204 		kvm_save_gc0_contextconfig(cop0);
2205 #ifdef CONFIG_64BIT
2206 	kvm_save_gc0_xcontext(cop0);
2207 	if (cpu_guest_has_contextconfig)
2208 		kvm_save_gc0_xcontextconfig(cop0);
2209 #endif
2210 	kvm_save_gc0_pagemask(cop0);
2211 	kvm_save_gc0_pagegrain(cop0);
2212 	kvm_save_gc0_wired(cop0);
2213 	/* allow wired TLB entries to be overwritten */
2214 	clear_gc0_wired(MIPSR6_WIRED_WIRED);
2215 	kvm_save_gc0_hwrena(cop0);
2216 	kvm_save_gc0_badvaddr(cop0);
2217 	kvm_save_gc0_entryhi(cop0);
2218 	kvm_save_gc0_status(cop0);
2219 	kvm_save_gc0_intctl(cop0);
2220 	kvm_save_gc0_epc(cop0);
2221 	kvm_write_sw_gc0_ebase(cop0, kvm_vz_read_gc0_ebase());
2222 	if (cpu_guest_has_userlocal)
2223 		kvm_save_gc0_userlocal(cop0);
2224 
2225 	/* only save implemented config registers */
2226 	kvm_save_gc0_config(cop0);
2227 	if (cpu_guest_has_conf1)
2228 		kvm_save_gc0_config1(cop0);
2229 	if (cpu_guest_has_conf2)
2230 		kvm_save_gc0_config2(cop0);
2231 	if (cpu_guest_has_conf3)
2232 		kvm_save_gc0_config3(cop0);
2233 	if (cpu_guest_has_conf4)
2234 		kvm_save_gc0_config4(cop0);
2235 	if (cpu_guest_has_conf5)
2236 		kvm_save_gc0_config5(cop0);
2237 	if (cpu_guest_has_conf6)
2238 		kvm_save_gc0_config6(cop0);
2239 	if (cpu_guest_has_conf7)
2240 		kvm_save_gc0_config7(cop0);
2241 
2242 	kvm_save_gc0_errorepc(cop0);
2243 
2244 	/* save KScratch registers if enabled in guest */
2245 	if (cpu_guest_has_conf4) {
2246 		if (cpu_guest_has_kscr(2))
2247 			kvm_save_gc0_kscratch1(cop0);
2248 		if (cpu_guest_has_kscr(3))
2249 			kvm_save_gc0_kscratch2(cop0);
2250 		if (cpu_guest_has_kscr(4))
2251 			kvm_save_gc0_kscratch3(cop0);
2252 		if (cpu_guest_has_kscr(5))
2253 			kvm_save_gc0_kscratch4(cop0);
2254 		if (cpu_guest_has_kscr(6))
2255 			kvm_save_gc0_kscratch5(cop0);
2256 		if (cpu_guest_has_kscr(7))
2257 			kvm_save_gc0_kscratch6(cop0);
2258 	}
2259 
2260 	if (cpu_guest_has_badinstr)
2261 		kvm_save_gc0_badinstr(cop0);
2262 	if (cpu_guest_has_badinstrp)
2263 		kvm_save_gc0_badinstrp(cop0);
2264 
2265 	if (cpu_guest_has_segments) {
2266 		kvm_save_gc0_segctl0(cop0);
2267 		kvm_save_gc0_segctl1(cop0);
2268 		kvm_save_gc0_segctl2(cop0);
2269 	}
2270 
2271 	kvm_vz_save_timer(vcpu);
2272 
2273 	/* save Root.GuestCtl2 in unused Guest guestctl2 register */
2274 	if (cpu_has_guestctl2)
2275 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] =
2276 			read_c0_guestctl2();
2277 
2278 	return 0;
2279 }
2280 
2281 /**
2282  * kvm_vz_resize_guest_vtlb() - Attempt to resize guest VTLB.
2283  * @size:	Number of guest VTLB entries (0 < @size <= root VTLB entries).
2284  *
2285  * Attempt to resize the guest VTLB by writing guest Config registers. This is
2286  * necessary for cores with a shared root/guest TLB to avoid overlap with wired
2287  * entries in the root VTLB.
