xref: /openbmc/linux/arch/mips/kvm/emulate.c (revision a16be368)
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/memblock.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 			fallthrough;
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 		fallthrough;
144 	case j_op:
145 		epc += 4;
146 		epc >>= 28;
147 		epc <<= 28;
148 		epc |= (insn.j_format.target << 2);
149 		nextpc = epc;
150 		break;
151 
152 		/* These are conditional and in i_format. */
153 	case beq_op:
154 	case beql_op:
155 		if (arch->gprs[insn.i_format.rs] ==
156 		    arch->gprs[insn.i_format.rt])
157 			epc = epc + 4 + (insn.i_format.simmediate << 2);
158 		else
159 			epc += 8;
160 		nextpc = epc;
161 		break;
162 
163 	case bne_op:
164 	case bnel_op:
165 		if (arch->gprs[insn.i_format.rs] !=
166 		    arch->gprs[insn.i_format.rt])
167 			epc = epc + 4 + (insn.i_format.simmediate << 2);
168 		else
169 			epc += 8;
170 		nextpc = epc;
171 		break;
172 
173 	case blez_op:	/* POP06 */
174 #ifndef CONFIG_CPU_MIPSR6
175 	case blezl_op:	/* removed in R6 */
176 #endif
177 		if (insn.i_format.rt != 0)
178 			goto compact_branch;
179 		if ((long)arch->gprs[insn.i_format.rs] <= 0)
180 			epc = epc + 4 + (insn.i_format.simmediate << 2);
181 		else
182 			epc += 8;
183 		nextpc = epc;
184 		break;
185 
186 	case bgtz_op:	/* POP07 */
187 #ifndef CONFIG_CPU_MIPSR6
188 	case bgtzl_op:	/* removed in R6 */
189 #endif
190 		if (insn.i_format.rt != 0)
191 			goto compact_branch;
192 		if ((long)arch->gprs[insn.i_format.rs] > 0)
193 			epc = epc + 4 + (insn.i_format.simmediate << 2);
194 		else
195 			epc += 8;
196 		nextpc = epc;
197 		break;
198 
199 		/* And now the FPA/cp1 branch instructions. */
200 	case cop1_op:
201 		kvm_err("%s: unsupported cop1_op\n", __func__);
202 		return -EINVAL;
203 
204 #ifdef CONFIG_CPU_MIPSR6
205 	/* R6 added the following compact branches with forbidden slots */
206 	case blezl_op:	/* POP26 */
207 	case bgtzl_op:	/* POP27 */
208 		/* only rt == 0 isn't compact branch */
209 		if (insn.i_format.rt != 0)
210 			goto compact_branch;
211 		return -EINVAL;
212 	case pop10_op:
213 	case pop30_op:
214 		/* only rs == rt == 0 is reserved, rest are compact branches */
215 		if (insn.i_format.rs != 0 || insn.i_format.rt != 0)
216 			goto compact_branch;
217 		return -EINVAL;
218 	case pop66_op:
219 	case pop76_op:
220 		/* only rs == 0 isn't compact branch */
221 		if (insn.i_format.rs != 0)
222 			goto compact_branch;
223 		return -EINVAL;
224 compact_branch:
225 		/*
226 		 * If we've hit an exception on the forbidden slot, then
227 		 * the branch must not have been taken.
228 		 */
229 		epc += 8;
230 		nextpc = epc;
231 		break;
232 #else
233 compact_branch:
234 		/* Fall through - Compact branches not supported before R6 */
235 #endif
236 	default:
237 		return -EINVAL;
238 	}
239 
240 	*out = nextpc;
241 	return 0;
242 }
243 
244 enum emulation_result update_pc(struct kvm_vcpu *vcpu, u32 cause)
245 {
246 	int err;
247 
248 	if (cause & CAUSEF_BD) {
249 		err = kvm_compute_return_epc(vcpu, vcpu->arch.pc,
250 					     &vcpu->arch.pc);
251 		if (err)
252 			return EMULATE_FAIL;
253 	} else {
254 		vcpu->arch.pc += 4;
255 	}
256 
257 	kvm_debug("update_pc(): New PC: %#lx\n", vcpu->arch.pc);
258 
259 	return EMULATE_DONE;
260 }
261 
262 /**
263  * kvm_get_badinstr() - Get bad instruction encoding.
264  * @opc:	Guest pointer to faulting instruction.
265  * @vcpu:	KVM VCPU information.
266  *
267  * Gets the instruction encoding of the faulting instruction, using the saved
268  * BadInstr register value if it exists, otherwise falling back to reading guest
269  * memory at @opc.
270  *
271  * Returns:	The instruction encoding of the faulting instruction.
272  */
273 int kvm_get_badinstr(u32 *opc, struct kvm_vcpu *vcpu, u32 *out)
274 {
275 	if (cpu_has_badinstr) {
276 		*out = vcpu->arch.host_cp0_badinstr;
277 		return 0;
278 	} else {
279 		return kvm_get_inst(opc, vcpu, out);
280 	}
281 }
282 
283 /**
284  * kvm_get_badinstrp() - Get bad prior instruction encoding.
285  * @opc:	Guest pointer to prior faulting instruction.
286  * @vcpu:	KVM VCPU information.
287  *
288  * Gets the instruction encoding of the prior faulting instruction (the branch
289  * containing the delay slot which faulted), using the saved BadInstrP register
290  * value if it exists, otherwise falling back to reading guest memory at @opc.
291  *
292  * Returns:	The instruction encoding of the prior faulting instruction.
293  */
294 int kvm_get_badinstrp(u32 *opc, struct kvm_vcpu *vcpu, u32 *out)
295 {
296 	if (cpu_has_badinstrp) {
297 		*out = vcpu->arch.host_cp0_badinstrp;
298 		return 0;
299 	} else {
300 		return kvm_get_inst(opc, vcpu, out);
301 	}
302 }
303 
304 /**
305  * kvm_mips_count_disabled() - Find whether the CP0_Count timer is disabled.
306  * @vcpu:	Virtual CPU.
307  *
308  * Returns:	1 if the CP0_Count timer is disabled by either the guest
309  *		CP0_Cause.DC bit or the count_ctl.DC bit.
310  *		0 otherwise (in which case CP0_Count timer is running).
