xref: /openbmc/linux/arch/mips/kvm/emulate.c (revision 60772e48)
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: Instruction/Exception emulation
7  *
8  * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
9  * Authors: Sanjay Lal <sanjayl@kymasys.com>
10  */
11 
12 #include <linux/errno.h>
13 #include <linux/err.h>
14 #include <linux/ktime.h>
15 #include <linux/kvm_host.h>
16 #include <linux/vmalloc.h>
17 #include <linux/fs.h>
18 #include <linux/bootmem.h>
19 #include <linux/random.h>
20 #include <asm/page.h>
21 #include <asm/cacheflush.h>
22 #include <asm/cacheops.h>
23 #include <asm/cpu-info.h>
24 #include <asm/mmu_context.h>
25 #include <asm/tlbflush.h>
26 #include <asm/inst.h>
27 
28 #undef CONFIG_MIPS_MT
29 #include <asm/r4kcache.h>
30 #define CONFIG_MIPS_MT
31 
32 #include "interrupt.h"
33 #include "commpage.h"
34 
35 #include "trace.h"
36 
37 /*
38  * Compute the return address and do emulate branch simulation, if required.
39  * This function should be called only in branch delay slot active.
40  */
41 static int kvm_compute_return_epc(struct kvm_vcpu *vcpu, unsigned long instpc,
42 				  unsigned long *out)
43 {
44 	unsigned int dspcontrol;
45 	union mips_instruction insn;
46 	struct kvm_vcpu_arch *arch = &vcpu->arch;
47 	long epc = instpc;
48 	long nextpc;
49 	int err;
50 
51 	if (epc & 3) {
52 		kvm_err("%s: unaligned epc\n", __func__);
53 		return -EINVAL;
54 	}
55 
56 	/* Read the instruction */
57 	err = kvm_get_badinstrp((u32 *)epc, vcpu, &insn.word);
58 	if (err)
59 		return err;
60 
61 	switch (insn.i_format.opcode) {
62 		/* jr and jalr are in r_format format. */
63 	case spec_op:
64 		switch (insn.r_format.func) {
65 		case jalr_op:
66 			arch->gprs[insn.r_format.rd] = epc + 8;
67 			/* Fall through */
68 		case jr_op:
69 			nextpc = arch->gprs[insn.r_format.rs];
70 			break;
71 		default:
72 			return -EINVAL;
73 		}
74 		break;
75 
76 		/*
77 		 * This group contains:
78 		 * bltz_op, bgez_op, bltzl_op, bgezl_op,
79 		 * bltzal_op, bgezal_op, bltzall_op, bgezall_op.
80 		 */
81 	case bcond_op:
82 		switch (insn.i_format.rt) {
83 		case bltz_op:
84 		case bltzl_op:
85 			if ((long)arch->gprs[insn.i_format.rs] < 0)
86 				epc = epc + 4 + (insn.i_format.simmediate << 2);
87 			else
88 				epc += 8;
89 			nextpc = epc;
90 			break;
91 
92 		case bgez_op:
93 		case bgezl_op:
94 			if ((long)arch->gprs[insn.i_format.rs] >= 0)
95 				epc = epc + 4 + (insn.i_format.simmediate << 2);
96 			else
97 				epc += 8;
98 			nextpc = epc;
99 			break;
100 
101 		case bltzal_op:
102 		case bltzall_op:
103 			arch->gprs[31] = epc + 8;
104 			if ((long)arch->gprs[insn.i_format.rs] < 0)
105 				epc = epc + 4 + (insn.i_format.simmediate << 2);
106 			else
107 				epc += 8;
108 			nextpc = epc;
109 			break;
110 
111 		case bgezal_op:
112 		case bgezall_op:
113 			arch->gprs[31] = epc + 8;
114 			if ((long)arch->gprs[insn.i_format.rs] >= 0)
115 				epc = epc + 4 + (insn.i_format.simmediate << 2);
116 			else
117 				epc += 8;
118 			nextpc = epc;
119 			break;
120 		case bposge32_op:
121 			if (!cpu_has_dsp) {
122 				kvm_err("%s: DSP branch but not DSP ASE\n",
123 					__func__);
124 				return -EINVAL;
125 			}
126 
127 			dspcontrol = rddsp(0x01);
128 
129 			if (dspcontrol >= 32)
130 				epc = epc + 4 + (insn.i_format.simmediate << 2);
131 			else
132 				epc += 8;
133 			nextpc = epc;
134 			break;
135 		default:
136 			return -EINVAL;
137 		}
138 		break;
139 
140 		/* These are unconditional and in j_format. */
141 	case jal_op:
142 		arch->gprs[31] = instpc + 8;
143 	case j_op:
144 		epc += 4;
145 		epc >>= 28;
146 		epc <<= 28;
147 		epc |= (insn.j_format.target << 2);
148 		nextpc = epc;
149 		break;
150 
151 		/* These are conditional and in i_format. */
152 	case beq_op:
153 	case beql_op:
154 		if (arch->gprs[insn.i_format.rs] ==
155 		    arch->gprs[insn.i_format.rt])
156 			epc = epc + 4 + (insn.i_format.simmediate << 2);
157 		else
158 			epc += 8;
159 		nextpc = epc;
160 		break;
161 
162 	case bne_op:
163 	case bnel_op:
164 		if (arch->gprs[insn.i_format.rs] !=
165 		    arch->gprs[insn.i_format.rt])
166 			epc = epc + 4 + (insn.i_format.simmediate << 2);
167 		else
168 			epc += 8;
169 		nextpc = epc;
170 		break;
171 
172 	case blez_op:	/* POP06 */
173 #ifndef CONFIG_CPU_MIPSR6
174 	case blezl_op:	/* removed in R6 */
175 #endif
176 		if (insn.i_format.rt != 0)
177 			goto compact_branch;
178 		if ((long)arch->gprs[insn.i_format.rs] <= 0)
179 			epc = epc + 4 + (insn.i_format.simmediate << 2);
180 		else
181 			epc += 8;
182 		nextpc = epc;
183 		break;
184 
185 	case bgtz_op:	/* POP07 */
186 #ifndef CONFIG_CPU_MIPSR6
187 	case bgtzl_op:	/* removed in R6 */
188 #endif
189 		if (insn.i_format.rt != 0)
190 			goto compact_branch;
191 		if ((long)arch->gprs[insn.i_format.rs] > 0)
192 			epc = epc + 4 + (insn.i_format.simmediate << 2);
193 		else
194 			epc += 8;
195 		nextpc = epc;
196 		break;
197 
198 		/* And now the FPA/cp1 branch instructions. */
199 	case cop1_op:
200 		kvm_err("%s: unsupported cop1_op\n", __func__);
201 		return -EINVAL;
202 
203 #ifdef CONFIG_CPU_MIPSR6
204 	/* R6 added the following compact branches with forbidden slots */
205 	case blezl_op:	/* POP26 */
206 	case bgtzl_op:	/* POP27 */
207 		/* only rt == 0 isn't compact branch */
208 		if (insn.i_format.rt != 0)
209 			goto compact_branch;
210 		return -EINVAL;
211 	case pop10_op:
212 	case pop30_op:
213 		/* only rs == rt == 0 is reserved, rest are compact branches */
214 		if (insn.i_format.rs != 0 || insn.i_format.rt != 0)
215 			goto compact_branch;
216 		return -EINVAL;
217 	case pop66_op:
218 	case pop76_op:
219 		/* only rs == 0 isn't compact branch */
220 		if (insn.i_format.rs != 0)
221 			goto compact_branch;
222 		return -EINVAL;
223 compact_branch:
224 		/*
225 		 * If we've hit an exception on the forbidden slot, then
226 		 * the branch must not have been taken.
227 		 */
228 		epc += 8;
229 		nextpc = epc;
230 		break;
231 #else
232 compact_branch:
233 		/* Fall through - Compact branches not supported before R6 */
234 #endif
235 	default:
236 		return -EINVAL;
237 	}
238 
239 	*out = nextpc;
240 	return 0;
241 }
242 
243 enum emulation_result update_pc(struct kvm_vcpu *vcpu, u32 cause)
244 {
245 	int err;
246 
247 	if (cause & CAUSEF_BD) {
248 		err = kvm_compute_return_epc(vcpu, vcpu->arch.pc,
249 					     &vcpu->arch.pc);
250 		if (err)
251 			return EMULATE_FAIL;
252 	} else {
253 		vcpu->arch.pc += 4;
254 	}
255 
256 	kvm_debug("update_pc(): New PC: %#lx\n", vcpu->arch.pc);
257 
258 	return EMULATE_DONE;
259 }
260 
261 /**
262  * kvm_get_badinstr() - Get bad instruction encoding.
263  * @opc:	Guest pointer to faulting instruction.
264  * @vcpu:	KVM VCPU information.
265  *
266  * Gets the instruction encoding of the faulting instruction, using the saved
267  * BadInstr register value if it exists, otherwise falling back to reading guest
268  * memory at @opc.
269  *
270  * Returns:	The instruction encoding of the faulting instruction.
271  */
272 int kvm_get_badinstr(u32 *opc, struct kvm_vcpu *vcpu, u32 *out)
273 {
274 	if (cpu_has_badinstr) {
275 		*out = vcpu->arch.host_cp0_badinstr;
276 		return 0;
277 	} else {
278 		return kvm_get_inst(opc, vcpu, out);
279 	}
280 }
281 
282 /**
283  * kvm_get_badinstrp() - Get bad prior instruction encoding.
284  * @opc:	Guest pointer to prior faulting instruction.
285  * @vcpu:	KVM VCPU information.
286  *
287  * Gets the instruction encoding of the prior faulting instruction (the branch
288  * containing the delay slot which faulted), using the saved BadInstrP register
289  * value if it exists, otherwise falling back to reading guest memory at @opc.
290  *
291  * Returns:	The instruction encoding of the prior faulting instruction.
292  */
293 int kvm_get_badinstrp(u32 *opc, struct kvm_vcpu *vcpu, u32 *out)
294 {
295 	if (cpu_has_badinstrp) {
296 		*out = vcpu->arch.host_cp0_badinstrp;
297 		return 0;
298 	} else {
299 		return kvm_get_inst(opc, vcpu, out);
300 	}
301 }
302 
303 /**
304  * kvm_mips_count_disabled() - Find whether the CP0_Count timer is disabled.
305  * @vcpu:	Virtual CPU.
306  *
307  * Returns:	1 if the CP0_Count timer is disabled by either the guest
308  *		CP0_Cause.DC bit or the count_ctl.DC bit.
309  *		0 otherwise (in which case CP0_Count timer is running).
310  */
311 int kvm_mips_count_disabled(struct kvm_vcpu *vcpu)
312 {
313 	struct mips_coproc *cop0 = vcpu->arch.cop0;
314 
315 	return	(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC) ||
316 		(kvm_read_c0_guest_cause(cop0) & CAUSEF_DC);
317 }
318 
319 /**
320  * kvm_mips_ktime_to_count() - Scale ktime_t to a 32-bit count.
321  *
322  * Caches the dynamic nanosecond bias in vcpu->arch.count_dyn_bias.
323  *
324  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
325  */
326 static u32 kvm_mips_ktime_to_count(struct kvm_vcpu *vcpu, ktime_t now)
327 {
328 	s64 now_ns, periods;
329 	u64 delta;
330 
331 	now_ns = ktime_to_ns(now);
332 	delta = now_ns + vcpu->arch.count_dyn_bias;
333 
334 	if (delta >= vcpu->arch.count_period) {
335 		/* If delta is out of safe range the bias needs adjusting */
336 		periods = div64_s64(now_ns, vcpu->arch.count_period);
337 		vcpu->arch.count_dyn_bias = -periods * vcpu->arch.count_period;
338 		/* Recalculate delta with new bias */
339 		delta = now_ns + vcpu->arch.count_dyn_bias;
340 	}
341 
342 	/*
343 	 * We've ensured that:
344 	 *   delta < count_period
345 	 *
346 	 * Therefore the intermediate delta*count_hz will never overflow since
347 	 * at the boundary condition:
348 	 *   delta = count_period
349 	 *   delta = NSEC_PER_SEC * 2^32 / count_hz
350 	 *   delta * count_hz = NSEC_PER_SEC * 2^32
351 	 */
352 	return div_u64(delta * vcpu->arch.count_hz, NSEC_PER_SEC);
353 }
354 
355 /**
356  * kvm_mips_count_time() - Get effective current time.
357  * @vcpu:	Virtual CPU.
358  *
359  * Get effective monotonic ktime. This is usually a straightforward ktime_get(),
360  * except when the master disable bit is set in count_ctl, in which case it is
361  * count_resume, i.e. the time that the count was disabled.
362  *
363  * Returns:	Effective monotonic ktime for CP0_Count.
364  */
365 static inline ktime_t kvm_mips_count_time(struct kvm_vcpu *vcpu)
366 {
367 	if (unlikely(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC))
368 		return vcpu->arch.count_resume;
369 
370 	return ktime_get();
371 }
372 
373 /**
374  * kvm_mips_read_count_running() - Read the current count value as if running.
375  * @vcpu:	Virtual CPU.
376  * @now:	Kernel time to read CP0_Count at.
377  *
378  * Returns the current guest CP0_Count register at time @now and handles if the
379  * timer interrupt is pending and hasn't been handled yet.
380  *
381  * Returns:	The current value of the guest CP0_Count register.