2288  *
2289  * Returns:	The resulting guest VTLB size.
2290  */
2291 static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
2292 {
2293 	unsigned int config4 = 0, ret = 0, limit;
2294 
2295 	/* Write MMUSize - 1 into guest Config registers */
2296 	if (cpu_guest_has_conf1)
2297 		change_gc0_config1(MIPS_CONF1_TLBS,
2298 				   (size - 1) << MIPS_CONF1_TLBS_SHIFT);
2299 	if (cpu_guest_has_conf4) {
2300 		config4 = read_gc0_config4();
2301 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2302 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT) {
2303 			config4 &= ~MIPS_CONF4_VTLBSIZEEXT;
2304 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2305 				MIPS_CONF4_VTLBSIZEEXT_SHIFT;
2306 		} else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2307 			   MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT) {
2308 			config4 &= ~MIPS_CONF4_MMUSIZEEXT;
2309 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2310 				MIPS_CONF4_MMUSIZEEXT_SHIFT;
2311 		}
2312 		write_gc0_config4(config4);
2313 	}
2314 
2315 	/*
2316 	 * Set Guest.Wired.Limit = 0 (no limit up to Guest.MMUSize-1), unless it
2317 	 * would exceed Root.Wired.Limit (clearing Guest.Wired.Wired so write
2318 	 * not dropped)
2319 	 */
2320 	if (cpu_has_mips_r6) {
2321 		limit = (read_c0_wired() & MIPSR6_WIRED_LIMIT) >>
2322 						MIPSR6_WIRED_LIMIT_SHIFT;
2323 		if (size - 1 <= limit)
2324 			limit = 0;
2325 		write_gc0_wired(limit << MIPSR6_WIRED_LIMIT_SHIFT);
2326 	}
2327 
2328 	/* Read back MMUSize - 1 */
2329 	back_to_back_c0_hazard();
2330 	if (cpu_guest_has_conf1)
2331 		ret = (read_gc0_config1() & MIPS_CONF1_TLBS) >>
2332 						MIPS_CONF1_TLBS_SHIFT;
2333 	if (config4) {
2334 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2335 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT)
2336 			ret |= ((config4 & MIPS_CONF4_VTLBSIZEEXT) >>
2337 				MIPS_CONF4_VTLBSIZEEXT_SHIFT) <<
2338 				MIPS_CONF1_TLBS_SIZE;
2339 		else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2340 			 MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT)
2341 			ret |= ((config4 & MIPS_CONF4_MMUSIZEEXT) >>
2342 				MIPS_CONF4_MMUSIZEEXT_SHIFT) <<
2343 				MIPS_CONF1_TLBS_SIZE;
2344 	}
2345 	return ret + 1;
2346 }
2347 
2348 static int kvm_vz_hardware_enable(void)
2349 {
2350 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
2351 
2352 	/*
2353 	 * ImgTec cores tend to use a shared root/guest TLB. To avoid overlap of
2354 	 * root wired and guest entries, the guest TLB may need resizing.
2355 	 */
2356 	mmu_size = current_cpu_data.tlbsizevtlb;
2357 	ftlb_size = current_cpu_data.tlbsize - mmu_size;
2358 
2359 	/* Try switching to maximum guest VTLB size for flush */
2360 	guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
2361 	current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2362 	kvm_vz_local_flush_guesttlb_all();
2363 
2364 	/*
2365 	 * Reduce to make space for root wired entries and at least 2 root
2366 	 * non-wired entries. This does assume that long-term wired entries
2367 	 * won't be added later.
2368 	 */
2369 	guest_mmu_size = mmu_size - num_wired_entries() - 2;
2370 	guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
2371 	current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2372 
2373 	/*
2374 	 * Write the VTLB size, but if another CPU has already written, check it
2375 	 * matches or we won't provide a consistent view to the guest. If this
2376 	 * ever happens it suggests an asymmetric number of wired entries.