311  */
312 int kvm_mips_count_disabled(struct kvm_vcpu *vcpu)
313 {
314 	struct mips_coproc *cop0 = vcpu->arch.cop0;
315 
316 	return	(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC) ||
317 		(kvm_read_c0_guest_cause(cop0) & CAUSEF_DC);
318 }
319 
320 /**
321  * kvm_mips_ktime_to_count() - Scale ktime_t to a 32-bit count.
322  *
323  * Caches the dynamic nanosecond bias in vcpu->arch.count_dyn_bias.
324  *
325  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
326  */
327 static u32 kvm_mips_ktime_to_count(struct kvm_vcpu *vcpu, ktime_t now)
328 {
329 	s64 now_ns, periods;
330 	u64 delta;
331 
332 	now_ns = ktime_to_ns(now);
333 	delta = now_ns + vcpu->arch.count_dyn_bias;
334 
335 	if (delta >= vcpu->arch.count_period) {
336 		/* If delta is out of safe range the bias needs adjusting */
337 		periods = div64_s64(now_ns, vcpu->arch.count_period);
338 		vcpu->arch.count_dyn_bias = -periods * vcpu->arch.count_period;
339 		/* Recalculate delta with new bias */
340 		delta = now_ns + vcpu->arch.count_dyn_bias;
341 	}
342 
343 	/*
344 	 * We've ensured that:
345 	 *   delta < count_period
346 	 *
347 	 * Therefore the intermediate delta*count_hz will never overflow since
348 	 * at the boundary condition:
349 	 *   delta = count_period
350 	 *   delta = NSEC_PER_SEC * 2^32 / count_hz
351 	 *   delta * count_hz = NSEC_PER_SEC * 2^32
352 	 */
353 	return div_u64(delta * vcpu->arch.count_hz, NSEC_PER_SEC);
354 }
355 
356 /**
357  * kvm_mips_count_time() - Get effective current time.
358  * @vcpu:	Virtual CPU.
359  *
360  * Get effective monotonic ktime. This is usually a straightforward ktime_get(),
361  * except when the master disable bit is set in count_ctl, in which case it is
362  * count_resume, i.e. the time that the count was disabled.
363  *
364  * Returns:	Effective monotonic ktime for CP0_Count.
365  */
366 static inline ktime_t kvm_mips_count_time(struct kvm_vcpu *vcpu)
367 {
368 	if (unlikely(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC))
369 		return vcpu->arch.count_resume;
370 
371 	return ktime_get();
372 }
373 
374 /**
375  * kvm_mips_read_count_running() - Read the current count value as if running.
376  * @vcpu:	Virtual CPU.
377  * @now:	Kernel time to read CP0_Count at.
378  *
379  * Returns the current guest CP0_Count register at time @now and handles if the
380  * timer interrupt is pending and hasn't been handled yet.
381  *
382  * Returns:	The current value of the guest CP0_Count register.
383  */
384 static u32 kvm_mips_read_count_running(struct kvm_vcpu *vcpu, ktime_t now)
385 {
386 	struct mips_coproc *cop0 = vcpu->arch.cop0;
387 	ktime_t expires, threshold;
388 	u32 count, compare;
389 	int running;
390 
391 	/* Calculate the biased and scaled guest CP0_Count */
392 	count = vcpu->arch.count_bias + kvm_mips_ktime_to_count(vcpu, now);
393 	compare = kvm_read_c0_guest_compare(cop0);
394 
395 	/*
396 	 * Find whether CP0_Count has reached the closest timer interrupt. If
397 	 * not, we shouldn't inject it.
398 	 */
399 	if ((s32)(count - compare) < 0)
400 		return count;
401 
402 	/*
403 	 * The CP0_Count we're going to return has already reached the closest
404 	 * timer interrupt. Quickly check if it really is a new interrupt by
405 	 * looking at whether the interval until the hrtimer expiry time is
406 	 * less than 1/4 of the timer period.
407 	 */
408 	expires = hrtimer_get_expires(&vcpu->arch.comparecount_timer);
409 	threshold = ktime_add_ns(now, vcpu->arch.count_period / 4);
410 	if (ktime_before(expires, threshold)) {
411 		/*
412 		 * Cancel it while we handle it so there's no chance of
413 		 * interference with the timeout handler.
414 		 */
415 		running = hrtimer_cancel(&vcpu->arch.comparecount_timer);
416 
417 		/* Nothing should be waiting on the timeout */
418 		kvm_mips_callbacks->queue_timer_int(vcpu);
419 
420 		/*
421 		 * Restart the timer if it was running based on the expiry time
422 		 * we read, so that we don't push it back 2 periods.
423 		 */
424 		if (running) {
425 			expires = ktime_add_ns(expires,
426 					       vcpu->arch.count_period);
427 			hrtimer_start(&vcpu->arch.comparecount_timer, expires,
428 				      HRTIMER_MODE_ABS);
429 		}
430 	}
431 
432 	return count;
433 }
434 
435 /**
436  * kvm_mips_read_count() - Read the current count value.
437  * @vcpu:	Virtual CPU.
438  *
439  * Read the current guest CP0_Count value, taking into account whether the timer
440  * is stopped.
441  *
442  * Returns:	The current guest CP0_Count value.
443  */
444 u32 kvm_mips_read_count(struct kvm_vcpu *vcpu)
445 {
446 	struct mips_coproc *cop0 = vcpu->arch.cop0;
447 
448 	/* If count disabled just read static copy of count */
449 	if (kvm_mips_count_disabled(vcpu))
450 		return kvm_read_c0_guest_count(cop0);
451 
452 	return kvm_mips_read_count_running(vcpu, ktime_get());
453 }
454 
455 /**
456  * kvm_mips_freeze_hrtimer() - Safely stop the hrtimer.
457  * @vcpu:	Virtual CPU.
458  * @count:	Output pointer for CP0_Count value at point of freeze.
459  *
460  * Freeze the hrtimer safely and return both the ktime and the CP0_Count value
461  * at the point it was frozen. It is guaranteed that any pending interrupts at
462  * the point it was frozen are handled, and none after that point.
463  *
464  * This is useful where the time/CP0_Count is needed in the calculation of the
465  * new parameters.
466  *
467  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
468  *
469  * Returns:	The ktime at the point of freeze.