382  */
383 static u32 kvm_mips_read_count_running(struct kvm_vcpu *vcpu, ktime_t now)
384 {
385 	struct mips_coproc *cop0 = vcpu->arch.cop0;
386 	ktime_t expires, threshold;
387 	u32 count, compare;
388 	int running;
389 
390 	/* Calculate the biased and scaled guest CP0_Count */
391 	count = vcpu->arch.count_bias + kvm_mips_ktime_to_count(vcpu, now);
392 	compare = kvm_read_c0_guest_compare(cop0);
393 
394 	/*
395 	 * Find whether CP0_Count has reached the closest timer interrupt. If
396 	 * not, we shouldn't inject it.
397 	 */
398 	if ((s32)(count - compare) < 0)
399 		return count;
400 
401 	/*
402 	 * The CP0_Count we're going to return has already reached the closest
403 	 * timer interrupt. Quickly check if it really is a new interrupt by
404 	 * looking at whether the interval until the hrtimer expiry time is
405 	 * less than 1/4 of the timer period.
406 	 */
407 	expires = hrtimer_get_expires(&vcpu->arch.comparecount_timer);
408 	threshold = ktime_add_ns(now, vcpu->arch.count_period / 4);
409 	if (ktime_before(expires, threshold)) {
410 		/*
411 		 * Cancel it while we handle it so there's no chance of
412 		 * interference with the timeout handler.
413 		 */
414 		running = hrtimer_cancel(&vcpu->arch.comparecount_timer);
415 
416 		/* Nothing should be waiting on the timeout */
417 		kvm_mips_callbacks->queue_timer_int(vcpu);
418 
419 		/*
420 		 * Restart the timer if it was running based on the expiry time
421 		 * we read, so that we don't push it back 2 periods.
422 		 */
423 		if (running) {
424 			expires = ktime_add_ns(expires,
425 					       vcpu->arch.count_period);
426 			hrtimer_start(&vcpu->arch.comparecount_timer, expires,
427 				      HRTIMER_MODE_ABS);
428 		}
429 	}
430 
431 	return count;
432 }
433 
434 /**
435  * kvm_mips_read_count() - Read the current count value.
436  * @vcpu:	Virtual CPU.
437  *
438  * Read the current guest CP0_Count value, taking into account whether the timer
439  * is stopped.
440  *
441  * Returns:	The current guest CP0_Count value.
442  */
443 u32 kvm_mips_read_count(struct kvm_vcpu *vcpu)
444 {
445 	struct mips_coproc *cop0 = vcpu->arch.cop0;
446 
447 	/* If count disabled just read static copy of count */
448 	if (kvm_mips_count_disabled(vcpu))
449 		return kvm_read_c0_guest_count(cop0);
450 
451 	return kvm_mips_read_count_running(vcpu, ktime_get());
452 }
453 
454 /**
455  * kvm_mips_freeze_hrtimer() - Safely stop the hrtimer.
456  * @vcpu:	Virtual CPU.
457  * @count:	Output pointer for CP0_Count value at point of freeze.
458  *
459  * Freeze the hrtimer safely and return both the ktime and the CP0_Count value
460  * at the point it was frozen. It is guaranteed that any pending interrupts at
461  * the point it was frozen are handled, and none after that point.
462  *
463  * This is useful where the time/CP0_Count is needed in the calculation of the
464  * new parameters.
465  *
466  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
467  *
468  * Returns:	The ktime at the point of freeze.
469  */
470 ktime_t kvm_mips_freeze_hrtimer(struct kvm_vcpu *vcpu, u32 *count)
471 {
472 	ktime_t now;
473 
474 	/* stop hrtimer before finding time */
475 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
476 	now = ktime_get();
477 
478 	/* find count at this point and handle pending hrtimer */
479 	*count = kvm_mips_read_count_running(vcpu, now);
480 
481 	return now;
482 }
483 
484 /**
485  * kvm_mips_resume_hrtimer() - Resume hrtimer, updating expiry.
486  * @vcpu:	Virtual CPU.
487  * @now:	ktime at point of resume.
488  * @count:	CP0_Count at point of resume.
489  *
490  * Resumes the timer and updates the timer expiry based on @now and @count.
491  * This can be used in conjunction with kvm_mips_freeze_timer() when timer
492  * parameters need to be changed.
493  *
494  * It is guaranteed that a timer interrupt immediately after resume will be
495  * handled, but not if CP_Compare is exactly at @count. That case is already
496  * handled by kvm_mips_freeze_timer().
497  *
498  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
499  */
500 static void kvm_mips_resume_hrtimer(struct kvm_vcpu *vcpu,
501 				    ktime_t now, u32 count)
502 {
503 	struct mips_coproc *cop0 = vcpu->arch.cop0;
504 	u32 compare;
505 	u64 delta;
506 	ktime_t expire;
507 
508 	/* Calculate timeout (wrap 0 to 2^32) */
509 	compare = kvm_read_c0_guest_compare(cop0);
510 	delta = (u64)(u32)(compare - count - 1) + 1;
511 	delta = div_u64(delta * NSEC_PER_SEC, vcpu->arch.count_hz);
512 	expire = ktime_add_ns(now, delta);
513 
514 	/* Update hrtimer to use new timeout */
515 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
516 	hrtimer_start(&vcpu->arch.comparecount_timer, expire, HRTIMER_MODE_ABS);
517 }
518 
519 /**
520  * kvm_mips_restore_hrtimer() - Restore hrtimer after a gap, updating expiry.
521  * @vcpu:	Virtual CPU.
522  * @before:	Time before Count was saved, lower bound of drift calculation.
523  * @count:	CP0_Count at point of restore.
524  * @min_drift:	Minimum amount of drift permitted before correction.
525  *		Must be <= 0.
526  *
527  * Restores the timer from a particular @count, accounting for drift. This can
528  * be used in conjunction with kvm_mips_freeze_timer() when a hardware timer is
529  * to be used for a period of time, but the exact ktime corresponding to the
530  * final Count that must be restored is not known.
531  *
532  * It is gauranteed that a timer interrupt immediately after restore will be
533  * handled, but not if CP0_Compare is exactly at @count. That case should
534  * already be handled when the hardware timer state is saved.
535  *
536  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is not
537  * stopped).
538  *
539  * Returns:	Amount of correction to count_bias due to drift.
540  */
541 int kvm_mips_restore_hrtimer(struct kvm_vcpu *vcpu, ktime_t before,
542 			     u32 count, int min_drift)
543 {
544 	ktime_t now, count_time;
545 	u32 now_count, before_count;
546 	u64 delta;
547 	int drift, ret = 0;
548 
549 	/* Calculate expected count at before */
550 	before_count = vcpu->arch.count_bias +
551 			kvm_mips_ktime_to_count(vcpu, before);
552 
553 	/*
554 	 * Detect significantly negative drift, where count is lower than
555 	 * expected. Some negative drift is expected when hardware counter is
556 	 * set after kvm_mips_freeze_timer(), and it is harmless to allow the
557 	 * time to jump forwards a little, within reason. If the drift is too
558 	 * significant, adjust the bias to avoid a big Guest.CP0_Count jump.
559 	 */
560 	drift = count - before_count;
561 	if (drift < min_drift) {
562 		count_time = before;
563 		vcpu->arch.count_bias += drift;
564 		ret = drift;
565 		goto resume;
566 	}
567 
568 	/* Calculate expected count right now */
569 	now = ktime_get();
570 	now_count = vcpu->arch.count_bias + kvm_mips_ktime_to_count(vcpu, now);
571 
572 	/*
573 	 * Detect positive drift, where count is higher than expected, and
574 	 * adjust the bias to avoid guest time going backwards.
575 	 */
576 	drift = count - now_count;
577 	if (drift > 0) {
578 		count_time = now;
579 		vcpu->arch.count_bias += drift;
580 		ret = drift;
581 		goto resume;
582 	}
583 
584 	/* Subtract nanosecond delta to find ktime when count was read */
585 	delta = (u64)(u32)(now_count - count);
586 	delta = div_u64(delta * NSEC_PER_SEC, vcpu->arch.count_hz);
587 	count_time = ktime_sub_ns(now, delta);
588 
589 resume:
590 	/* Resume using the calculated ktime */
591 	kvm_mips_resume_hrtimer(vcpu, count_time, count);
592 	return ret;
593 }
594 
595 /**
596  * kvm_mips_write_count() - Modify the count and update timer.
597  * @vcpu:	Virtual CPU.
598  * @count:	Guest CP0_Count value to set.
599  *
600  * Sets the CP0_Count value and updates the timer accordingly.
601  */
602 void kvm_mips_write_count(struct kvm_vcpu *vcpu, u32 count)
603 {
604 	struct mips_coproc *cop0 = vcpu->arch.cop0;
605 	ktime_t now;
606 
607 	/* Calculate bias */
608 	now = kvm_mips_count_time(vcpu);
609 	vcpu->arch.count_bias = count - kvm_mips_ktime_to_count(vcpu, now);
610 
611 	if (kvm_mips_count_disabled(vcpu))
612 		/* The timer's disabled, adjust the static count */
613 		kvm_write_c0_guest_count(cop0, count);
614 	else
615 		/* Update timeout */
616 		kvm_mips_resume_hrtimer(vcpu, now, count);
617 }
618 
619 /**
620  * kvm_mips_init_count() - Initialise timer.
621  * @vcpu:	Virtual CPU.
622  * @count_hz:	Frequency of timer.
623  *
624  * Initialise the timer to the specified frequency, zero it, and set it going if
625  * it's enabled.
626  */
627 void kvm_mips_init_count(struct kvm_vcpu *vcpu, unsigned long count_hz)
628 {
629 	vcpu->arch.count_hz = count_hz;
630 	vcpu->arch.count_period = div_u64((u64)NSEC_PER_SEC << 32, count_hz);
631 	vcpu->arch.count_dyn_bias = 0;
632 
633 	/* Starting at 0 */
634 	kvm_mips_write_count(vcpu, 0);
635 }
636 
637 /**
638  * kvm_mips_set_count_hz() - Update the frequency of the timer.
639  * @vcpu:	Virtual CPU.
640  * @count_hz:	Frequency of CP0_Count timer in Hz.
641  *
642  * Change the frequency of the CP0_Count timer. This is done atomically so that
643  * CP0_Count is continuous and no timer interrupt is lost.
644  *
645  * Returns:	-EINVAL if @count_hz is out of range.
646  *		0 on success.
647  */
648 int kvm_mips_set_count_hz(struct kvm_vcpu *vcpu, s64 count_hz)
649 {
650 	struct mips_coproc *cop0 = vcpu->arch.cop0;
651 	int dc;
652 	ktime_t now;
653 	u32 count;
654 
655 	/* ensure the frequency is in a sensible range... */
656 	if (count_hz <= 0 || count_hz > NSEC_PER_SEC)
657 		return -EINVAL;
658 	/* ... and has actually changed */
659 	if (vcpu->arch.count_hz == count_hz)
660 		return 0;
661 
662 	/* Safely freeze timer so we can keep it continuous */
663 	dc = kvm_mips_count_disabled(vcpu);
664 	if (dc) {
665 		now = kvm_mips_count_time(vcpu);
666 		count = kvm_read_c0_guest_count(cop0);
667 	} else {
668 		now = kvm_mips_freeze_hrtimer(vcpu, &count);
669 	}
670 
671 	/* Update the frequency */
672 	vcpu->arch.count_hz = count_hz;
673 	vcpu->arch.count_period = div_u64((u64)NSEC_PER_SEC << 32, count_hz);
674 	vcpu->arch.count_dyn_bias = 0;
675 
676 	/* Calculate adjusted bias so dynamic count is unchanged */
677 	vcpu->arch.count_bias = count - kvm_mips_ktime_to_count(vcpu, now);
678 
679 	/* Update and resume hrtimer */
680 	if (!dc)
681 		kvm_mips_resume_hrtimer(vcpu, now, count);
682 	return 0;
683 }
684 
685 /**
686  * kvm_mips_write_compare() - Modify compare and update timer.
687  * @vcpu:	Virtual CPU.
688  * @compare:	New CP0_Compare value.
689  * @ack:	Whether to acknowledge timer interrupt.
690  *
691  * Update CP0_Compare to a new value and update the timeout.
692  * If @ack, atomically acknowledge any pending timer interrupt, otherwise ensure
693  * any pending timer interrupt is preserved.
694  */
695 void kvm_mips_write_compare(struct kvm_vcpu *vcpu, u32 compare, bool ack)
696 {
697 	struct mips_coproc *cop0 = vcpu->arch.cop0;
698 	int dc;
699 	u32 old_compare = kvm_read_c0_guest_compare(cop0);
700 	s32 delta = compare - old_compare;
701 	u32 cause;
702 	ktime_t now = ktime_set(0, 0); /* silence bogus GCC warning */
703 	u32 count;
704 
705 	/* if unchanged, must just be an ack */
706 	if (old_compare == compare) {
707 		if (!ack)
708 			return;
709 		kvm_mips_callbacks->dequeue_timer_int(vcpu);
710 		kvm_write_c0_guest_compare(cop0, compare);
711 		return;
712 	}
713 
714 	/*
715 	 * If guest CP0_Compare moves forward, CP0_GTOffset should be adjusted
716 	 * too to prevent guest CP0_Count hitting guest CP0_Compare.
717 	 *
718 	 * The new GTOffset corresponds to the new value of CP0_Compare, and is
719 	 * set prior to it being written into the guest context. We disable
720 	 * preemption until the new value is written to prevent restore of a
721 	 * GTOffset corresponding to the old CP0_Compare value.