2377 	 */
2378 	if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
2379 	    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
2380 		 "Available guest VTLB size mismatch"))
2381 		return -EINVAL;
2382 
2383 	/*
2384 	 * Enable virtualization features granting guest direct control of
2385 	 * certain features:
2386 	 * CP0=1:	Guest coprocessor 0 context.
2387 	 * AT=Guest:	Guest MMU.
2388 	 * CG=1:	Hit (virtual address) CACHE operations (optional).
2389 	 * CF=1:	Guest Config registers.
2390 	 * CGI=1:	Indexed flush CACHE operations (optional).
2391 	 */
2392 	write_c0_guestctl0(MIPS_GCTL0_CP0 |
2393 			   (MIPS_GCTL0_AT_GUEST << MIPS_GCTL0_AT_SHIFT) |
2394 			   MIPS_GCTL0_CG | MIPS_GCTL0_CF);
2395 	if (cpu_has_guestctl0ext)
2396 		set_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2397 
2398 	if (cpu_has_guestid) {
2399 		write_c0_guestctl1(0);
2400 		kvm_vz_local_flush_roottlb_all_guests();
2401 
2402 		GUESTID_MASK = current_cpu_data.guestid_mask;
2403 		GUESTID_FIRST_VERSION = GUESTID_MASK + 1;
2404 		GUESTID_VERSION_MASK = ~GUESTID_MASK;
2405 
2406 		current_cpu_data.guestid_cache = GUESTID_FIRST_VERSION;
2407 	}
2408 
2409 	/* clear any pending injected virtual guest interrupts */
2410 	if (cpu_has_guestctl2)
2411 		clear_c0_guestctl2(0x3f << 10);
2412 
2413 	return 0;
2414 }
2415 
2416 static void kvm_vz_hardware_disable(void)
2417 {
2418 	kvm_vz_local_flush_guesttlb_all();
2419 
2420 	if (cpu_has_guestid) {
2421 		write_c0_guestctl1(0);
2422 		kvm_vz_local_flush_roottlb_all_guests();
2423 	}
2424 }
2425 
2426 static int kvm_vz_check_extension(struct kvm *kvm, long ext)
2427 {
2428 	int r;
2429 
2430 	switch (ext) {
2431 	case KVM_CAP_MIPS_VZ:
2432 		/* we wouldn't be here unless cpu_has_vz */
2433 		r = 1;
2434 		break;
2435 #ifdef CONFIG_64BIT
2436 	case KVM_CAP_MIPS_64BIT:
2437 		/* We support 64-bit registers/operations and addresses */
2438 		r = 2;
2439 		break;
2440 #endif
2441 	default:
2442 		r = 0;
2443 		break;
2444 	}
2445 
2446 	return r;
2447 }
2448 
2449 static int kvm_vz_vcpu_init(struct kvm_vcpu *vcpu)
2450 {
2451 	int i;
2452 
2453 	for_each_possible_cpu(i)
2454 		vcpu->arch.vzguestid[i] = 0;
2455 
2456 	return 0;
2457 }
2458 
2459 static void kvm_vz_vcpu_uninit(struct kvm_vcpu *vcpu)
2460 {
2461 	int cpu;
2462 
2463 	/*
2464 	 * If the VCPU is freed and reused as another VCPU, we don't want the
2465 	 * matching pointer wrongly hanging around in last_vcpu[] or
2466 	 * last_exec_vcpu[].
2467 	 */
2468 	for_each_possible_cpu(cpu) {
2469 		if (last_vcpu[cpu] == vcpu)
2470 			last_vcpu[cpu] = NULL;
2471 		if (last_exec_vcpu[cpu] == vcpu)
2472 			last_exec_vcpu[cpu] = NULL;
2473 	}
2474 }
2475 
2476 static int kvm_vz_vcpu_setup(struct kvm_vcpu *vcpu)
2477 {
2478 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2479 	unsigned long count_hz = 100*1000*1000; /* default to 100 MHz */
2480 
2481 	/*
2482 	 * Start off the timer at the same frequency as the host timer, but the
2483 	 * soft timer doesn't handle frequencies greater than 1GHz yet.