470  */
471 ktime_t kvm_mips_freeze_hrtimer(struct kvm_vcpu *vcpu, u32 *count)
472 {
473 	ktime_t now;
474 
475 	/* stop hrtimer before finding time */
476 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
477 	now = ktime_get();
478 
479 	/* find count at this point and handle pending hrtimer */
480 	*count = kvm_mips_read_count_running(vcpu, now);
481 
482 	return now;
483 }
484 
485 /**
486  * kvm_mips_resume_hrtimer() - Resume hrtimer, updating expiry.
487  * @vcpu:	Virtual CPU.
488  * @now:	ktime at point of resume.
489  * @count:	CP0_Count at point of resume.
490  *
491  * Resumes the timer and updates the timer expiry based on @now and @count.
492  * This can be used in conjunction with kvm_mips_freeze_timer() when timer
493  * parameters need to be changed.
494  *
495  * It is guaranteed that a timer interrupt immediately after resume will be
496  * handled, but not if CP_Compare is exactly at @count. That case is already
497  * handled by kvm_mips_freeze_timer().
498  *
499  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
500  */
501 static void kvm_mips_resume_hrtimer(struct kvm_vcpu *vcpu,
502 				    ktime_t now, u32 count)
503 {
504 	struct mips_coproc *cop0 = vcpu->arch.cop0;
505 	u32 compare;
506 	u64 delta;
507 	ktime_t expire;
508 
509 	/* Calculate timeout (wrap 0 to 2^32) */
510 	compare = kvm_read_c0_guest_compare(cop0);
511 	delta = (u64)(u32)(compare - count - 1) + 1;
512 	delta = div_u64(delta * NSEC_PER_SEC, vcpu->arch.count_hz);
513 	expire = ktime_add_ns(now, delta);
514 
515 	/* Update hrtimer to use new timeout */
516 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
517 	hrtimer_start(&vcpu->arch.comparecount_timer, expire, HRTIMER_MODE_ABS);
518 }
519 
520 /**
521  * kvm_mips_restore_hrtimer() - Restore hrtimer after a gap, updating expiry.
522  * @vcpu:	Virtual CPU.
523  * @before:	Time before Count was saved, lower bound of drift calculation.
524  * @count:	CP0_Count at point of restore.
525  * @min_drift:	Minimum amount of drift permitted before correction.
526  *		Must be <= 0.
527  *
528  * Restores the timer from a particular @count, accounting for drift. This can
529  * be used in conjunction with kvm_mips_freeze_timer() when a hardware timer is
530  * to be used for a period of time, but the exact ktime corresponding to the
531  * final Count that must be restored is not known.
532  *
533  * It is gauranteed that a timer interrupt immediately after restore will be
534  * handled, but not if CP0_Compare is exactly at @count. That case should
535  * already be handled when the hardware timer state is saved.
536  *
537  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is not
538  * stopped).
539  *
540  * Returns:	Amount of correction to count_bias due to drift.
541  */
542 int kvm_mips_restore_hrtimer(struct kvm_vcpu *vcpu, ktime_t before,
543 			     u32 count, int min_drift)
544 {
545 	ktime_t now, count_time;
546 	u32 now_count, before_count;
547 	u64 delta;
548 	int drift, ret = 0;
549 
550 	/* Calculate expected count at before */
551 	before_count = vcpu->arch.count_bias +
552 			kvm_mips_ktime_to_count(vcpu, before);
553 
554 	/*
555 	 * Detect significantly negative drift, where count is lower than
556 	 * expected. Some negative drift is expected when hardware counter is
557 	 * set after kvm_mips_freeze_timer(), and it is harmless to allow the
558 	 * time to jump forwards a little, within reason. If the drift is too
559 	 * significant, adjust the bias to avoid a big Guest.CP0_Count jump.
560 	 */
561 	drift = count - before_count;
562 	if (drift < min_drift) {
563 		count_time = before;
564 		vcpu->arch.count_bias += drift;
565 		ret = drift;
566 		goto resume;
567 	}
568 
569 	/* Calculate expected count right now */
570 	now = ktime_get();
571 	now_count = vcpu->arch.count_bias + kvm_mips_ktime_to_count(vcpu, now);
572 
573 	/*
574 	 * Detect positive drift, where count is higher than expected, and
575 	 * adjust the bias to avoid guest time going backwards.
576 	 */
577 	drift = count - now_count;
578 	if (drift > 0) {
579 		count_time = now;
580 		vcpu->arch.count_bias += drift;
581 		ret = drift;
582 		goto resume;
583 	}
584 
585 	/* Subtract nanosecond delta to find ktime when count was read */
586 	delta = (u64)(u32)(now_count - count);
587 	delta = div_u64(delta * NSEC_PER_SEC, vcpu->arch.count_hz);
588 	count_time = ktime_sub_ns(now, delta);
589 
590 resume:
591 	/* Resume using the calculated ktime */
592 	kvm_mips_resume_hrtimer(vcpu, count_time, count);
593 	return ret;
594 }
595 
596 /**
597  * kvm_mips_write_count() - Modify the count and update timer.
598  * @vcpu:	Virtual CPU.
599  * @count:	Guest CP0_Count value to set.
600  *
601  * Sets the CP0_Count value and updates the timer accordingly.
602  */
603 void kvm_mips_write_count(struct kvm_vcpu *vcpu, u32 count)
604 {
605 	struct mips_coproc *cop0 = vcpu->arch.cop0;
606 	ktime_t now;
607 
608 	/* Calculate bias */
609 	now = kvm_mips_count_time(vcpu);
610 	vcpu->arch.count_bias = count - kvm_mips_ktime_to_count(vcpu, now);
611 
612 	if (kvm_mips_count_disabled(vcpu))
613 		/* The timer's disabled, adjust the static count */
614 		kvm_write_c0_guest_count(cop0, count);
615 	else
616 		/* Update timeout */
617 		kvm_mips_resume_hrtimer(vcpu, now, count);
618 }
619 
620 /**
621  * kvm_mips_init_count() - Initialise timer.
622  * @vcpu:	Virtual CPU.
623  * @count_hz:	Frequency of timer.
624  *
625  * Initialise the timer to the specified frequency, zero it, and set it going if
626  * it's enabled.