722 	 */
723 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ) && delta > 0) {
724 		preempt_disable();
725 		write_c0_gtoffset(compare - read_c0_count());
726 		back_to_back_c0_hazard();
727 	}
728 
729 	/* freeze_hrtimer() takes care of timer interrupts <= count */
730 	dc = kvm_mips_count_disabled(vcpu);
731 	if (!dc)
732 		now = kvm_mips_freeze_hrtimer(vcpu, &count);
733 
734 	if (ack)
735 		kvm_mips_callbacks->dequeue_timer_int(vcpu);
736 	else if (IS_ENABLED(CONFIG_KVM_MIPS_VZ))
737 		/*
738 		 * With VZ, writing CP0_Compare acks (clears) CP0_Cause.TI, so
739 		 * preserve guest CP0_Cause.TI if we don't want to ack it.
740 		 */
741 		cause = kvm_read_c0_guest_cause(cop0);
742 
743 	kvm_write_c0_guest_compare(cop0, compare);
744 
745 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
746 		if (delta > 0)
747 			preempt_enable();
748 
749 		back_to_back_c0_hazard();
750 
751 		if (!ack && cause & CAUSEF_TI)
752 			kvm_write_c0_guest_cause(cop0, cause);
753 	}
754 
755 	/* resume_hrtimer() takes care of timer interrupts > count */
756 	if (!dc)
757 		kvm_mips_resume_hrtimer(vcpu, now, count);
758 
759 	/*
760 	 * If guest CP0_Compare is moving backward, we delay CP0_GTOffset change
761 	 * until after the new CP0_Compare is written, otherwise new guest
762 	 * CP0_Count could hit new guest CP0_Compare.
763 	 */
764 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ) && delta <= 0)
765 		write_c0_gtoffset(compare - read_c0_count());
766 }
767 
768 /**
769  * kvm_mips_count_disable() - Disable count.
770  * @vcpu:	Virtual CPU.
771  *
772  * Disable the CP0_Count timer. A timer interrupt on or before the final stop
773  * time will be handled but not after.
774  *
775  * Assumes CP0_Count was previously enabled but now Guest.CP0_Cause.DC or
776  * count_ctl.DC has been set (count disabled).
777  *
778  * Returns:	The time that the timer was stopped.
779  */
780 static ktime_t kvm_mips_count_disable(struct kvm_vcpu *vcpu)
781 {
782 	struct mips_coproc *cop0 = vcpu->arch.cop0;
783 	u32 count;
784 	ktime_t now;
785 
786 	/* Stop hrtimer */
787 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
788 
789 	/* Set the static count from the dynamic count, handling pending TI */
790 	now = ktime_get();
791 	count = kvm_mips_read_count_running(vcpu, now);
792 	kvm_write_c0_guest_count(cop0, count);
793 
794 	return now;
795 }
796 
797 /**
798  * kvm_mips_count_disable_cause() - Disable count using CP0_Cause.DC.
799  * @vcpu:	Virtual CPU.
800  *
801  * Disable the CP0_Count timer and set CP0_Cause.DC. A timer interrupt on or
802  * before the final stop time will be handled if the timer isn't disabled by
803  * count_ctl.DC, but not after.
804  *
805  * Assumes CP0_Cause.DC is clear (count enabled).
806  */
807 void kvm_mips_count_disable_cause(struct kvm_vcpu *vcpu)
808 {
809 	struct mips_coproc *cop0 = vcpu->arch.cop0;
810 
811 	kvm_set_c0_guest_cause(cop0, CAUSEF_DC);
812 	if (!(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC))
813 		kvm_mips_count_disable(vcpu);
814 }
815 
816 /**
817  * kvm_mips_count_enable_cause() - Enable count using CP0_Cause.DC.
818  * @vcpu:	Virtual CPU.
819  *
820  * Enable the CP0_Count timer and clear CP0_Cause.DC. A timer interrupt after
821  * the start time will be handled if the timer isn't disabled by count_ctl.DC,
822  * potentially before even returning, so the caller should be careful with
823  * ordering of CP0_Cause modifications so as not to lose it.
824  *
825  * Assumes CP0_Cause.DC is set (count disabled).
826  */
827 void kvm_mips_count_enable_cause(struct kvm_vcpu *vcpu)
828 {
829 	struct mips_coproc *cop0 = vcpu->arch.cop0;
830 	u32 count;
831 
832 	kvm_clear_c0_guest_cause(cop0, CAUSEF_DC);
833 
834 	/*
835 	 * Set the dynamic count to match the static count.
836 	 * This starts the hrtimer if count_ctl.DC allows it.
837 	 * Otherwise it conveniently updates the biases.
838 	 */
839 	count = kvm_read_c0_guest_count(cop0);
840 	kvm_mips_write_count(vcpu, count);
841 }
842 
843 /**
844  * kvm_mips_set_count_ctl() - Update the count control KVM register.
845  * @vcpu:	Virtual CPU.
846  * @count_ctl:	Count control register new value.
847  *
848  * Set the count control KVM register. The timer is updated accordingly.
849  *
850  * Returns:	-EINVAL if reserved bits are set.
851  *		0 on success.
852  */
853 int kvm_mips_set_count_ctl(struct kvm_vcpu *vcpu, s64 count_ctl)
854 {
855 	struct mips_coproc *cop0 = vcpu->arch.cop0;
856 	s64 changed = count_ctl ^ vcpu->arch.count_ctl;
857 	s64 delta;
858 	ktime_t expire, now;
859 	u32 count, compare;
860 
861 	/* Only allow defined bits to be changed */
862 	if (changed & ~(s64)(KVM_REG_MIPS_COUNT_CTL_DC))
863 		return -EINVAL;
864 
865 	/* Apply new value */
866 	vcpu->arch.count_ctl = count_ctl;
867 
868 	/* Master CP0_Count disable */
869 	if (changed & KVM_REG_MIPS_COUNT_CTL_DC) {
870 		/* Is CP0_Cause.DC already disabling CP0_Count? */
871 		if (kvm_read_c0_guest_cause(cop0) & CAUSEF_DC) {
872 			if (count_ctl & KVM_REG_MIPS_COUNT_CTL_DC)
873 				/* Just record the current time */
874 				vcpu->arch.count_resume = ktime_get();
875 		} else if (count_ctl & KVM_REG_MIPS_COUNT_CTL_DC) {
876 			/* disable timer and record current time */
877 			vcpu->arch.count_resume = kvm_mips_count_disable(vcpu);
878 		} else {
879 			/*
880 			 * Calculate timeout relative to static count at resume
881 			 * time (wrap 0 to 2^32).
882 			 */
883 			count = kvm_read_c0_guest_count(cop0);
884 			compare = kvm_read_c0_guest_compare(cop0);
885 			delta = (u64)(u32)(compare - count - 1) + 1;
886 			delta = div_u64(delta * NSEC_PER_SEC,
887 					vcpu->arch.count_hz);
888 			expire = ktime_add_ns(vcpu->arch.count_resume, delta);
889 
890 			/* Handle pending interrupt */
891 			now = ktime_get();
892 			if (ktime_compare(now, expire) >= 0)
893 				/* Nothing should be waiting on the timeout */
894 				kvm_mips_callbacks->queue_timer_int(vcpu);
895 
896 			/* Resume hrtimer without changing bias */
897 			count = kvm_mips_read_count_running(vcpu, now);
898 			kvm_mips_resume_hrtimer(vcpu, now, count);
899 		}
900 	}
901 
902 	return 0;
903 }
904 
905 /**
906  * kvm_mips_set_count_resume() - Update the count resume KVM register.
907  * @vcpu:		Virtual CPU.
908  * @count_resume:	Count resume register new value.
909  *
910  * Set the count resume KVM register.
911  *
912  * Returns:	-EINVAL if out of valid range (0..now).
913  *		0 on success.
914  */
915 int kvm_mips_set_count_resume(struct kvm_vcpu *vcpu, s64 count_resume)
916 {
917 	/*
918 	 * It doesn't make sense for the resume time to be in the future, as it
919 	 * would be possible for the next interrupt to be more than a full
920 	 * period in the future.
921 	 */
922 	if (count_resume < 0 || count_resume > ktime_to_ns(ktime_get()))
923 		return -EINVAL;
924 
925 	vcpu->arch.count_resume = ns_to_ktime(count_resume);
926 	return 0;
927 }
928 
929 /**
930  * kvm_mips_count_timeout() - Push timer forward on timeout.
931  * @vcpu:	Virtual CPU.
932  *
933  * Handle an hrtimer event by push the hrtimer forward a period.
934  *
935  * Returns:	The hrtimer_restart value to return to the hrtimer subsystem.
936  */
937 enum hrtimer_restart kvm_mips_count_timeout(struct kvm_vcpu *vcpu)
938 {
939 	/* Add the Count period to the current expiry time */
940 	hrtimer_add_expires_ns(&vcpu->arch.comparecount_timer,
941 			       vcpu->arch.count_period);
942 	return HRTIMER_RESTART;
943 }
944 
945 enum emulation_result kvm_mips_emul_eret(struct kvm_vcpu *vcpu)
946 {
947 	struct mips_coproc *cop0 = vcpu->arch.cop0;
948 	enum emulation_result er = EMULATE_DONE;
949 
950 	if (kvm_read_c0_guest_status(cop0) & ST0_ERL) {
951 		kvm_clear_c0_guest_status(cop0, ST0_ERL);
952 		vcpu->arch.pc = kvm_read_c0_guest_errorepc(cop0);
953 	} else if (kvm_read_c0_guest_status(cop0) & ST0_EXL) {
954 		kvm_debug("[%#lx] ERET to %#lx\n", vcpu->arch.pc,
955 			  kvm_read_c0_guest_epc(cop0));
956 		kvm_clear_c0_guest_status(cop0, ST0_EXL);
957 		vcpu->arch.pc = kvm_read_c0_guest_epc(cop0);
958 
959 	} else {
960 		kvm_err("[%#lx] ERET when MIPS_SR_EXL|MIPS_SR_ERL == 0\n",
961 			vcpu->arch.pc);
962 		er = EMULATE_FAIL;
963 	}
964 
965 	return er;
966 }
967 
968 enum emulation_result kvm_mips_emul_wait(struct kvm_vcpu *vcpu)
969 {
970 	kvm_debug("[%#lx] !!!WAIT!!! (%#lx)\n", vcpu->arch.pc,
971 		  vcpu->arch.pending_exceptions);
972 
973 	++vcpu->stat.wait_exits;
974 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_WAIT);
975 	if (!vcpu->arch.pending_exceptions) {
976 		kvm_vz_lose_htimer(vcpu);
977 		vcpu->arch.wait = 1;
978 		kvm_vcpu_block(vcpu);
979 
980 		/*
981 		 * We we are runnable, then definitely go off to user space to
982 		 * check if any I/O interrupts are pending.
983 		 */
984 		if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
985 			kvm_clear_request(KVM_REQ_UNHALT, vcpu);
986 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
987 		}
988 	}
989 
990 	return EMULATE_DONE;
991 }
992 
993 static void kvm_mips_change_entryhi(struct kvm_vcpu *vcpu,
994 				    unsigned long entryhi)
995 {
996 	struct mips_coproc *cop0 = vcpu->arch.cop0;
997 	struct mm_struct *kern_mm = &vcpu->arch.guest_kernel_mm;
998 	int cpu, i;
999 	u32 nasid = entryhi & KVM_ENTRYHI_ASID;
1000 
1001 	if (((kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID) != nasid)) {
1002 		trace_kvm_asid_change(vcpu, kvm_read_c0_guest_entryhi(cop0) &
1003 				      KVM_ENTRYHI_ASID, nasid);
1004 
1005 		/*
1006 		 * Flush entries from the GVA page tables.
1007 		 * Guest user page table will get flushed lazily on re-entry to
1008 		 * guest user if the guest ASID actually changes.
1009 		 */
1010 		kvm_mips_flush_gva_pt(kern_mm->pgd, KMF_KERN);
1011 
1012 		/*
1013 		 * Regenerate/invalidate kernel MMU context.
1014 		 * The user MMU context will be regenerated lazily on re-entry
1015 		 * to guest user if the guest ASID actually changes.
1016 		 */
1017 		preempt_disable();
1018 		cpu = smp_processor_id();
1019 		get_new_mmu_context(kern_mm, cpu);
1020 		for_each_possible_cpu(i)
1021 			if (i != cpu)
1022 				cpu_context(i, kern_mm) = 0;
1023 		preempt_enable();
1024 	}
1025 	kvm_write_c0_guest_entryhi(cop0, entryhi);
1026 }
1027 
1028 enum emulation_result kvm_mips_emul_tlbr(struct kvm_vcpu *vcpu)
1029 {
1030 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1031 	struct kvm_mips_tlb *tlb;
1032 	unsigned long pc = vcpu->arch.pc;
1033 	int index;
1034 
1035 	index = kvm_read_c0_guest_index(cop0);
1036 	if (index < 0 || index >= KVM_MIPS_GUEST_TLB_SIZE) {
1037 		/* UNDEFINED */
1038 		kvm_debug("[%#lx] TLBR Index %#x out of range\n", pc, index);
1039 		index &= KVM_MIPS_GUEST_TLB_SIZE - 1;
1040 	}
1041 
1042 	tlb = &vcpu->arch.guest_tlb[index];
1043 	kvm_write_c0_guest_pagemask(cop0, tlb->tlb_mask);
1044 	kvm_write_c0_guest_entrylo0(cop0, tlb->tlb_lo[0]);
1045 	kvm_write_c0_guest_entrylo1(cop0, tlb->tlb_lo[1]);
1046 	kvm_mips_change_entryhi(vcpu, tlb->tlb_hi);
1047 
1048 	return EMULATE_DONE;
1049 }
1050 
1051 /**
1052  * kvm_mips_invalidate_guest_tlb() - Indicates a change in guest MMU map.