2484 	 */
2485 	if (mips_hpt_frequency && mips_hpt_frequency <= NSEC_PER_SEC)
2486 		count_hz = mips_hpt_frequency;
2487 	kvm_mips_init_count(vcpu, count_hz);
2488 
2489 	/*
2490 	 * Initialize guest register state to valid architectural reset state.
2491 	 */
2492 
2493 	/* PageGrain */
2494 	if (cpu_has_mips_r6)
2495 		kvm_write_sw_gc0_pagegrain(cop0, PG_RIE | PG_XIE | PG_IEC);
2496 	/* Wired */
2497 	if (cpu_has_mips_r6)
2498 		kvm_write_sw_gc0_wired(cop0,
2499 				       read_gc0_wired() & MIPSR6_WIRED_LIMIT);
2500 	/* Status */
2501 	kvm_write_sw_gc0_status(cop0, ST0_BEV | ST0_ERL);
2502 	if (cpu_has_mips_r6)
2503 		kvm_change_sw_gc0_status(cop0, ST0_FR, read_gc0_status());
2504 	/* IntCtl */
2505 	kvm_write_sw_gc0_intctl(cop0, read_gc0_intctl() &
2506 				(INTCTLF_IPFDC | INTCTLF_IPPCI | INTCTLF_IPTI));
2507 	/* PRId */
2508 	kvm_write_sw_gc0_prid(cop0, boot_cpu_data.processor_id);
2509 	/* EBase */
2510 	kvm_write_sw_gc0_ebase(cop0, (s32)0x80000000 | vcpu->vcpu_id);
2511 	/* Config */
2512 	kvm_save_gc0_config(cop0);
2513 	/* architecturally writable (e.g. from guest) */
2514 	kvm_change_sw_gc0_config(cop0, CONF_CM_CMASK,
2515 				 _page_cachable_default >> _CACHE_SHIFT);
2516 	/* architecturally read only, but maybe writable from root */
2517 	kvm_change_sw_gc0_config(cop0, MIPS_CONF_MT, read_c0_config());
2518 	if (cpu_guest_has_conf1) {
2519 		kvm_set_sw_gc0_config(cop0, MIPS_CONF_M);
2520 		/* Config1 */
2521 		kvm_save_gc0_config1(cop0);
2522 		/* architecturally read only, but maybe writable from root */
2523 		kvm_clear_sw_gc0_config1(cop0, MIPS_CONF1_C2	|
2524 					       MIPS_CONF1_MD	|
2525 					       MIPS_CONF1_PC	|
2526 					       MIPS_CONF1_WR	|
2527 					       MIPS_CONF1_CA	|
2528 					       MIPS_CONF1_FP);
2529 	}
2530 	if (cpu_guest_has_conf2) {
2531 		kvm_set_sw_gc0_config1(cop0, MIPS_CONF_M);
2532 		/* Config2 */
2533 		kvm_save_gc0_config2(cop0);
2534 	}
2535 	if (cpu_guest_has_conf3) {
2536 		kvm_set_sw_gc0_config2(cop0, MIPS_CONF_M);
2537 		/* Config3 */
2538 		kvm_save_gc0_config3(cop0);
2539 		/* architecturally writable (e.g. from guest) */
2540 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_ISA_OE);
2541 		/* architecturally read only, but maybe writable from root */
2542 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_MSA	|
2543 					       MIPS_CONF3_BPG	|
2544 					       MIPS_CONF3_ULRI	|
2545 					       MIPS_CONF3_DSP	|
2546 					       MIPS_CONF3_CTXTC	|
2547 					       MIPS_CONF3_ITL	|
2548 					       MIPS_CONF3_LPA	|
2549 					       MIPS_CONF3_VEIC	|