627  */
628 void kvm_mips_init_count(struct kvm_vcpu *vcpu, unsigned long count_hz)
629 {
630 	vcpu->arch.count_hz = count_hz;
631 	vcpu->arch.count_period = div_u64((u64)NSEC_PER_SEC << 32, count_hz);
632 	vcpu->arch.count_dyn_bias = 0;
633 
634 	/* Starting at 0 */
635 	kvm_mips_write_count(vcpu, 0);
636 }
637 
638 /**
639  * kvm_mips_set_count_hz() - Update the frequency of the timer.
640  * @vcpu:	Virtual CPU.
641  * @count_hz:	Frequency of CP0_Count timer in Hz.
642  *
643  * Change the frequency of the CP0_Count timer. This is done atomically so that
644  * CP0_Count is continuous and no timer interrupt is lost.
645  *
646  * Returns:	-EINVAL if @count_hz is out of range.
647  *		0 on success.
648  */
649 int kvm_mips_set_count_hz(struct kvm_vcpu *vcpu, s64 count_hz)
650 {
651 	struct mips_coproc *cop0 = vcpu->arch.cop0;
652 	int dc;
653 	ktime_t now;
654 	u32 count;
655 
656 	/* ensure the frequency is in a sensible range... */
657 	if (count_hz <= 0 || count_hz > NSEC_PER_SEC)
658 		return -EINVAL;
659 	/* ... and has actually changed */
660 	if (vcpu->arch.count_hz == count_hz)
661 		return 0;
662 
663 	/* Safely freeze timer so we can keep it continuous */
664 	dc = kvm_mips_count_disabled(vcpu);
665 	if (dc) {
666 		now = kvm_mips_count_time(vcpu);
667 		count = kvm_read_c0_guest_count(cop0);
668 	} else {
669 		now = kvm_mips_freeze_hrtimer(vcpu, &count);
670 	}
671 
672 	/* Update the frequency */
673 	vcpu->arch.count_hz = count_hz;
674 	vcpu->arch.count_period = div_u64((u64)NSEC_PER_SEC << 32, count_hz);
675 	vcpu->arch.count_dyn_bias = 0;
676 
677 	/* Calculate adjusted bias so dynamic count is unchanged */
678 	vcpu->arch.count_bias = count - kvm_mips_ktime_to_count(vcpu, now);
679 
680 	/* Update and resume hrtimer */
681 	if (!dc)
682 		kvm_mips_resume_hrtimer(vcpu, now, count);
683 	return 0;
684 }
685 
686 /**
687  * kvm_mips_write_compare() - Modify compare and update timer.
688  * @vcpu:	Virtual CPU.
689  * @compare:	New CP0_Compare value.
690  * @ack:	Whether to acknowledge timer interrupt.
691  *
692  * Update CP0_Compare to a new value and update the timeout.
693  * If @ack, atomically acknowledge any pending timer interrupt, otherwise ensure
694  * any pending timer interrupt is preserved.
695  */
696 void kvm_mips_write_compare(struct kvm_vcpu *vcpu, u32 compare, bool ack)
697 {
698 	struct mips_coproc *cop0 = vcpu->arch.cop0;
699 	int dc;
700 	u32 old_compare = kvm_read_c0_guest_compare(cop0);
701 	s32 delta = compare - old_compare;
702 	u32 cause;
703 	ktime_t now = ktime_set(0, 0); /* silence bogus GCC warning */
704 	u32 count;
705 
706 	/* if unchanged, must just be an ack */
707 	if (old_compare == compare) {
708 		if (!ack)
709 			return;
710 		kvm_mips_callbacks->dequeue_timer_int(vcpu);
711 		kvm_write_c0_guest_compare(cop0, compare);
712 		return;
713 	}
714 
715 	/*
716 	 * If guest CP0_Compare moves forward, CP0_GTOffset should be adjusted
717 	 * too to prevent guest CP0_Count hitting guest CP0_Compare.
718 	 *
719 	 * The new GTOffset corresponds to the new value of CP0_Compare, and is
720 	 * set prior to it being written into the guest context. We disable
721 	 * preemption until the new value is written to prevent restore of a
722 	 * GTOffset corresponding to the old CP0_Compare value.
723 	 */
724 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ) && delta > 0) {
725 		preempt_disable();
726 		write_c0_gtoffset(compare - read_c0_count());
727 		back_to_back_c0_hazard();
728 	}
729 
730 	/* freeze_hrtimer() takes care of timer interrupts <= count */
731 	dc = kvm_mips_count_disabled(vcpu);
732 	if (!dc)
733 		now = kvm_mips_freeze_hrtimer(vcpu, &count);
734 
735 	if (ack)
736 		kvm_mips_callbacks->dequeue_timer_int(vcpu);
737 	else if (IS_ENABLED(CONFIG_KVM_MIPS_VZ))
738 		/*
739 		 * With VZ, writing CP0_Compare acks (clears) CP0_Cause.TI, so
740 		 * preserve guest CP0_Cause.TI if we don't want to ack it.
741 		 */
742 		cause = kvm_read_c0_guest_cause(cop0);
743 
744 	kvm_write_c0_guest_compare(cop0, compare);
745 
746 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
747 		if (delta > 0)
748 			preempt_enable();
749 
750 		back_to_back_c0_hazard();
751 
752 		if (!ack && cause & CAUSEF_TI)
753 			kvm_write_c0_guest_cause(cop0, cause);
754 	}
755 
756 	/* resume_hrtimer() takes care of timer interrupts > count */
757 	if (!dc)
758 		kvm_mips_resume_hrtimer(vcpu, now, count);
759 
760 	/*
761 	 * If guest CP0_Compare is moving backward, we delay CP0_GTOffset change
762 	 * until after the new CP0_Compare is written, otherwise new guest
763 	 * CP0_Count could hit new guest CP0_Compare.
764 	 */
765 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ) && delta <= 0)
766 		write_c0_gtoffset(compare - read_c0_count());
767 }
768 
769 /**
770  * kvm_mips_count_disable() - Disable count.
771  * @vcpu:	Virtual CPU.
772  *
773  * Disable the CP0_Count timer. A timer interrupt on or before the final stop
774  * time will be handled but not after.
775  *
776  * Assumes CP0_Count was previously enabled but now Guest.CP0_Cause.DC or
777  * count_ctl.DC has been set (count disabled).
778  *
779  * Returns:	The time that the timer was stopped.