1053  * @vcpu:	VCPU with changed mappings.
1054  * @tlb:	TLB entry being removed.
1055  *
1056  * This is called to indicate a single change in guest MMU mappings, so that we
1057  * can arrange TLB flushes on this and other CPUs.
1058  */
1059 static void kvm_mips_invalidate_guest_tlb(struct kvm_vcpu *vcpu,
1060 					  struct kvm_mips_tlb *tlb)
1061 {
1062 	struct mm_struct *kern_mm = &vcpu->arch.guest_kernel_mm;
1063 	struct mm_struct *user_mm = &vcpu->arch.guest_user_mm;
1064 	int cpu, i;
1065 	bool user;
1066 
1067 	/* No need to flush for entries which are already invalid */
1068 	if (!((tlb->tlb_lo[0] | tlb->tlb_lo[1]) & ENTRYLO_V))
1069 		return;
1070 	/* Don't touch host kernel page tables or TLB mappings */
1071 	if ((unsigned long)tlb->tlb_hi > 0x7fffffff)
1072 		return;
1073 	/* User address space doesn't need flushing for KSeg2/3 changes */
1074 	user = tlb->tlb_hi < KVM_GUEST_KSEG0;
1075 
1076 	preempt_disable();
1077 
1078 	/* Invalidate page table entries */
1079 	kvm_trap_emul_invalidate_gva(vcpu, tlb->tlb_hi & VPN2_MASK, user);
1080 
1081 	/*
1082 	 * Probe the shadow host TLB for the entry being overwritten, if one
1083 	 * matches, invalidate it
1084 	 */
1085 	kvm_mips_host_tlb_inv(vcpu, tlb->tlb_hi, user, true);
1086 
1087 	/* Invalidate the whole ASID on other CPUs */
1088 	cpu = smp_processor_id();
1089 	for_each_possible_cpu(i) {
1090 		if (i == cpu)
1091 			continue;
1092 		if (user)
1093 			cpu_context(i, user_mm) = 0;
1094 		cpu_context(i, kern_mm) = 0;
1095 	}
1096 
1097 	preempt_enable();
1098 }
1099 
1100 /* Write Guest TLB Entry @ Index */
1101 enum emulation_result kvm_mips_emul_tlbwi(struct kvm_vcpu *vcpu)
1102 {
1103 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1104 	int index = kvm_read_c0_guest_index(cop0);
1105 	struct kvm_mips_tlb *tlb = NULL;
1106 	unsigned long pc = vcpu->arch.pc;
1107 
1108 	if (index < 0 || index >= KVM_MIPS_GUEST_TLB_SIZE) {
1109 		kvm_debug("%s: illegal index: %d\n", __func__, index);
1110 		kvm_debug("[%#lx] COP0_TLBWI [%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx, mask: %#lx)\n",
1111 			  pc, index, kvm_read_c0_guest_entryhi(cop0),
1112 			  kvm_read_c0_guest_entrylo0(cop0),
1113 			  kvm_read_c0_guest_entrylo1(cop0),
1114 			  kvm_read_c0_guest_pagemask(cop0));
1115 		index = (index & ~0x80000000) % KVM_MIPS_GUEST_TLB_SIZE;
1116 	}
1117 
1118 	tlb = &vcpu->arch.guest_tlb[index];
1119 
1120 	kvm_mips_invalidate_guest_tlb(vcpu, tlb);
1121 
1122 	tlb->tlb_mask = kvm_read_c0_guest_pagemask(cop0);
1123 	tlb->tlb_hi = kvm_read_c0_guest_entryhi(cop0);
1124 	tlb->tlb_lo[0] = kvm_read_c0_guest_entrylo0(cop0);
1125 	tlb->tlb_lo[1] = kvm_read_c0_guest_entrylo1(cop0);
1126 
1127 	kvm_debug("[%#lx] COP0_TLBWI [%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx, mask: %#lx)\n",
1128 		  pc, index, kvm_read_c0_guest_entryhi(cop0),
1129 		  kvm_read_c0_guest_entrylo0(cop0),
1130 		  kvm_read_c0_guest_entrylo1(cop0),
1131 		  kvm_read_c0_guest_pagemask(cop0));
1132 
1133 	return EMULATE_DONE;
1134 }
1135 
1136 /* Write Guest TLB Entry @ Random Index */
1137 enum emulation_result kvm_mips_emul_tlbwr(struct kvm_vcpu *vcpu)
1138 {
1139 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1140 	struct kvm_mips_tlb *tlb = NULL;
1141 	unsigned long pc = vcpu->arch.pc;
1142 	int index;
1143 
1144 	get_random_bytes(&index, sizeof(index));
1145 	index &= (KVM_MIPS_GUEST_TLB_SIZE - 1);
1146 
1147 	tlb = &vcpu->arch.guest_tlb[index];
1148 
1149 	kvm_mips_invalidate_guest_tlb(vcpu, tlb);
1150 
1151 	tlb->tlb_mask = kvm_read_c0_guest_pagemask(cop0);
1152 	tlb->tlb_hi = kvm_read_c0_guest_entryhi(cop0);
1153 	tlb->tlb_lo[0] = kvm_read_c0_guest_entrylo0(cop0);
1154 	tlb->tlb_lo[1] = kvm_read_c0_guest_entrylo1(cop0);
1155 
1156 	kvm_debug("[%#lx] COP0_TLBWR[%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx)\n",
1157 		  pc, index, kvm_read_c0_guest_entryhi(cop0),
1158 		  kvm_read_c0_guest_entrylo0(cop0),
1159 		  kvm_read_c0_guest_entrylo1(cop0));
1160 
1161 	return EMULATE_DONE;
1162 }
1163 
1164 enum emulation_result kvm_mips_emul_tlbp(struct kvm_vcpu *vcpu)
1165 {
1166 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1167 	long entryhi = kvm_read_c0_guest_entryhi(cop0);
1168 	unsigned long pc = vcpu->arch.pc;
1169 	int index = -1;
1170 
1171 	index = kvm_mips_guest_tlb_lookup(vcpu, entryhi);
1172 
1173 	kvm_write_c0_guest_index(cop0, index);
1174 
1175 	kvm_debug("[%#lx] COP0_TLBP (entryhi: %#lx), index: %d\n", pc, entryhi,
1176 		  index);
1177 
1178 	return EMULATE_DONE;
1179 }
1180 
1181 /**
1182  * kvm_mips_config1_wrmask() - Find mask of writable bits in guest Config1
1183  * @vcpu:	Virtual CPU.
1184  *
1185  * Finds the mask of bits which are writable in the guest's Config1 CP0
1186  * register, by userland (currently read-only to the guest).
1187  */
1188 unsigned int kvm_mips_config1_wrmask(struct kvm_vcpu *vcpu)
1189 {
1190 	unsigned int mask = 0;
1191 
1192 	/* Permit FPU to be present if FPU is supported */
1193 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch))
1194 		mask |= MIPS_CONF1_FP;
1195 
1196 	return mask;
1197 }
1198 
1199 /**
1200  * kvm_mips_config3_wrmask() - Find mask of writable bits in guest Config3
1201  * @vcpu:	Virtual CPU.
1202  *
1203  * Finds the mask of bits which are writable in the guest's Config3 CP0
1204  * register, by userland (currently read-only to the guest).
1205  */
1206 unsigned int kvm_mips_config3_wrmask(struct kvm_vcpu *vcpu)
1207 {
1208 	/* Config4 and ULRI are optional */
1209 	unsigned int mask = MIPS_CONF_M | MIPS_CONF3_ULRI;
1210 
1211 	/* Permit MSA to be present if MSA is supported */
1212 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
1213 		mask |= MIPS_CONF3_MSA;
1214 
1215 	return mask;
1216 }
1217 
1218 /**
1219  * kvm_mips_config4_wrmask() - Find mask of writable bits in guest Config4
1220  * @vcpu:	Virtual CPU.
1221  *
1222  * Finds the mask of bits which are writable in the guest's Config4 CP0
1223  * register, by userland (currently read-only to the guest).
1224  */
1225 unsigned int kvm_mips_config4_wrmask(struct kvm_vcpu *vcpu)
1226 {
1227 	/* Config5 is optional */
1228 	unsigned int mask = MIPS_CONF_M;
1229 
1230 	/* KScrExist */
1231 	mask |= 0xfc << MIPS_CONF4_KSCREXIST_SHIFT;
1232 
1233 	return mask;
1234 }
1235 
1236 /**
1237  * kvm_mips_config5_wrmask() - Find mask of writable bits in guest Config5
1238  * @vcpu:	Virtual CPU.
1239  *
1240  * Finds the mask of bits which are writable in the guest's Config5 CP0
1241  * register, by the guest itself.
1242  */
1243 unsigned int kvm_mips_config5_wrmask(struct kvm_vcpu *vcpu)
1244 {
1245 	unsigned int mask = 0;
1246 
1247 	/* Permit MSAEn changes if MSA supported and enabled */
1248 	if (kvm_mips_guest_has_msa(&vcpu->arch))
1249 		mask |= MIPS_CONF5_MSAEN;
1250 
1251 	/*
1252 	 * Permit guest FPU mode changes if FPU is enabled and the relevant
1253 	 * feature exists according to FIR register.
1254 	 */
1255 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1256 		if (cpu_has_fre)
1257 			mask |= MIPS_CONF5_FRE;
1258 		/* We don't support UFR or UFE */
1259 	}
1260 
1261 	return mask;
1262 }
1263 
1264 enum emulation_result kvm_mips_emulate_CP0(union mips_instruction inst,
1265 					   u32 *opc, u32 cause,
1266 					   struct kvm_run *run,
1267 					   struct kvm_vcpu *vcpu)
1268 {
1269 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1270 	enum emulation_result er = EMULATE_DONE;
1271 	u32 rt, rd, sel;
1272 	unsigned long curr_pc;
1273 
1274 	/*
1275 	 * Update PC and hold onto current PC in case there is
1276 	 * an error and we want to rollback the PC
1277 	 */
1278 	curr_pc = vcpu->arch.pc;
1279 	er = update_pc(vcpu, cause);
1280 	if (er == EMULATE_FAIL)
1281 		return er;
1282 
1283 	if (inst.co_format.co) {
1284 		switch (inst.co_format.func) {
1285 		case tlbr_op:	/*  Read indexed TLB entry  */
1286 			er = kvm_mips_emul_tlbr(vcpu);
1287 			break;
1288 		case tlbwi_op:	/*  Write indexed  */
1289 			er = kvm_mips_emul_tlbwi(vcpu);
1290 			break;
1291 		case tlbwr_op:	/*  Write random  */
1292 			er = kvm_mips_emul_tlbwr(vcpu);
1293 			break;
1294 		case tlbp_op:	/* TLB Probe */
1295 			er = kvm_mips_emul_tlbp(vcpu);
1296 			break;
1297 		case rfe_op:
1298 			kvm_err("!!!COP0_RFE!!!\n");
1299 			break;
1300 		case eret_op:
1301 			er = kvm_mips_emul_eret(vcpu);
1302 			goto dont_update_pc;
1303 		case wait_op:
1304 			er = kvm_mips_emul_wait(vcpu);
1305 			break;
1306 		case hypcall_op:
1307 			er = kvm_mips_emul_hypcall(vcpu, inst);
1308 			break;
1309 		}
1310 	} else {
1311 		rt = inst.c0r_format.rt;
1312 		rd = inst.c0r_format.rd;
1313 		sel = inst.c0r_format.sel;
1314 
1315 		switch (inst.c0r_format.rs) {
1316 		case mfc_op:
1317 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
1318 			cop0->stat[rd][sel]++;
1319 #endif
1320 			/* Get reg */
1321 			if ((rd == MIPS_CP0_COUNT) && (sel == 0)) {
1322 				vcpu->arch.gprs[rt] =
1323 				    (s32)kvm_mips_read_count(vcpu);
1324 			} else if ((rd == MIPS_CP0_ERRCTL) && (sel == 0)) {
1325 				vcpu->arch.gprs[rt] = 0x0;
1326 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1327 				kvm_mips_trans_mfc0(inst, opc, vcpu);
1328 #endif
1329 			} else {
1330 				vcpu->arch.gprs[rt] = (s32)cop0->reg[rd][sel];
1331 
1332 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1333 				kvm_mips_trans_mfc0(inst, opc, vcpu);
1334 #endif
1335 			}
1336 
1337 			trace_kvm_hwr(vcpu, KVM_TRACE_MFC0,
1338 				      KVM_TRACE_COP0(rd, sel),
1339 				      vcpu->arch.gprs[rt]);
1340 			break;
1341 
1342 		case dmfc_op:
1343 			vcpu->arch.gprs[rt] = cop0->reg[rd][sel];
1344 
1345 			trace_kvm_hwr(vcpu, KVM_TRACE_DMFC0,
1346 				      KVM_TRACE_COP0(rd, sel),
1347 				      vcpu->arch.gprs[rt]);
1348 			break;
1349 
1350 		case mtc_op:
1351 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
1352 			cop0->stat[rd][sel]++;
1353 #endif
1354 			trace_kvm_hwr(vcpu, KVM_TRACE_MTC0,
1355 				      KVM_TRACE_COP0(rd, sel),
1356 				      vcpu->arch.gprs[rt]);
1357 
1358 			if ((rd == MIPS_CP0_TLB_INDEX)
1359 			    && (vcpu->arch.gprs[rt] >=
1360 				KVM_MIPS_GUEST_TLB_SIZE)) {
1361 				kvm_err("Invalid TLB Index: %ld",
1362 					vcpu->arch.gprs[rt]);
1363 				er = EMULATE_FAIL;
1364 				break;
1365 			}
1366 			if ((rd == MIPS_CP0_PRID) && (sel == 1)) {
1367 				/*
1368 				 * Preserve core number, and keep the exception
1369 				 * base in guest KSeg0.