2550 					       MIPS_CONF3_VINT	|
2551 					       MIPS_CONF3_SP	|
2552 					       MIPS_CONF3_CDMM	|
2553 					       MIPS_CONF3_MT	|
2554 					       MIPS_CONF3_SM	|
2555 					       MIPS_CONF3_TL);
2556 	}
2557 	if (cpu_guest_has_conf4) {
2558 		kvm_set_sw_gc0_config3(cop0, MIPS_CONF_M);
2559 		/* Config4 */
2560 		kvm_save_gc0_config4(cop0);
2561 	}
2562 	if (cpu_guest_has_conf5) {
2563 		kvm_set_sw_gc0_config4(cop0, MIPS_CONF_M);
2564 		/* Config5 */
2565 		kvm_save_gc0_config5(cop0);
2566 		/* architecturally writable (e.g. from guest) */
2567 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_K	|
2568 					       MIPS_CONF5_CV	|
2569 					       MIPS_CONF5_MSAEN	|
2570 					       MIPS_CONF5_UFE	|
2571 					       MIPS_CONF5_FRE	|
2572 					       MIPS_CONF5_SBRI	|
2573 					       MIPS_CONF5_UFR);
2574 		/* architecturally read only, but maybe writable from root */
2575 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_MRP);
2576 	}
2577 
2578 	if (cpu_guest_has_contextconfig) {
2579 		/* ContextConfig */
2580 		kvm_write_sw_gc0_contextconfig(cop0, 0x007ffff0);
2581 #ifdef CONFIG_64BIT
2582 		/* XContextConfig */
2583 		/* bits SEGBITS-13+3:4 set */
2584 		kvm_write_sw_gc0_xcontextconfig(cop0,
2585 					((1ull << (cpu_vmbits - 13)) - 1) << 4);
2586 #endif
2587 	}
2588 
2589 	/* Implementation dependent, use the legacy layout */
2590 	if (cpu_guest_has_segments) {
2591 		/* SegCtl0, SegCtl1, SegCtl2 */
2592 		kvm_write_sw_gc0_segctl0(cop0, 0x00200010);
2593 		kvm_write_sw_gc0_segctl1(cop0, 0x00000002 |
2594 				(_page_cachable_default >> _CACHE_SHIFT) <<
2595 						(16 + MIPS_SEGCFG_C_SHIFT));
2596 		kvm_write_sw_gc0_segctl2(cop0, 0x00380438);
2597 	}
2598 
2599 	/* start with no pending virtual guest interrupts */
2600 	if (cpu_has_guestctl2)
2601 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] = 0;
2602 
2603 	/* Put PC at reset vector */
2604 	vcpu->arch.pc = CKSEG1ADDR(0x1fc00000);
2605 
2606 	return 0;
2607 }
2608 
2609 static void kvm_vz_flush_shadow_all(struct kvm *kvm)
2610 {
2611 	if (cpu_has_guestid) {
2612 		/* Flush GuestID for each VCPU individually */
2613 		kvm_flush_remote_tlbs(kvm);
2614 	} else {
2615 		/*
2616 		 * For each CPU there is a single GPA ASID used by all VCPUs in
2617 		 * the VM, so it doesn't make sense for the VCPUs to handle
2618 		 * invalidation of these ASIDs individually.
2619 		 *
2620 		 * Instead mark all CPUs as needing ASID invalidation in
2621 		 * asid_flush_mask, and just use kvm_flush_remote_tlbs(kvm) to
2622 		 * kick any running VCPUs so they check asid_flush_mask.