780  */
781 static ktime_t kvm_mips_count_disable(struct kvm_vcpu *vcpu)
782 {
783 	struct mips_coproc *cop0 = vcpu->arch.cop0;
784 	u32 count;
785 	ktime_t now;
786 
787 	/* Stop hrtimer */
788 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
789 
790 	/* Set the static count from the dynamic count, handling pending TI */
791 	now = ktime_get();
792 	count = kvm_mips_read_count_running(vcpu, now);
793 	kvm_write_c0_guest_count(cop0, count);
794 
795 	return now;
796 }
797 
798 /**
799  * kvm_mips_count_disable_cause() - Disable count using CP0_Cause.DC.
800  * @vcpu:	Virtual CPU.
801  *
802  * Disable the CP0_Count timer and set CP0_Cause.DC. A timer interrupt on or
803  * before the final stop time will be handled if the timer isn't disabled by
804  * count_ctl.DC, but not after.
805  *
806  * Assumes CP0_Cause.DC is clear (count enabled).
807  */
808 void kvm_mips_count_disable_cause(struct kvm_vcpu *vcpu)
809 {
810 	struct mips_coproc *cop0 = vcpu->arch.cop0;
811 
812 	kvm_set_c0_guest_cause(cop0, CAUSEF_DC);
813 	if (!(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC))
814 		kvm_mips_count_disable(vcpu);
815 }
816 
817 /**
818  * kvm_mips_count_enable_cause() - Enable count using CP0_Cause.DC.
819  * @vcpu:	Virtual CPU.
820  *
821  * Enable the CP0_Count timer and clear CP0_Cause.DC. A timer interrupt after
822  * the start time will be handled if the timer isn't disabled by count_ctl.DC,
823  * potentially before even returning, so the caller should be careful with
824  * ordering of CP0_Cause modifications so as not to lose it.
825  *
826  * Assumes CP0_Cause.DC is set (count disabled).
827  */
828 void kvm_mips_count_enable_cause(struct kvm_vcpu *vcpu)
829 {
830 	struct mips_coproc *cop0 = vcpu->arch.cop0;
831 	u32 count;
832 
833 	kvm_clear_c0_guest_cause(cop0, CAUSEF_DC);
834 
835 	/*
836 	 * Set the dynamic count to match the static count.
837 	 * This starts the hrtimer if count_ctl.DC allows it.
838 	 * Otherwise it conveniently updates the biases.
839 	 */
840 	count = kvm_read_c0_guest_count(cop0);
841 	kvm_mips_write_count(vcpu, count);
842 }
843 
844 /**
845  * kvm_mips_set_count_ctl() - Update the count control KVM register.
846  * @vcpu:	Virtual CPU.
847  * @count_ctl:	Count control register new value.
848  *
849  * Set the count control KVM register. The timer is updated accordingly.
850  *
851  * Returns:	-EINVAL if reserved bits are set.
852  *		0 on success.
853  */
854 int kvm_mips_set_count_ctl(struct kvm_vcpu *vcpu, s64 count_ctl)
855 {
856 	struct mips_coproc *cop0 = vcpu->arch.cop0;
857 	s64 changed = count_ctl ^ vcpu->arch.count_ctl;
858 	s64 delta;
859 	ktime_t expire, now;
860 	u32 count, compare;
861 
862 	/* Only allow defined bits to be changed */
863 	if (changed & ~(s64)(KVM_REG_MIPS_COUNT_CTL_DC))
864 		return -EINVAL;
865 
866 	/* Apply new value */
867 	vcpu->arch.count_ctl = count_ctl;
868 
869 	/* Master CP0_Count disable */
870 	if (changed & KVM_REG_MIPS_COUNT_CTL_DC) {
871 		/* Is CP0_Cause.DC already disabling CP0_Count? */
872 		if (kvm_read_c0_guest_cause(cop0) & CAUSEF_DC) {
873 			if (count_ctl & KVM_REG_MIPS_COUNT_CTL_DC)
874 				/* Just record the current time */
875 				vcpu->arch.count_resume = ktime_get();
876 		} else if (count_ctl & KVM_REG_MIPS_COUNT_CTL_DC) {
877 			/* disable timer and record current time */
878 			vcpu->arch.count_resume = kvm_mips_count_disable(vcpu);
879 		} else {
880 			/*
881 			 * Calculate timeout relative to static count at resume
882 			 * time (wrap 0 to 2^32).
883 			 */
884 			count = kvm_read_c0_guest_count(cop0);
885 			compare = kvm_read_c0_guest_compare(cop0);
886 			delta = (u64)(u32)(compare - count - 1) + 1;
887 			delta = div_u64(delta * NSEC_PER_SEC,
888 					vcpu->arch.count_hz);
889 			expire = ktime_add_ns(vcpu->arch.count_resume, delta);
890 
891 			/* Handle pending interrupt */
892 			now = ktime_get();
893 			if (ktime_compare(now, expire) >= 0)
894 				/* Nothing should be waiting on the timeout */
895 				kvm_mips_callbacks->queue_timer_int(vcpu);
896 
897 			/* Resume hrtimer without changing bias */
898 			count = kvm_mips_read_count_running(vcpu, now);
899 			kvm_mips_resume_hrtimer(vcpu, now, count);
900 		}
901 	}
902 
903 	return 0;
904 }
905 
906 /**
907  * kvm_mips_set_count_resume() - Update the count resume KVM register.
908  * @vcpu:		Virtual CPU.
909  * @count_resume:	Count resume register new value.
910  *
911  * Set the count resume KVM register.
912  *
913  * Returns:	-EINVAL if out of valid range (0..now).
914  *		0 on success.
915  */
916 int kvm_mips_set_count_resume(struct kvm_vcpu *vcpu, s64 count_resume)
917 {
918 	/*
919 	 * It doesn't make sense for the resume time to be in the future, as it
920 	 * would be possible for the next interrupt to be more than a full
921 	 * period in the future.
922 	 */
923 	if (count_resume < 0 || count_resume > ktime_to_ns(ktime_get()))
924 		return -EINVAL;
925 
926 	vcpu->arch.count_resume = ns_to_ktime(count_resume);
927 	return 0;
928 }
929 
930 /**
931  * kvm_mips_count_timeout() - Push timer forward on timeout.
932  * @vcpu:	Virtual CPU.
933  *
934  * Handle an hrtimer event by push the hrtimer forward a period.
935  *
936  * Returns:	The hrtimer_restart value to return to the hrtimer subsystem.