1370 				 */
1371 				kvm_change_c0_guest_ebase(cop0, 0x1ffff000,
1372 							  vcpu->arch.gprs[rt]);
1373 			} else if (rd == MIPS_CP0_TLB_HI && sel == 0) {
1374 				kvm_mips_change_entryhi(vcpu,
1375 							vcpu->arch.gprs[rt]);
1376 			}
1377 			/* Are we writing to COUNT */
1378 			else if ((rd == MIPS_CP0_COUNT) && (sel == 0)) {
1379 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
1380 				goto done;
1381 			} else if ((rd == MIPS_CP0_COMPARE) && (sel == 0)) {
1382 				/* If we are writing to COMPARE */
1383 				/* Clear pending timer interrupt, if any */
1384 				kvm_mips_write_compare(vcpu,
1385 						       vcpu->arch.gprs[rt],
1386 						       true);
1387 			} else if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1388 				unsigned int old_val, val, change;
1389 
1390 				old_val = kvm_read_c0_guest_status(cop0);
1391 				val = vcpu->arch.gprs[rt];
1392 				change = val ^ old_val;
1393 
1394 				/* Make sure that the NMI bit is never set */
1395 				val &= ~ST0_NMI;
1396 
1397 				/*
1398 				 * Don't allow CU1 or FR to be set unless FPU
1399 				 * capability enabled and exists in guest
1400 				 * configuration.
1401 				 */
1402 				if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1403 					val &= ~(ST0_CU1 | ST0_FR);
1404 
1405 				/*
1406 				 * Also don't allow FR to be set if host doesn't
1407 				 * support it.
1408 				 */
1409 				if (!(current_cpu_data.fpu_id & MIPS_FPIR_F64))
1410 					val &= ~ST0_FR;
1411 
1412 
1413 				/* Handle changes in FPU mode */
1414 				preempt_disable();
1415 
1416 				/*
1417 				 * FPU and Vector register state is made
1418 				 * UNPREDICTABLE by a change of FR, so don't
1419 				 * even bother saving it.
1420 				 */
1421 				if (change & ST0_FR)
1422 					kvm_drop_fpu(vcpu);
1423 
1424 				/*
1425 				 * If MSA state is already live, it is undefined
1426 				 * how it interacts with FR=0 FPU state, and we
1427 				 * don't want to hit reserved instruction
1428 				 * exceptions trying to save the MSA state later
1429 				 * when CU=1 && FR=1, so play it safe and save
1430 				 * it first.
1431 				 */
1432 				if (change & ST0_CU1 && !(val & ST0_FR) &&
1433 				    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1434 					kvm_lose_fpu(vcpu);
1435 
1436 				/*
1437 				 * Propagate CU1 (FPU enable) changes
1438 				 * immediately if the FPU context is already
1439 				 * loaded. When disabling we leave the context
1440 				 * loaded so it can be quickly enabled again in
1441 				 * the near future.
1442 				 */
1443 				if (change & ST0_CU1 &&
1444 				    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1445 					change_c0_status(ST0_CU1, val);
1446 
1447 				preempt_enable();
1448 
1449 				kvm_write_c0_guest_status(cop0, val);
1450 
1451 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1452 				/*
1453 				 * If FPU present, we need CU1/FR bits to take
1454 				 * effect fairly soon.
1455 				 */
1456 				if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1457 					kvm_mips_trans_mtc0(inst, opc, vcpu);
1458 #endif
1459 			} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1460 				unsigned int old_val, val, change, wrmask;
1461 
1462 				old_val = kvm_read_c0_guest_config5(cop0);
1463 				val = vcpu->arch.gprs[rt];
1464 
1465 				/* Only a few bits are writable in Config5 */
1466 				wrmask = kvm_mips_config5_wrmask(vcpu);
1467 				change = (val ^ old_val) & wrmask;
1468 				val = old_val ^ change;
1469 
1470 
1471 				/* Handle changes in FPU/MSA modes */
1472 				preempt_disable();
1473 
1474 				/*
1475 				 * Propagate FRE changes immediately if the FPU
1476 				 * context is already loaded.
1477 				 */
1478 				if (change & MIPS_CONF5_FRE &&
1479 				    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1480 					change_c0_config5(MIPS_CONF5_FRE, val);
1481 
1482 				/*
1483 				 * Propagate MSAEn changes immediately if the
1484 				 * MSA context is already loaded. When disabling
1485 				 * we leave the context loaded so it can be
1486 				 * quickly enabled again in the near future.
1487 				 */
1488 				if (change & MIPS_CONF5_MSAEN &&
1489 				    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1490 					change_c0_config5(MIPS_CONF5_MSAEN,
1491 							  val);
1492 
1493 				preempt_enable();
1494 
1495 				kvm_write_c0_guest_config5(cop0, val);
1496 			} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1497 				u32 old_cause, new_cause;
1498 
1499 				old_cause = kvm_read_c0_guest_cause(cop0);
1500 				new_cause = vcpu->arch.gprs[rt];
1501 				/* Update R/W bits */
1502 				kvm_change_c0_guest_cause(cop0, 0x08800300,
1503 							  new_cause);
1504 				/* DC bit enabling/disabling timer? */
1505 				if ((old_cause ^ new_cause) & CAUSEF_DC) {
1506 					if (new_cause & CAUSEF_DC)
1507 						kvm_mips_count_disable_cause(vcpu);
1508 					else
1509 						kvm_mips_count_enable_cause(vcpu);
1510 				}
1511 			} else if ((rd == MIPS_CP0_HWRENA) && (sel == 0)) {
1512 				u32 mask = MIPS_HWRENA_CPUNUM |
1513 					   MIPS_HWRENA_SYNCISTEP |
1514 					   MIPS_HWRENA_CC |
1515 					   MIPS_HWRENA_CCRES;
1516 
1517 				if (kvm_read_c0_guest_config3(cop0) &
1518 				    MIPS_CONF3_ULRI)
1519 					mask |= MIPS_HWRENA_ULR;
1520 				cop0->reg[rd][sel] = vcpu->arch.gprs[rt] & mask;
1521 			} else {
1522 				cop0->reg[rd][sel] = vcpu->arch.gprs[rt];
1523 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1524 				kvm_mips_trans_mtc0(inst, opc, vcpu);
1525 #endif
1526 			}
1527 			break;
1528 
1529 		case dmtc_op:
1530 			kvm_err("!!!!!!![%#lx]dmtc_op: rt: %d, rd: %d, sel: %d!!!!!!\n",
1531 				vcpu->arch.pc, rt, rd, sel);
1532 			trace_kvm_hwr(vcpu, KVM_TRACE_DMTC0,
1533 				      KVM_TRACE_COP0(rd, sel),
1534 				      vcpu->arch.gprs[rt]);
1535 			er = EMULATE_FAIL;
1536 			break;
1537 
1538 		case mfmc0_op:
1539 #ifdef KVM_MIPS_DEBUG_COP0_COUNTERS
1540 			cop0->stat[MIPS_CP0_STATUS][0]++;
1541 #endif
1542 			if (rt != 0)
1543 				vcpu->arch.gprs[rt] =
1544 				    kvm_read_c0_guest_status(cop0);
1545 			/* EI */
1546 			if (inst.mfmc0_format.sc) {
1547 				kvm_debug("[%#lx] mfmc0_op: EI\n",
1548 					  vcpu->arch.pc);
1549 				kvm_set_c0_guest_status(cop0, ST0_IE);
1550 			} else {
1551 				kvm_debug("[%#lx] mfmc0_op: DI\n",
1552 					  vcpu->arch.pc);
1553 				kvm_clear_c0_guest_status(cop0, ST0_IE);
1554 			}
1555 
1556 			break;
1557 
1558 		case wrpgpr_op:
1559 			{
1560 				u32 css = cop0->reg[MIPS_CP0_STATUS][2] & 0xf;
1561 				u32 pss =
1562 				    (cop0->reg[MIPS_CP0_STATUS][2] >> 6) & 0xf;
1563 				/*
1564 				 * We don't support any shadow register sets, so
1565 				 * SRSCtl[PSS] == SRSCtl[CSS] = 0
1566 				 */
1567 				if (css || pss) {
1568 					er = EMULATE_FAIL;
1569 					break;
1570 				}
1571 				kvm_debug("WRPGPR[%d][%d] = %#lx\n", pss, rd,
1572 					  vcpu->arch.gprs[rt]);
1573 				vcpu->arch.gprs[rd] = vcpu->arch.gprs[rt];
1574 			}
1575 			break;
1576 		default:
1577 			kvm_err("[%#lx]MachEmulateCP0: unsupported COP0, copz: 0x%x\n",
1578 				vcpu->arch.pc, inst.c0r_format.rs);
1579 			er = EMULATE_FAIL;
1580 			break;
1581 		}
1582 	}
1583 
1584 done:
1585 	/* Rollback PC only if emulation was unsuccessful */
1586 	if (er == EMULATE_FAIL)
1587 		vcpu->arch.pc = curr_pc;
1588 
1589 dont_update_pc:
1590 	/*
1591 	 * This is for special instructions whose emulation
1592 	 * updates the PC, so do not overwrite the PC under
1593 	 * any circumstances
1594 	 */
1595 
1596 	return er;
1597 }
1598 
1599 enum emulation_result kvm_mips_emulate_store(union mips_instruction inst,
1600 					     u32 cause,
1601 					     struct kvm_run *run,
1602 					     struct kvm_vcpu *vcpu)
1603 {
1604 	enum emulation_result er;
1605 	u32 rt;
1606 	void *data = run->mmio.data;
1607 	unsigned long curr_pc;
1608 
1609 	/*
1610 	 * Update PC and hold onto current PC in case there is
1611 	 * an error and we want to rollback the PC
1612 	 */
1613 	curr_pc = vcpu->arch.pc;
1614 	er = update_pc(vcpu, cause);
1615 	if (er == EMULATE_FAIL)
1616 		return er;
1617 
1618 	rt = inst.i_format.rt;
1619 
1620 	run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1621 						vcpu->arch.host_cp0_badvaddr);
1622 	if (run->mmio.phys_addr == KVM_INVALID_ADDR)
1623 		goto out_fail;
1624 
1625 	switch (inst.i_format.opcode) {
1626 #if defined(CONFIG_64BIT) && defined(CONFIG_KVM_MIPS_VZ)
1627 	case sd_op:
1628 		run->mmio.len = 8;
1629 		*(u64 *)data = vcpu->arch.gprs[rt];
1630 
1631 		kvm_debug("[%#lx] OP_SD: eaddr: %#lx, gpr: %#lx, data: %#llx\n",
1632 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1633 			  vcpu->arch.gprs[rt], *(u64 *)data);
1634 		break;
1635 #endif
1636 
1637 	case sw_op:
1638 		run->mmio.len = 4;
1639 		*(u32 *)data = vcpu->arch.gprs[rt];
1640 
1641 		kvm_debug("[%#lx] OP_SW: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1642 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1643 			  vcpu->arch.gprs[rt], *(u32 *)data);
1644 		break;
1645 
1646 	case sh_op:
1647 		run->mmio.len = 2;
1648 		*(u16 *)data = vcpu->arch.gprs[rt];
1649 
1650 		kvm_debug("[%#lx] OP_SH: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1651 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1652 			  vcpu->arch.gprs[rt], *(u16 *)data);
1653 		break;
1654 
1655 	case sb_op:
1656 		run->mmio.len = 1;
1657 		*(u8 *)data = vcpu->arch.gprs[rt];
1658 
1659 		kvm_debug("[%#lx] OP_SB: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1660 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1661 			  vcpu->arch.gprs[rt], *(u8 *)data);
1662 		break;
1663 
1664 	default:
1665 		kvm_err("Store not yet supported (inst=0x%08x)\n",
1666 			inst.word);
1667 		goto out_fail;
1668 	}
1669 
1670 	run->mmio.is_write = 1;
1671 	vcpu->mmio_needed = 1;
1672 	vcpu->mmio_is_write = 1;
1673 	return EMULATE_DO_MMIO;
1674 
1675 out_fail:
1676 	/* Rollback PC if emulation was unsuccessful */
1677 	vcpu->arch.pc = curr_pc;
1678 	return EMULATE_FAIL;
1679 }
1680 
1681 enum emulation_result kvm_mips_emulate_load(union mips_instruction inst,
1682 					    u32 cause, struct kvm_run *run,
1683 					    struct kvm_vcpu *vcpu)
1684 {
1685 	enum emulation_result er;
1686 	unsigned long curr_pc;
1687 	u32 op, rt;
1688 
1689 	rt = inst.i_format.rt;
1690 	op = inst.i_format.opcode;
1691 
1692 	/*
1693 	 * Find the resume PC now while we have safe and easy access to the
1694 	 * prior branch instruction, and save it for
1695 	 * kvm_mips_complete_mmio_load() to restore later.