2623 		 */
2624 		cpumask_setall(&kvm->arch.asid_flush_mask);
2625 		kvm_flush_remote_tlbs(kvm);
2626 	}
2627 }
2628 
2629 static void kvm_vz_flush_shadow_memslot(struct kvm *kvm,
2630 					const struct kvm_memory_slot *slot)
2631 {
2632 	kvm_vz_flush_shadow_all(kvm);
2633 }
2634 
2635 static void kvm_vz_vcpu_reenter(struct kvm_run *run, struct kvm_vcpu *vcpu)
2636 {
2637 	int cpu = smp_processor_id();
2638 	int preserve_guest_tlb;
2639 
2640 	preserve_guest_tlb = kvm_vz_check_requests(vcpu, cpu);
2641 
2642 	if (preserve_guest_tlb)
2643 		kvm_vz_vcpu_save_wired(vcpu);
2644 
2645 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
2646 
2647 	if (preserve_guest_tlb)
2648 		kvm_vz_vcpu_load_wired(vcpu);
2649 }
2650 
2651 static int kvm_vz_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
2652 {
2653 	int cpu = smp_processor_id();
2654 	int r;
2655 
2656 	/* Check if we have any exceptions/interrupts pending */
2657 	kvm_mips_deliver_interrupts(vcpu, read_gc0_cause());
2658 
2659 	kvm_vz_check_requests(vcpu, cpu);
2660 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
2661 	kvm_vz_vcpu_load_wired(vcpu);
2662 
2663 	r = vcpu->arch.vcpu_run(run, vcpu);
2664 
2665 	kvm_vz_vcpu_save_wired(vcpu);
2666 
2667 	return r;
2668 }
2669 
2670 static struct kvm_mips_callbacks kvm_vz_callbacks = {
2671 	.handle_cop_unusable = kvm_trap_vz_handle_cop_unusable,
2672 	.handle_tlb_mod = kvm_trap_vz_handle_tlb_st_miss,
2673 	.handle_tlb_ld_miss = kvm_trap_vz_handle_tlb_ld_miss,
2674 	.handle_tlb_st_miss = kvm_trap_vz_handle_tlb_st_miss,
2675 	.handle_addr_err_st = kvm_trap_vz_no_handler,
2676 	.handle_addr_err_ld = kvm_trap_vz_no_handler,
2677 	.handle_syscall = kvm_trap_vz_no_handler,
2678 	.handle_res_inst = kvm_trap_vz_no_handler,
2679 	.handle_break = kvm_trap_vz_no_handler,
2680 	.handle_msa_disabled = kvm_trap_vz_handle_msa_disabled,
2681 	.handle_guest_exit = kvm_trap_vz_handle_guest_exit,
2682 
2683 	.hardware_enable = kvm_vz_hardware_enable,
2684 	.hardware_disable = kvm_vz_hardware_disable,
2685 	.check_extension = kvm_vz_check_extension,
2686 	.vcpu_init = kvm_vz_vcpu_init,
2687 	.vcpu_uninit = kvm_vz_vcpu_uninit,
2688 	.vcpu_setup = kvm_vz_vcpu_setup,
2689 	.flush_shadow_all = kvm_vz_flush_shadow_all,
2690 	.flush_shadow_memslot = kvm_vz_flush_shadow_memslot,
2691 	.gva_to_gpa = kvm_vz_gva_to_gpa_cb,
2692 	.queue_timer_int = kvm_vz_queue_timer_int_cb,
2693 	.dequeue_timer_int = kvm_vz_dequeue_timer_int_cb,
2694 	.queue_io_int = kvm_vz_queue_io_int_cb,
2695 	.dequeue_io_int = kvm_vz_dequeue_io_int_cb,
2696 	.irq_deliver = kvm_vz_irq_deliver_cb,
2697 	.irq_clear = kvm_vz_irq_clear_cb,
2698 	.num_regs = kvm_vz_num_regs,
2699 	.copy_reg_indices = kvm_vz_copy_reg_indices,
2700 	.get_one_reg = kvm_vz_get_one_reg,
2701 	.set_one_reg = kvm_vz_set_one_reg,
2702 	.vcpu_load = kvm_vz_vcpu_load,
2703 	.vcpu_put = kvm_vz_vcpu_put,
2704 	.vcpu_run = kvm_vz_vcpu_run,
2705 	.vcpu_reenter = kvm_vz_vcpu_reenter,
2706 };
2707 
2708 int kvm_mips_emulation_init(struct kvm_mips_callbacks **install_callbacks)
2709 {
2710 	if (!cpu_has_vz)
2711 		return -ENODEV;
2712 
2713 	/*
2714 	 * VZ requires at least 2 KScratch registers, so it should have been
2715 	 * possible to allocate pgd_reg.
2716 	 */
2717 	if (WARN(pgd_reg == -1,
2718 		 "pgd_reg not allocated even though cpu_has_vz\n"))
2719 		return -ENODEV;
2720 
2721 	pr_info("Starting KVM with MIPS VZ extensions\n");
2722 
2723 	*install_callbacks = &kvm_vz_callbacks;
2724 	return 0;
2725 }
2726