937  */
938 enum hrtimer_restart kvm_mips_count_timeout(struct kvm_vcpu *vcpu)
939 {
940 	/* Add the Count period to the current expiry time */
941 	hrtimer_add_expires_ns(&vcpu->arch.comparecount_timer,
942 			       vcpu->arch.count_period);
943 	return HRTIMER_RESTART;
944 }
945 
946 enum emulation_result kvm_mips_emul_eret(struct kvm_vcpu *vcpu)
947 {
948 	struct mips_coproc *cop0 = vcpu->arch.cop0;
949 	enum emulation_result er = EMULATE_DONE;
950 
951 	if (kvm_read_c0_guest_status(cop0) & ST0_ERL) {
952 		kvm_clear_c0_guest_status(cop0, ST0_ERL);
953 		vcpu->arch.pc = kvm_read_c0_guest_errorepc(cop0);
954 	} else if (kvm_read_c0_guest_status(cop0) & ST0_EXL) {
955 		kvm_debug("[%#lx] ERET to %#lx\n", vcpu->arch.pc,
956 			  kvm_read_c0_guest_epc(cop0));
957 		kvm_clear_c0_guest_status(cop0, ST0_EXL);
958 		vcpu->arch.pc = kvm_read_c0_guest_epc(cop0);
959 
960 	} else {
961 		kvm_err("[%#lx] ERET when MIPS_SR_EXL|MIPS_SR_ERL == 0\n",
962 			vcpu->arch.pc);
963 		er = EMULATE_FAIL;
964 	}
965 
966 	return er;
967 }
968 
969 enum emulation_result kvm_mips_emul_wait(struct kvm_vcpu *vcpu)
970 {
971 	kvm_debug("[%#lx] !!!WAIT!!! (%#lx)\n", vcpu->arch.pc,
972 		  vcpu->arch.pending_exceptions);
973 
974 	++vcpu->stat.wait_exits;
975 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_WAIT);
976 	if (!vcpu->arch.pending_exceptions) {
977 		kvm_vz_lose_htimer(vcpu);
978 		vcpu->arch.wait = 1;
979 		kvm_vcpu_block(vcpu);
980 
981 		/*
982 		 * We we are runnable, then definitely go off to user space to
983 		 * check if any I/O interrupts are pending.
984 		 */
985 		if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
986 			kvm_clear_request(KVM_REQ_UNHALT, vcpu);
987 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
988 		}
989 	}
990 
991 	return EMULATE_DONE;
992 }
993 
994 static void kvm_mips_change_entryhi(struct kvm_vcpu *vcpu,
995 				    unsigned long entryhi)
996 {
997 	struct mips_coproc *cop0 = vcpu->arch.cop0;
998 	struct mm_struct *kern_mm = &vcpu->arch.guest_kernel_mm;
999 	int cpu, i;
1000 	u32 nasid = entryhi & KVM_ENTRYHI_ASID;
1001 
1002 	if (((kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID) != nasid)) {
1003 		trace_kvm_asid_change(vcpu, kvm_read_c0_guest_entryhi(cop0) &
1004 				      KVM_ENTRYHI_ASID, nasid);
1005 
1006 		/*
1007 		 * Flush entries from the GVA page tables.
1008 		 * Guest user page table will get flushed lazily on re-entry to
1009 		 * guest user if the guest ASID actually changes.
1010 		 */
1011 		kvm_mips_flush_gva_pt(kern_mm->pgd, KMF_KERN);
1012 
1013 		/*
1014 		 * Regenerate/invalidate kernel MMU context.
1015 		 * The user MMU context will be regenerated lazily on re-entry
1016 		 * to guest user if the guest ASID actually changes.
1017 		 */
1018 		preempt_disable();
1019 		cpu = smp_processor_id();
1020 		get_new_mmu_context(kern_mm);
1021 		for_each_possible_cpu(i)
1022 			if (i != cpu)
1023 				set_cpu_context(i, kern_mm, 0);
1024 		preempt_enable();
1025 	}
1026 	kvm_write_c0_guest_entryhi(cop0, entryhi);
1027 }
1028 
1029 enum emulation_result kvm_mips_emul_tlbr(struct kvm_vcpu *vcpu)
1030 {
1031 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1032 	struct kvm_mips_tlb *tlb;
1033 	unsigned long pc = vcpu->arch.pc;
1034 	int index;
1035 
1036 	index = kvm_read_c0_guest_index(cop0);
1037 	if (index < 0 || index >= KVM_MIPS_GUEST_TLB_SIZE) {
1038 		/* UNDEFINED */
1039 		kvm_debug("[%#lx] TLBR Index %#x out of range\n", pc, index);
1040 		index &= KVM_MIPS_GUEST_TLB_SIZE - 1;
1041 	}
1042 
1043 	tlb = &vcpu->arch.guest_tlb[index];
1044 	kvm_write_c0_guest_pagemask(cop0, tlb->tlb_mask);
1045 	kvm_write_c0_guest_entrylo0(cop0, tlb->tlb_lo[0]);
1046 	kvm_write_c0_guest_entrylo1(cop0, tlb->tlb_lo[1]);
1047 	kvm_mips_change_entryhi(vcpu, tlb->tlb_hi);
1048 
1049 	return EMULATE_DONE;
1050 }
1051 
1052 /**
1053  * kvm_mips_invalidate_guest_tlb() - Indicates a change in guest MMU map.
1054  * @vcpu:	VCPU with changed mappings.
1055  * @tlb:	TLB entry being removed.
1056  *
1057  * This is called to indicate a single change in guest MMU mappings, so that we
1058  * can arrange TLB flushes on this and other CPUs.