1696 	 */
1697 	curr_pc = vcpu->arch.pc;
1698 	er = update_pc(vcpu, cause);
1699 	if (er == EMULATE_FAIL)
1700 		return er;
1701 	vcpu->arch.io_pc = vcpu->arch.pc;
1702 	vcpu->arch.pc = curr_pc;
1703 
1704 	vcpu->arch.io_gpr = rt;
1705 
1706 	run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1707 						vcpu->arch.host_cp0_badvaddr);
1708 	if (run->mmio.phys_addr == KVM_INVALID_ADDR)
1709 		return EMULATE_FAIL;
1710 
1711 	vcpu->mmio_needed = 2;	/* signed */
1712 	switch (op) {
1713 #if defined(CONFIG_64BIT) && defined(CONFIG_KVM_MIPS_VZ)
1714 	case ld_op:
1715 		run->mmio.len = 8;
1716 		break;
1717 
1718 	case lwu_op:
1719 		vcpu->mmio_needed = 1;	/* unsigned */
1720 		/* fall through */
1721 #endif
1722 	case lw_op:
1723 		run->mmio.len = 4;
1724 		break;
1725 
1726 	case lhu_op:
1727 		vcpu->mmio_needed = 1;	/* unsigned */
1728 		/* fall through */
1729 	case lh_op:
1730 		run->mmio.len = 2;
1731 		break;
1732 
1733 	case lbu_op:
1734 		vcpu->mmio_needed = 1;	/* unsigned */
1735 		/* fall through */
1736 	case lb_op:
1737 		run->mmio.len = 1;
1738 		break;
1739 
1740 	default:
1741 		kvm_err("Load not yet supported (inst=0x%08x)\n",
1742 			inst.word);
1743 		vcpu->mmio_needed = 0;
1744 		return EMULATE_FAIL;
1745 	}
1746 
1747 	run->mmio.is_write = 0;
1748 	vcpu->mmio_is_write = 0;
1749 	return EMULATE_DO_MMIO;
1750 }
1751 
1752 #ifndef CONFIG_KVM_MIPS_VZ
1753 static enum emulation_result kvm_mips_guest_cache_op(int (*fn)(unsigned long),
1754 						     unsigned long curr_pc,
1755 						     unsigned long addr,
1756 						     struct kvm_run *run,
1757 						     struct kvm_vcpu *vcpu,
1758 						     u32 cause)
1759 {
1760 	int err;
1761 
1762 	for (;;) {
1763 		/* Carefully attempt the cache operation */
1764 		kvm_trap_emul_gva_lockless_begin(vcpu);
1765 		err = fn(addr);
1766 		kvm_trap_emul_gva_lockless_end(vcpu);
1767 
1768 		if (likely(!err))
1769 			return EMULATE_DONE;
1770 
1771 		/*
1772 		 * Try to handle the fault and retry, maybe we just raced with a
1773 		 * GVA invalidation.
1774 		 */
1775 		switch (kvm_trap_emul_gva_fault(vcpu, addr, false)) {
1776 		case KVM_MIPS_GVA:
1777 		case KVM_MIPS_GPA:
1778 			/* bad virtual or physical address */
1779 			return EMULATE_FAIL;
1780 		case KVM_MIPS_TLB:
1781 			/* no matching guest TLB */
1782 			vcpu->arch.host_cp0_badvaddr = addr;
1783 			vcpu->arch.pc = curr_pc;
1784 			kvm_mips_emulate_tlbmiss_ld(cause, NULL, run, vcpu);
1785 			return EMULATE_EXCEPT;
1786 		case KVM_MIPS_TLBINV:
1787 			/* invalid matching guest TLB */
1788 			vcpu->arch.host_cp0_badvaddr = addr;
1789 			vcpu->arch.pc = curr_pc;
1790 			kvm_mips_emulate_tlbinv_ld(cause, NULL, run, vcpu);
1791 			return EMULATE_EXCEPT;
1792 		default:
1793 			break;
1794 		};
1795 	}
1796 }
1797 
1798 enum emulation_result kvm_mips_emulate_cache(union mips_instruction inst,
1799 					     u32 *opc, u32 cause,
1800 					     struct kvm_run *run,
1801 					     struct kvm_vcpu *vcpu)
1802 {
1803 	enum emulation_result er = EMULATE_DONE;
1804 	u32 cache, op_inst, op, base;
1805 	s16 offset;
1806 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1807 	unsigned long va;
1808 	unsigned long curr_pc;
1809 
1810 	/*
1811 	 * Update PC and hold onto current PC in case there is
1812 	 * an error and we want to rollback the PC
1813 	 */
1814 	curr_pc = vcpu->arch.pc;
1815 	er = update_pc(vcpu, cause);
1816 	if (er == EMULATE_FAIL)
1817 		return er;
1818 
1819 	base = inst.i_format.rs;
1820 	op_inst = inst.i_format.rt;
1821 	if (cpu_has_mips_r6)
1822 		offset = inst.spec3_format.simmediate;
1823 	else
1824 		offset = inst.i_format.simmediate;
1825 	cache = op_inst & CacheOp_Cache;
1826 	op = op_inst & CacheOp_Op;
1827 
1828 	va = arch->gprs[base] + offset;
1829 
1830 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1831 		  cache, op, base, arch->gprs[base], offset);
1832 
1833 	/*
1834 	 * Treat INDEX_INV as a nop, basically issued by Linux on startup to
1835 	 * invalidate the caches entirely by stepping through all the
1836 	 * ways/indexes
1837 	 */
1838 	if (op == Index_Writeback_Inv) {
1839 		kvm_debug("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1840 			  vcpu->arch.pc, vcpu->arch.gprs[31], cache, op, base,
1841 			  arch->gprs[base], offset);
1842 
1843 		if (cache == Cache_D) {
1844 #ifdef CONFIG_CPU_R4K_CACHE_TLB
1845 			r4k_blast_dcache();
1846 #else
1847 			switch (boot_cpu_type()) {
1848 			case CPU_CAVIUM_OCTEON3:
1849 				/* locally flush icache */
1850 				local_flush_icache_range(0, 0);
1851 				break;
1852 			default:
1853 				__flush_cache_all();
1854 				break;
1855 			}
1856 #endif
1857 		} else if (cache == Cache_I) {
1858 #ifdef CONFIG_CPU_R4K_CACHE_TLB
1859 			r4k_blast_icache();
1860 #else
1861 			switch (boot_cpu_type()) {
1862 			case CPU_CAVIUM_OCTEON3:
1863 				/* locally flush icache */
1864 				local_flush_icache_range(0, 0);
1865 				break;
1866 			default:
1867 				flush_icache_all();
1868 				break;
1869 			}
1870 #endif
1871 		} else {
1872 			kvm_err("%s: unsupported CACHE INDEX operation\n",
1873 				__func__);
1874 			return EMULATE_FAIL;
1875 		}
1876 
1877 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1878 		kvm_mips_trans_cache_index(inst, opc, vcpu);
1879 #endif
1880 		goto done;
1881 	}
1882 
1883 	/* XXXKYMA: Only a subset of cache ops are supported, used by Linux */
1884 	if (op_inst == Hit_Writeback_Inv_D || op_inst == Hit_Invalidate_D) {
1885 		/*
1886 		 * Perform the dcache part of icache synchronisation on the
1887 		 * guest's behalf.
1888 		 */
1889 		er = kvm_mips_guest_cache_op(protected_writeback_dcache_line,
1890 					     curr_pc, va, run, vcpu, cause);
1891 		if (er != EMULATE_DONE)
1892 			goto done;
1893 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1894 		/*
1895 		 * Replace the CACHE instruction, with a SYNCI, not the same,
1896 		 * but avoids a trap
1897 		 */
1898 		kvm_mips_trans_cache_va(inst, opc, vcpu);
1899 #endif
1900 	} else if (op_inst == Hit_Invalidate_I) {
1901 		/* Perform the icache synchronisation on the guest's behalf */
1902 		er = kvm_mips_guest_cache_op(protected_writeback_dcache_line,
1903 					     curr_pc, va, run, vcpu, cause);
1904 		if (er != EMULATE_DONE)
1905 			goto done;
1906 		er = kvm_mips_guest_cache_op(protected_flush_icache_line,
1907 					     curr_pc, va, run, vcpu, cause);
1908 		if (er != EMULATE_DONE)
1909 			goto done;
1910 
1911 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1912 		/* Replace the CACHE instruction, with a SYNCI */
1913 		kvm_mips_trans_cache_va(inst, opc, vcpu);
1914 #endif
1915 	} else {
1916 		kvm_err("NO-OP CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1917 			cache, op, base, arch->gprs[base], offset);
1918 		er = EMULATE_FAIL;
1919 	}
1920 
1921 done:
1922 	/* Rollback PC only if emulation was unsuccessful */
1923 	if (er == EMULATE_FAIL)
1924 		vcpu->arch.pc = curr_pc;
1925 	/* Guest exception needs guest to resume */
1926 	if (er == EMULATE_EXCEPT)
1927 		er = EMULATE_DONE;
1928 
1929 	return er;
1930 }
1931 
1932 enum emulation_result kvm_mips_emulate_inst(u32 cause, u32 *opc,
1933 					    struct kvm_run *run,
1934 					    struct kvm_vcpu *vcpu)
1935 {
1936 	union mips_instruction inst;
1937 	enum emulation_result er = EMULATE_DONE;
1938 	int err;
1939 
1940 	/* Fetch the instruction. */
1941 	if (cause & CAUSEF_BD)
1942 		opc += 1;
1943 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1944 	if (err)
1945 		return EMULATE_FAIL;
1946 
1947 	switch (inst.r_format.opcode) {
1948 	case cop0_op:
1949 		er = kvm_mips_emulate_CP0(inst, opc, cause, run, vcpu);
1950 		break;
1951 
1952 #ifndef CONFIG_CPU_MIPSR6
1953 	case cache_op:
1954 		++vcpu->stat.cache_exits;
1955 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1956 		er = kvm_mips_emulate_cache(inst, opc, cause, run, vcpu);
1957 		break;
1958 #else
1959 	case spec3_op:
1960 		switch (inst.spec3_format.func) {
1961 		case cache6_op:
1962 			++vcpu->stat.cache_exits;
1963 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1964 			er = kvm_mips_emulate_cache(inst, opc, cause, run,
1965 						    vcpu);
1966 			break;
1967 		default:
1968 			goto unknown;
1969 		};
1970 		break;
1971 unknown:
1972 #endif
1973 
1974 	default:
1975 		kvm_err("Instruction emulation not supported (%p/%#x)\n", opc,
1976 			inst.word);
1977 		kvm_arch_vcpu_dump_regs(vcpu);
1978 		er = EMULATE_FAIL;
1979 		break;
1980 	}
1981 
1982 	return er;
1983 }
1984 #endif /* CONFIG_KVM_MIPS_VZ */
1985 
1986 /**
1987  * kvm_mips_guest_exception_base() - Find guest exception vector base address.
1988  *
1989  * Returns:	The base address of the current guest exception vector, taking
1990  *		both Guest.CP0_Status.BEV and Guest.CP0_EBase into account.