1059  */
1060 static void kvm_mips_invalidate_guest_tlb(struct kvm_vcpu *vcpu,
1061 					  struct kvm_mips_tlb *tlb)
1062 {
1063 	struct mm_struct *kern_mm = &vcpu->arch.guest_kernel_mm;
1064 	struct mm_struct *user_mm = &vcpu->arch.guest_user_mm;
1065 	int cpu, i;
1066 	bool user;
1067 
1068 	/* No need to flush for entries which are already invalid */
1069 	if (!((tlb->tlb_lo[0] | tlb->tlb_lo[1]) & ENTRYLO_V))
1070 		return;
1071 	/* Don't touch host kernel page tables or TLB mappings */
1072 	if ((unsigned long)tlb->tlb_hi > 0x7fffffff)
1073 		return;
1074 	/* User address space doesn't need flushing for KSeg2/3 changes */
1075 	user = tlb->tlb_hi < KVM_GUEST_KSEG0;
1076 
1077 	preempt_disable();
1078 
1079 	/* Invalidate page table entries */
1080 	kvm_trap_emul_invalidate_gva(vcpu, tlb->tlb_hi & VPN2_MASK, user);
1081 
1082 	/*
1083 	 * Probe the shadow host TLB for the entry being overwritten, if one
1084 	 * matches, invalidate it
1085 	 */
1086 	kvm_mips_host_tlb_inv(vcpu, tlb->tlb_hi, user, true);
1087 
1088 	/* Invalidate the whole ASID on other CPUs */
1089 	cpu = smp_processor_id();
1090 	for_each_possible_cpu(i) {
1091 		if (i == cpu)
1092 			continue;
1093 		if (user)
1094 			set_cpu_context(i, user_mm, 0);
1095 		set_cpu_context(i, kern_mm, 0);
1096 	}
1097 
1098 	preempt_enable();
1099 }
1100 
1101 /* Write Guest TLB Entry @ Index */
1102 enum emulation_result kvm_mips_emul_tlbwi(struct kvm_vcpu *vcpu)
1103 {
1104 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1105 	int index = kvm_read_c0_guest_index(cop0);
1106 	struct kvm_mips_tlb *tlb = NULL;
1107 	unsigned long pc = vcpu->arch.pc;
1108 
1109 	if (index < 0 || index >= KVM_MIPS_GUEST_TLB_SIZE) {
1110 		kvm_debug("%s: illegal index: %d\n", __func__, index);
1111 		kvm_debug("[%#lx] COP0_TLBWI [%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx, mask: %#lx)\n",
1112 			  pc, index, kvm_read_c0_guest_entryhi(cop0),
1113 			  kvm_read_c0_guest_entrylo0(cop0),
1114 			  kvm_read_c0_guest_entrylo1(cop0),
1115 			  kvm_read_c0_guest_pagemask(cop0));
1116 		index = (index & ~0x80000000) % KVM_MIPS_GUEST_TLB_SIZE;
1117 	}
1118 
1119 	tlb = &vcpu->arch.guest_tlb[index];
1120 
1121 	kvm_mips_invalidate_guest_tlb(vcpu, tlb);
1122 
1123 	tlb->tlb_mask = kvm_read_c0_guest_pagemask(cop0);
1124 	tlb->tlb_hi = kvm_read_c0_guest_entryhi(cop0);
1125 	tlb->tlb_lo[0] = kvm_read_c0_guest_entrylo0(cop0);
1126 	tlb->tlb_lo[1] = kvm_read_c0_guest_entrylo1(cop0);
1127 
1128 	kvm_debug("[%#lx] COP0_TLBWI [%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx, mask: %#lx)\n",
1129 		  pc, index, kvm_read_c0_guest_entryhi(cop0),
1130 		  kvm_read_c0_guest_entrylo0(cop0),
1131 		  kvm_read_c0_guest_entrylo1(cop0),
1132 		  kvm_read_c0_guest_pagemask(cop0));
1133 
1134 	return EMULATE_DONE;
1135 }
1136 
1137 /* Write Guest TLB Entry @ Random Index */
1138 enum emulation_result kvm_mips_emul_tlbwr(struct kvm_vcpu *vcpu)
1139 {
1140 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1141 	struct kvm_mips_tlb *tlb = NULL;
1142 	unsigned long pc = vcpu->arch.pc;
1143 	int index;
1144 
1145 	index = prandom_u32_max(KVM_MIPS_GUEST_TLB_SIZE);
1146 	tlb = &vcpu->arch.guest_tlb[index];
1147 
1148 	kvm_mips_invalidate_guest_tlb(vcpu, tlb);
1149 
1150 	tlb->tlb_mask = kvm_read_c0_guest_pagemask(cop0);
1151 	tlb->tlb_hi = kvm_read_c0_guest_entryhi(cop0);
1152 	tlb->tlb_lo[0] = kvm_read_c0_guest_entrylo0(cop0);
1153 	tlb->tlb_lo[1] = kvm_read_c0_guest_entrylo1(cop0);
1154 
1155 	kvm_debug("[%#lx] COP0_TLBWR[%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx)\n",
1156 		  pc, index, kvm_read_c0_guest_entryhi(cop0),
1157 		  kvm_read_c0_guest_entrylo0(cop0),
1158 		  kvm_read_c0_guest_entrylo1(cop0));
1159 
1160 	return EMULATE_DONE;
1161 }
1162 
1163 enum emulation_result kvm_mips_emul_tlbp(struct kvm_vcpu *vcpu)
1164 {
1165 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1166 	long entryhi = kvm_read_c0_guest_entryhi(cop0);
1167 	unsigned long pc = vcpu->arch.pc;
1168 	int index = -1;
1169 
1170 	index = kvm_mips_guest_tlb_lookup(vcpu, entryhi);
1171 
1172 	kvm_write_c0_guest_index(cop0, index);
1173 
1174 	kvm_debug("[%#lx] COP0_TLBP (entryhi: %#lx), index: %d\n", pc, entryhi,
1175 		  index);
1176 
1177 	return EMULATE_DONE;
1178 }
1179 
1180 /**
1181  * kvm_mips_config1_wrmask() - Find mask of writable bits in guest Config1
1182  * @vcpu:	Virtual CPU.
1183  *
1184  * Finds the mask of bits which are writable in the guest's Config1 CP0
1185  * register, by userland (currently read-only to the guest).
1186  */
1187 unsigned int kvm_mips_config1_wrmask(struct kvm_vcpu *vcpu)
1188 {
1189 	unsigned int mask = 0;
1190 
1191 	/* Permit FPU to be present if FPU is supported */
1192 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch))
1193 		mask |= MIPS_CONF1_FP;
1194 
1195 	return mask;
1196 }
1197 
1198 /**
1199  * kvm_mips_config3_wrmask() - Find mask of writable bits in guest Config3
1200  * @vcpu:	Virtual CPU.
1201  *
1202  * Finds the mask of bits which are writable in the guest's Config3 CP0
1203  * register, by userland (currently read-only to the guest).