1991  */
1992 long kvm_mips_guest_exception_base(struct kvm_vcpu *vcpu)
1993 {
1994 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1995 
1996 	if (kvm_read_c0_guest_status(cop0) & ST0_BEV)
1997 		return KVM_GUEST_CKSEG1ADDR(0x1fc00200);
1998 	else
1999 		return kvm_read_c0_guest_ebase(cop0) & MIPS_EBASE_BASE;
2000 }
2001 
2002 enum emulation_result kvm_mips_emulate_syscall(u32 cause,
2003 					       u32 *opc,
2004 					       struct kvm_run *run,
2005 					       struct kvm_vcpu *vcpu)
2006 {
2007 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2008 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2009 	enum emulation_result er = EMULATE_DONE;
2010 
2011 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2012 		/* save old pc */
2013 		kvm_write_c0_guest_epc(cop0, arch->pc);
2014 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2015 
2016 		if (cause & CAUSEF_BD)
2017 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2018 		else
2019 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2020 
2021 		kvm_debug("Delivering SYSCALL @ pc %#lx\n", arch->pc);
2022 
2023 		kvm_change_c0_guest_cause(cop0, (0xff),
2024 					  (EXCCODE_SYS << CAUSEB_EXCCODE));
2025 
2026 		/* Set PC to the exception entry point */
2027 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2028 
2029 	} else {
2030 		kvm_err("Trying to deliver SYSCALL when EXL is already set\n");
2031 		er = EMULATE_FAIL;
2032 	}
2033 
2034 	return er;
2035 }
2036 
2037 enum emulation_result kvm_mips_emulate_tlbmiss_ld(u32 cause,
2038 						  u32 *opc,
2039 						  struct kvm_run *run,
2040 						  struct kvm_vcpu *vcpu)
2041 {
2042 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2043 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2044 	unsigned long entryhi = (vcpu->arch.  host_cp0_badvaddr & VPN2_MASK) |
2045 			(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2046 
2047 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2048 		/* save old pc */
2049 		kvm_write_c0_guest_epc(cop0, arch->pc);
2050 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2051 
2052 		if (cause & CAUSEF_BD)
2053 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2054 		else
2055 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2056 
2057 		kvm_debug("[EXL == 0] delivering TLB MISS @ pc %#lx\n",
2058 			  arch->pc);
2059 
2060 		/* set pc to the exception entry point */
2061 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x0;
2062 
2063 	} else {
2064 		kvm_debug("[EXL == 1] delivering TLB MISS @ pc %#lx\n",
2065 			  arch->pc);
2066 
2067 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2068 	}
2069 
2070 	kvm_change_c0_guest_cause(cop0, (0xff),
2071 				  (EXCCODE_TLBL << CAUSEB_EXCCODE));
2072 
2073 	/* setup badvaddr, context and entryhi registers for the guest */
2074 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2075 	/* XXXKYMA: is the context register used by linux??? */
2076 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2077 
2078 	return EMULATE_DONE;
2079 }
2080 
2081 enum emulation_result kvm_mips_emulate_tlbinv_ld(u32 cause,
2082 						 u32 *opc,
2083 						 struct kvm_run *run,
2084 						 struct kvm_vcpu *vcpu)
2085 {
2086 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2087 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2088 	unsigned long entryhi =
2089 		(vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2090 		(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2091 
2092 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2093 		/* save old pc */
2094 		kvm_write_c0_guest_epc(cop0, arch->pc);
2095 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2096 
2097 		if (cause & CAUSEF_BD)
2098 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2099 		else
2100 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2101 
2102 		kvm_debug("[EXL == 0] delivering TLB INV @ pc %#lx\n",
2103 			  arch->pc);
2104 	} else {
2105 		kvm_debug("[EXL == 1] delivering TLB MISS @ pc %#lx\n",
2106 			  arch->pc);
2107 	}
2108 
2109 	/* set pc to the exception entry point */
2110 	arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2111 
2112 	kvm_change_c0_guest_cause(cop0, (0xff),
2113 				  (EXCCODE_TLBL << CAUSEB_EXCCODE));
2114 
2115 	/* setup badvaddr, context and entryhi registers for the guest */
2116 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2117 	/* XXXKYMA: is the context register used by linux??? */
2118 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2119 
2120 	return EMULATE_DONE;
2121 }
2122 
2123 enum emulation_result kvm_mips_emulate_tlbmiss_st(u32 cause,
2124 						  u32 *opc,
2125 						  struct kvm_run *run,
2126 						  struct kvm_vcpu *vcpu)
2127 {
2128 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2129 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2130 	unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2131 			(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2132 
2133 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2134 		/* save old pc */
2135 		kvm_write_c0_guest_epc(cop0, arch->pc);
2136 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2137 
2138 		if (cause & CAUSEF_BD)
2139 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2140 		else
2141 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2142 
2143 		kvm_debug("[EXL == 0] Delivering TLB MISS @ pc %#lx\n",
2144 			  arch->pc);
2145 
2146 		/* Set PC to the exception entry point */
2147 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x0;
2148 	} else {
2149 		kvm_debug("[EXL == 1] Delivering TLB MISS @ pc %#lx\n",
2150 			  arch->pc);
2151 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2152 	}
2153 
2154 	kvm_change_c0_guest_cause(cop0, (0xff),
2155 				  (EXCCODE_TLBS << CAUSEB_EXCCODE));
2156 
2157 	/* setup badvaddr, context and entryhi registers for the guest */
2158 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2159 	/* XXXKYMA: is the context register used by linux??? */
2160 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2161 
2162 	return EMULATE_DONE;
2163 }
2164 
2165 enum emulation_result kvm_mips_emulate_tlbinv_st(u32 cause,
2166 						 u32 *opc,
2167 						 struct kvm_run *run,
2168 						 struct kvm_vcpu *vcpu)
2169 {
2170 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2171 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2172 	unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2173 		(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2174 
2175 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2176 		/* save old pc */
2177 		kvm_write_c0_guest_epc(cop0, arch->pc);
2178 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2179 
2180 		if (cause & CAUSEF_BD)
2181 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2182 		else
2183 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2184 
2185 		kvm_debug("[EXL == 0] Delivering TLB MISS @ pc %#lx\n",
2186 			  arch->pc);
2187 	} else {
2188 		kvm_debug("[EXL == 1] Delivering TLB MISS @ pc %#lx\n",
2189 			  arch->pc);
2190 	}
2191 
2192 	/* Set PC to the exception entry point */
2193 	arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2194 
2195 	kvm_change_c0_guest_cause(cop0, (0xff),
2196 				  (EXCCODE_TLBS << CAUSEB_EXCCODE));
2197 
2198 	/* setup badvaddr, context and entryhi registers for the guest */
2199 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2200 	/* XXXKYMA: is the context register used by linux??? */
2201 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2202 
2203 	return EMULATE_DONE;
2204 }
2205 
2206 enum emulation_result kvm_mips_emulate_tlbmod(u32 cause,
2207 					      u32 *opc,
2208 					      struct kvm_run *run,
2209 					      struct kvm_vcpu *vcpu)
2210 {
2211 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2212 	unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2213 			(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2214 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2215 
2216 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2217 		/* save old pc */
2218 		kvm_write_c0_guest_epc(cop0, arch->pc);
2219 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2220 
2221 		if (cause & CAUSEF_BD)
2222 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2223 		else
2224 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2225 
2226 		kvm_debug("[EXL == 0] Delivering TLB MOD @ pc %#lx\n",
2227 			  arch->pc);
2228 	} else {
2229 		kvm_debug("[EXL == 1] Delivering TLB MOD @ pc %#lx\n",
2230 			  arch->pc);
2231 	}
2232 
2233 	arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2234 
2235 	kvm_change_c0_guest_cause(cop0, (0xff),
2236 				  (EXCCODE_MOD << CAUSEB_EXCCODE));
2237 
2238 	/* setup badvaddr, context and entryhi registers for the guest */
2239 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2240 	/* XXXKYMA: is the context register used by linux??? */
2241 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2242 
2243 	return EMULATE_DONE;
2244 }
2245 
2246 enum emulation_result kvm_mips_emulate_fpu_exc(u32 cause,
2247 					       u32 *opc,
2248 					       struct kvm_run *run,
2249 					       struct kvm_vcpu *vcpu)
2250 {
2251 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2252 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2253 
2254 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2255 		/* save old pc */
2256 		kvm_write_c0_guest_epc(cop0, arch->pc);
2257 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2258 
2259 		if (cause & CAUSEF_BD)
2260 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2261 		else
2262 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2263 
2264 	}
2265 
2266 	arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2267 
2268 	kvm_change_c0_guest_cause(cop0, (0xff),
2269 				  (EXCCODE_CPU << CAUSEB_EXCCODE));
2270 	kvm_change_c0_guest_cause(cop0, (CAUSEF_CE), (0x1 << CAUSEB_CE));
2271 
2272 	return EMULATE_DONE;
2273 }
2274 
2275 enum emulation_result kvm_mips_emulate_ri_exc(u32 cause,
2276 					      u32 *opc,
2277 					      struct kvm_run *run,
2278 					      struct kvm_vcpu *vcpu)
2279 {
2280 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2281 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2282 	enum emulation_result er = EMULATE_DONE;
2283 
2284 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2285 		/* save old pc */
2286 		kvm_write_c0_guest_epc(cop0, arch->pc);
2287 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2288 
2289 		if (cause & CAUSEF_BD)
2290 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2291 		else
2292 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2293 
2294 		kvm_debug("Delivering RI @ pc %#lx\n", arch->pc);
2295 
2296 		kvm_change_c0_guest_cause(cop0, (0xff),
2297 					  (EXCCODE_RI << CAUSEB_EXCCODE));
2298 
2299 		/* Set PC to the exception entry point */
2300 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2301 
2302 	} else {
2303 		kvm_err("Trying to deliver RI when EXL is already set\n");
2304 		er = EMULATE_FAIL;
2305 	}
2306 
2307 	return er;
2308 }
2309 
2310 enum emulation_result kvm_mips_emulate_bp_exc(u32 cause,
2311 					      u32 *opc,
2312 					      struct kvm_run *run,
2313 					      struct kvm_vcpu *vcpu)
2314 {
2315 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2316 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2317 	enum emulation_result er = EMULATE_DONE;
2318 
2319 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2320 		/* save old pc */
2321 		kvm_write_c0_guest_epc(cop0, arch->pc);
2322 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2323 
2324 		if (cause & CAUSEF_BD)
2325 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2326 		else
2327 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2328 
2329 		kvm_debug("Delivering BP @ pc %#lx\n", arch->pc);
2330 
2331 		kvm_change_c0_guest_cause(cop0, (0xff),
2332 					  (EXCCODE_BP << CAUSEB_EXCCODE));
2333 
2334 		/* Set PC to the exception entry point */
2335 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2336 
2337 	} else {
2338 		kvm_err("Trying to deliver BP when EXL is already set\n");
2339 		er = EMULATE_FAIL;
2340 	}
2341 
2342 	return er;
2343 }
2344 
2345 enum emulation_result kvm_mips_emulate_trap_exc(u32 cause,
2346 						u32 *opc,
2347 						struct kvm_run *run,
2348 						struct kvm_vcpu *vcpu)
2349 {
2350 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2351 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2352 	enum emulation_result er = EMULATE_DONE;
2353 
2354 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2355 		/* save old pc */
2356 		kvm_write_c0_guest_epc(cop0, arch->pc);
2357 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2358 
2359 		if (cause & CAUSEF_BD)
2360 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2361 		else
2362 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2363 
2364 		kvm_debug("Delivering TRAP @ pc %#lx\n", arch->pc);
2365 
2366 		kvm_change_c0_guest_cause(cop0, (0xff),
2367 					  (EXCCODE_TR << CAUSEB_EXCCODE));
2368 
2369 		/* Set PC to the exception entry point */
2370 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2371 
2372 	} else {
2373 		kvm_err("Trying to deliver TRAP when EXL is already set\n");
2374 		er = EMULATE_FAIL;
2375 	}
2376 
2377 	return er;
2378 }
2379 
2380 enum emulation_result kvm_mips_emulate_msafpe_exc(u32 cause,
2381 						  u32 *opc,
2382 						  struct kvm_run *run,
2383 						  struct kvm_vcpu *vcpu)
2384 {
2385 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2386 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2387 	enum emulation_result er = EMULATE_DONE;
2388 
2389 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2390 		/* save old pc */
2391 		kvm_write_c0_guest_epc(cop0, arch->pc);
2392 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2393 
2394 		if (cause & CAUSEF_BD)
2395 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2396 		else
2397 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2398 
2399 		kvm_debug("Delivering MSAFPE @ pc %#lx\n", arch->pc);
2400 
2401 		kvm_change_c0_guest_cause(cop0, (0xff),
2402 					  (EXCCODE_MSAFPE << CAUSEB_EXCCODE));
2403 
2404 		/* Set PC to the exception entry point */
2405 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2406 
2407 	} else {
2408 		kvm_err("Trying to deliver MSAFPE when EXL is already set\n");
2409 		er = EMULATE_FAIL;
2410 	}
2411 
2412 	return er;
2413 }
2414 
2415 enum emulation_result kvm_mips_emulate_fpe_exc(u32 cause,
2416 					       u32 *opc,
2417 					       struct kvm_run *run,
2418 					       struct kvm_vcpu *vcpu)
2419 {
2420 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2421 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2422 	enum emulation_result er = EMULATE_DONE;
2423 
2424 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2425 		/* save old pc */
2426 		kvm_write_c0_guest_epc(cop0, arch->pc);
2427 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2428 
2429 		if (cause & CAUSEF_BD)
2430 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2431 		else
2432 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2433 
2434 		kvm_debug("Delivering FPE @ pc %#lx\n", arch->pc);
2435 
2436 		kvm_change_c0_guest_cause(cop0, (0xff),
2437 					  (EXCCODE_FPE << CAUSEB_EXCCODE));
2438 
2439 		/* Set PC to the exception entry point */
2440 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2441 
2442 	} else {
2443 		kvm_err("Trying to deliver FPE when EXL is already set\n");
2444 		er = EMULATE_FAIL;
2445 	}
2446 
2447 	return er;
2448 }
2449 
2450 enum emulation_result kvm_mips_emulate_msadis_exc(u32 cause,
2451 						  u32 *opc,
2452 						  struct kvm_run *run,
2453 						  struct kvm_vcpu *vcpu)
2454 {
2455 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2456 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2457 	enum emulation_result er = EMULATE_DONE;
2458 
2459 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2460 		/* save old pc */
2461 		kvm_write_c0_guest_epc(cop0, arch->pc);
2462 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2463 
2464 		if (cause & CAUSEF_BD)
2465 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2466 		else
2467 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2468 
2469 		kvm_debug("Delivering MSADIS @ pc %#lx\n", arch->pc);
2470 
2471 		kvm_change_c0_guest_cause(cop0, (0xff),
2472 					  (EXCCODE_MSADIS << CAUSEB_EXCCODE));
2473 
2474 		/* Set PC to the exception entry point */
2475 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2476 
2477 	} else {
2478 		kvm_err("Trying to deliver MSADIS when EXL is already set\n");
2479 		er = EMULATE_FAIL;
2480 	}
2481 
2482 	return er;
2483 }
2484 
2485 enum emulation_result kvm_mips_handle_ri(u32 cause, u32 *opc,
2486 					 struct kvm_run *run,
2487 					 struct kvm_vcpu *vcpu)
2488 {
2489 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2490 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2491 	enum emulation_result er = EMULATE_DONE;
2492 	unsigned long curr_pc;
2493 	union mips_instruction inst;
2494 	int err;
2495 
2496 	/*
2497 	 * Update PC and hold onto current PC in case there is
2498 	 * an error and we want to rollback the PC
2499 	 */
2500 	curr_pc = vcpu->arch.pc;
2501 	er = update_pc(vcpu, cause);
2502 	if (er == EMULATE_FAIL)
2503 		return er;
2504 
2505 	/* Fetch the instruction. */
2506 	if (cause & CAUSEF_BD)
2507 		opc += 1;
2508 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
2509 	if (err) {
2510 		kvm_err("%s: Cannot get inst @ %p (%d)\n", __func__, opc, err);
2511 		return EMULATE_FAIL;
2512 	}
2513 
2514 	if (inst.r_format.opcode == spec3_op &&
2515 	    inst.r_format.func == rdhwr_op &&
2516 	    inst.r_format.rs == 0 &&
2517 	    (inst.r_format.re >> 3) == 0) {
2518 		int usermode = !KVM_GUEST_KERNEL_MODE(vcpu);
2519 		int rd = inst.r_format.rd;
2520 		int rt = inst.r_format.rt;
2521 		int sel = inst.r_format.re & 0x7;
2522 
2523 		/* If usermode, check RDHWR rd is allowed by guest HWREna */
2524 		if (usermode && !(kvm_read_c0_guest_hwrena(cop0) & BIT(rd))) {
2525 			kvm_debug("RDHWR %#x disallowed by HWREna @ %p\n",
2526 				  rd, opc);
2527 			goto emulate_ri;
2528 		}
2529 		switch (rd) {
2530 		case MIPS_HWR_CPUNUM:		/* CPU number */
2531 			arch->gprs[rt] = vcpu->vcpu_id;
2532 			break;
2533 		case MIPS_HWR_SYNCISTEP:	/* SYNCI length */
2534 			arch->gprs[rt] = min(current_cpu_data.dcache.linesz,
2535 					     current_cpu_data.icache.linesz);
2536 			break;
2537 		case MIPS_HWR_CC:		/* Read count register */
2538 			arch->gprs[rt] = (s32)kvm_mips_read_count(vcpu);
2539 			break;
2540 		case MIPS_HWR_CCRES:		/* Count register resolution */
2541 			switch (current_cpu_data.cputype) {
2542 			case CPU_20KC:
2543 			case CPU_25KF:
2544 				arch->gprs[rt] = 1;
2545 				break;
2546 			default:
2547 				arch->gprs[rt] = 2;
2548 			}
2549 			break;
2550 		case MIPS_HWR_ULR:		/* Read UserLocal register */
2551 			arch->gprs[rt] = kvm_read_c0_guest_userlocal(cop0);
2552 			break;
2553 
2554 		default:
2555 			kvm_debug("RDHWR %#x not supported @ %p\n", rd, opc);
2556 			goto emulate_ri;
2557 		}
2558 
2559 		trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR, KVM_TRACE_HWR(rd, sel),
2560 			      vcpu->arch.gprs[rt]);
2561 	} else {
2562 		kvm_debug("Emulate RI not supported @ %p: %#x\n",
2563 			  opc, inst.word);
2564 		goto emulate_ri;
2565 	}
2566 
2567 	return EMULATE_DONE;
2568 
2569 emulate_ri:
2570 	/*
2571 	 * Rollback PC (if in branch delay slot then the PC already points to
2572 	 * branch target), and pass the RI exception to the guest OS.
2573 	 */
2574 	vcpu->arch.pc = curr_pc;
2575 	return kvm_mips_emulate_ri_exc(cause, opc, run, vcpu);
2576 }
2577 
2578 enum emulation_result kvm_mips_complete_mmio_load(struct kvm_vcpu *vcpu,
2579 						  struct kvm_run *run)
2580 {
2581 	unsigned long *gpr = &vcpu->arch.gprs[vcpu->arch.io_gpr];
2582 	enum emulation_result er = EMULATE_DONE;
2583 
2584 	if (run->mmio.len > sizeof(*gpr)) {
2585 		kvm_err("Bad MMIO length: %d", run->mmio.len);
2586 		er = EMULATE_FAIL;
2587 		goto done;
2588 	}
2589 
2590 	/* Restore saved resume PC */
2591 	vcpu->arch.pc = vcpu->arch.io_pc;
2592 
2593 	switch (run->mmio.len) {
2594 	case 8:
2595 		*gpr = *(s64 *)run->mmio.data;
2596 		break;
2597 
2598 	case 4:
2599 		if (vcpu->mmio_needed == 2)
2600 			*gpr = *(s32 *)run->mmio.data;
2601 		else
2602 			*gpr = *(u32 *)run->mmio.data;
2603 		break;
2604 
2605 	case 2:
2606 		if (vcpu->mmio_needed == 2)
2607 			*gpr = *(s16 *) run->mmio.data;
2608 		else
2609 			*gpr = *(u16 *)run->mmio.data;
2610 
2611 		break;
2612 	case 1:
2613 		if (vcpu->mmio_needed == 2)
2614 			*gpr = *(s8 *) run->mmio.data;
2615 		else
2616 			*gpr = *(u8 *) run->mmio.data;
2617 		break;
2618 	}
2619 
2620 done:
2621 	return er;
2622 }
2623 
2624 static enum emulation_result kvm_mips_emulate_exc(u32 cause,
2625 						  u32 *opc,
2626 						  struct kvm_run *run,
2627 						  struct kvm_vcpu *vcpu)
2628 {
2629 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2630 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2631 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2632 	enum emulation_result er = EMULATE_DONE;
2633 
2634 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2635 		/* save old pc */
2636 		kvm_write_c0_guest_epc(cop0, arch->pc);
2637 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2638 
2639 		if (cause & CAUSEF_BD)
2640 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2641 		else
2642 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2643 
2644 		kvm_change_c0_guest_cause(cop0, (0xff),
2645 					  (exccode << CAUSEB_EXCCODE));
2646 
2647 		/* Set PC to the exception entry point */
2648 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2649 		kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2650 
2651 		kvm_debug("Delivering EXC %d @ pc %#lx, badVaddr: %#lx\n",
2652 			  exccode, kvm_read_c0_guest_epc(cop0),
2653 			  kvm_read_c0_guest_badvaddr(cop0));
2654 	} else {
2655 		kvm_err("Trying to deliver EXC when EXL is already set\n");
2656 		er = EMULATE_FAIL;
2657 	}
2658 
2659 	return er;
2660 }
2661 
2662 enum emulation_result kvm_mips_check_privilege(u32 cause,
2663 					       u32 *opc,
2664 					       struct kvm_run *run,
2665 					       struct kvm_vcpu *vcpu)
2666 {
2667 	enum emulation_result er = EMULATE_DONE;
2668 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2669 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
2670 
2671 	int usermode = !KVM_GUEST_KERNEL_MODE(vcpu);
2672 
2673 	if (usermode) {
2674 		switch (exccode) {
2675 		case EXCCODE_INT:
2676 		case EXCCODE_SYS:
2677 		case EXCCODE_BP:
2678 		case EXCCODE_RI:
2679 		case EXCCODE_TR:
2680 		case EXCCODE_MSAFPE:
2681 		case EXCCODE_FPE:
2682 		case EXCCODE_MSADIS:
2683 			break;
2684 
2685 		case EXCCODE_CPU:
2686 			if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 0)
2687 				er = EMULATE_PRIV_FAIL;
2688 			break;
2689 
2690 		case EXCCODE_MOD:
2691 			break;
2692 
2693 		case EXCCODE_TLBL:
2694 			/*
2695 			 * We we are accessing Guest kernel space, then send an
2696 			 * address error exception to the guest
2697 			 */
2698 			if (badvaddr >= (unsigned long) KVM_GUEST_KSEG0) {
2699 				kvm_debug("%s: LD MISS @ %#lx\n", __func__,
2700 					  badvaddr);
2701 				cause &= ~0xff;
2702 				cause |= (EXCCODE_ADEL << CAUSEB_EXCCODE);
2703 				er = EMULATE_PRIV_FAIL;
2704 			}
2705 			break;
2706 
2707 		case EXCCODE_TLBS:
2708 			/*
2709 			 * We we are accessing Guest kernel space, then send an
2710 			 * address error exception to the guest
2711 			 */
2712 			if (badvaddr >= (unsigned long) KVM_GUEST_KSEG0) {
2713 				kvm_debug("%s: ST MISS @ %#lx\n", __func__,
2714 					  badvaddr);
2715 				cause &= ~0xff;
2716 				cause |= (EXCCODE_ADES << CAUSEB_EXCCODE);
2717 				er = EMULATE_PRIV_FAIL;
2718 			}
2719 			break;
2720 
2721 		case EXCCODE_ADES:
2722 			kvm_debug("%s: address error ST @ %#lx\n", __func__,
2723 				  badvaddr);
2724 			if ((badvaddr & PAGE_MASK) == KVM_GUEST_COMMPAGE_ADDR) {
2725 				cause &= ~0xff;
2726 				cause |= (EXCCODE_TLBS << CAUSEB_EXCCODE);
2727 			}
2728 			er = EMULATE_PRIV_FAIL;
2729 			break;
2730 		case EXCCODE_ADEL:
2731 			kvm_debug("%s: address error LD @ %#lx\n", __func__,
2732 				  badvaddr);
2733 			if ((badvaddr & PAGE_MASK) == KVM_GUEST_COMMPAGE_ADDR) {
2734 				cause &= ~0xff;
2735 				cause |= (EXCCODE_TLBL << CAUSEB_EXCCODE);
2736 			}
2737 			er = EMULATE_PRIV_FAIL;
2738 			break;
2739 		default:
2740 			er = EMULATE_PRIV_FAIL;
2741 			break;
2742 		}
2743 	}
2744 
2745 	if (er == EMULATE_PRIV_FAIL)
2746 		kvm_mips_emulate_exc(cause, opc, run, vcpu);
2747 
2748 	return er;
2749 }
2750 
2751 /*
2752  * User Address (UA) fault, this could happen if
2753  * (1) TLB entry not present/valid in both Guest and shadow host TLBs, in this
2754  *     case we pass on the fault to the guest kernel and let it handle it.
2755  * (2) TLB entry is present in the Guest TLB but not in the shadow, in this
2756  *     case we inject the TLB from the Guest TLB into the shadow host TLB
2757  */
2758 enum emulation_result kvm_mips_handle_tlbmiss(u32 cause,
2759 					      u32 *opc,
2760 					      struct kvm_run *run,
2761 					      struct kvm_vcpu *vcpu,
2762 					      bool write_fault)
2763 {
2764 	enum emulation_result er = EMULATE_DONE;
2765 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2766 	unsigned long va = vcpu->arch.host_cp0_badvaddr;
2767 	int index;
2768 
2769 	kvm_debug("kvm_mips_handle_tlbmiss: badvaddr: %#lx\n",
2770 		  vcpu->arch.host_cp0_badvaddr);
2771 
2772 	/*
2773 	 * KVM would not have got the exception if this entry was valid in the
2774 	 * shadow host TLB. Check the Guest TLB, if the entry is not there then
2775 	 * send the guest an exception. The guest exc handler should then inject
2776 	 * an entry into the guest TLB.
2777 	 */
2778 	index = kvm_mips_guest_tlb_lookup(vcpu,
2779 		      (va & VPN2_MASK) |
2780 		      (kvm_read_c0_guest_entryhi(vcpu->arch.cop0) &
2781 		       KVM_ENTRYHI_ASID));
2782 	if (index < 0) {
2783 		if (exccode == EXCCODE_TLBL) {
2784 			er = kvm_mips_emulate_tlbmiss_ld(cause, opc, run, vcpu);
2785 		} else if (exccode == EXCCODE_TLBS) {
2786 			er = kvm_mips_emulate_tlbmiss_st(cause, opc, run, vcpu);
2787 		} else {
2788 			kvm_err("%s: invalid exc code: %d\n", __func__,
2789 				exccode);
2790 			er = EMULATE_FAIL;
2791 		}
2792 	} else {
2793 		struct kvm_mips_tlb *tlb = &vcpu->arch.guest_tlb[index];
2794 
2795 		/*
2796 		 * Check if the entry is valid, if not then setup a TLB invalid
2797 		 * exception to the guest
2798 		 */
2799 		if (!TLB_IS_VALID(*tlb, va)) {
2800 			if (exccode == EXCCODE_TLBL) {
2801 				er = kvm_mips_emulate_tlbinv_ld(cause, opc, run,
2802 								vcpu);
2803 			} else if (exccode == EXCCODE_TLBS) {
2804 				er = kvm_mips_emulate_tlbinv_st(cause, opc, run,
2805 								vcpu);
2806 			} else {
2807 				kvm_err("%s: invalid exc code: %d\n", __func__,
2808 					exccode);
2809 				er = EMULATE_FAIL;
2810 			}
2811 		} else {
2812 			kvm_debug("Injecting hi: %#lx, lo0: %#lx, lo1: %#lx into shadow host TLB\n",
2813 				  tlb->tlb_hi, tlb->tlb_lo[0], tlb->tlb_lo[1]);
2814 			/*
2815 			 * OK we have a Guest TLB entry, now inject it into the
2816 			 * shadow host TLB
2817 			 */
2818 			if (kvm_mips_handle_mapped_seg_tlb_fault(vcpu, tlb, va,
2819 								 write_fault)) {
2820 				kvm_err("%s: handling mapped seg tlb fault for %lx, index: %u, vcpu: %p, ASID: %#lx\n",
2821 					__func__, va, index, vcpu,
2822 					read_c0_entryhi());
2823 				er = EMULATE_FAIL;
2824 			}
2825 		}
2826 	}
2827 
2828 	return er;
2829 }
2830