1204  */
1205 unsigned int kvm_mips_config3_wrmask(struct kvm_vcpu *vcpu)
1206 {
1207 	/* Config4 and ULRI are optional */
1208 	unsigned int mask = MIPS_CONF_M | MIPS_CONF3_ULRI;
1209 
1210 	/* Permit MSA to be present if MSA is supported */
1211 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
1212 		mask |= MIPS_CONF3_MSA;
1213 
1214 	return mask;
1215 }
1216 
1217 /**
1218  * kvm_mips_config4_wrmask() - Find mask of writable bits in guest Config4
1219  * @vcpu:	Virtual CPU.
1220  *
1221  * Finds the mask of bits which are writable in the guest's Config4 CP0
1222  * register, by userland (currently read-only to the guest).
1223  */
1224 unsigned int kvm_mips_config4_wrmask(struct kvm_vcpu *vcpu)
1225 {
1226 	/* Config5 is optional */
1227 	unsigned int mask = MIPS_CONF_M;
1228 
1229 	/* KScrExist */
1230 	mask |= 0xfc << MIPS_CONF4_KSCREXIST_SHIFT;
1231 
1232 	return mask;
1233 }
1234 
1235 /**
1236  * kvm_mips_config5_wrmask() - Find mask of writable bits in guest Config5
1237  * @vcpu:	Virtual CPU.
1238  *
1239  * Finds the mask of bits which are writable in the guest's Config5 CP0
1240  * register, by the guest itself.
1241  */
1242 unsigned int kvm_mips_config5_wrmask(struct kvm_vcpu *vcpu)
1243 {
1244 	unsigned int mask = 0;
1245 
1246 	/* Permit MSAEn changes if MSA supported and enabled */
1247 	if (kvm_mips_guest_has_msa(&vcpu->arch))
1248 		mask |= MIPS_CONF5_MSAEN;
1249 
1250 	/*
1251 	 * Permit guest FPU mode changes if FPU is enabled and the relevant
1252 	 * feature exists according to FIR register.
1253 	 */
1254 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1255 		if (cpu_has_fre)
1256 			mask |= MIPS_CONF5_FRE;
1257 		/* We don't support UFR or UFE */
1258 	}
1259 
1260 	return mask;
1261 }
1262 
1263 enum emulation_result kvm_mips_emulate_CP0(union mips_instruction inst,
1264 					   u32 *opc, u32 cause,
1265 					   struct kvm_run *run,
1266 					   struct kvm_vcpu *vcpu)
1267 {
1268 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1269 	enum emulation_result er = EMULATE_DONE;
1270 	u32 rt, rd, sel;
1271 	unsigned long curr_pc;
1272 
1273 	/*
1274 	 * Update PC and hold onto current PC in case there is
1275 	 * an error and we want to rollback the PC
1276 	 */
1277 	curr_pc = vcpu->arch.pc;
1278 	er = update_pc(vcpu, cause);
1279 	if (er == EMULATE_FAIL)
1280 		return er;
1281 
1282 	if (inst.co_format.co) {
1283 		switch (inst.co_format.func) {
1284 		case tlbr_op:	/*  Read indexed TLB entry  */
1285 			er = kvm_mips_emul_tlbr(vcpu);
1286 			break;
1287 		case tlbwi_op:	/*  Write indexed  */
1288 			er = kvm_mips_emul_tlbwi(vcpu);
1289 			break;
1290 		case tlbwr_op:	/*  Write random  */
1291 			er = kvm_mips_emul_tlbwr(vcpu);
1292 			break;
1293 		case tlbp_op:	/* TLB Probe */
1294 			er = kvm_mips_emul_tlbp(vcpu);
1295 			break;
1296 		case rfe_op:
1297 			kvm_err("!!!COP0_RFE!!!\n");
1298 			break;
1299 		case eret_op:
1300 			er = kvm_mips_emul_eret(vcpu);
1301 			goto dont_update_pc;
1302 		case wait_op:
1303 			er = kvm_mips_emul_wait(vcpu);
1304 			break;
1305 		case hypcall_op:
1306 			er = kvm_mips_emul_hypcall(vcpu, inst);
1307 			break;
1308 		}
1309 	} else {
1310 		rt = inst.c0r_format.rt;
1311 		rd = inst.c0r_format.rd;
1312 		sel = inst.c0r_format.sel;
1313 
1314 		switch (inst.c0r_format.rs) {
1315 		case mfc_op:
1316 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
1317 			cop0->stat[rd][sel]++;
1318 #endif
1319 			/* Get reg */
1320 			if ((rd == MIPS_CP0_COUNT) && (sel == 0)) {
1321 				vcpu->arch.gprs[rt] =
1322 				    (s32)kvm_mips_read_count(vcpu);
1323 			} else if ((rd == MIPS_CP0_ERRCTL) && (sel == 0)) {
1324 				vcpu->arch.gprs[rt] = 0x0;
1325 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1326 				kvm_mips_trans_mfc0(inst, opc, vcpu);
1327 #endif
1328 			} else {
1329 				vcpu->arch.gprs[rt] = (s32)cop0->reg[rd][sel];
1330 
1331 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1332 				kvm_mips_trans_mfc0(inst, opc, vcpu);
1333 #endif
1334 			}
1335 
1336 			trace_kvm_hwr(vcpu, KVM_TRACE_MFC0,
1337 				      KVM_TRACE_COP0(rd, sel),
1338 				      vcpu->arch.gprs[rt]);
1339 			break;
1340 
1341 		case dmfc_op:
1342 			vcpu->arch.gprs[rt] = cop0->reg[rd][sel];
1343 
1344 			trace_kvm_hwr(vcpu, KVM_TRACE_DMFC0,
1345 				      KVM_TRACE_COP0(rd, sel),
1346 				      vcpu->arch.gprs[rt]);
1347 			break;
1348 
1349 		case mtc_op:
1350 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
1351 			cop0->stat[rd][sel]++;
1352 #endif
1353 			trace_kvm_hwr(vcpu, KVM_TRACE_MTC0,
1354 				      KVM_TRACE_COP0(rd, sel),
1355 				      vcpu->arch.gprs[rt]);
1356 
1357 			if ((rd == MIPS_CP0_TLB_INDEX)
1358 			    && (vcpu->arch.gprs[rt] >=
1359 				KVM_MIPS_GUEST_TLB_SIZE)) {
1360 				kvm_err("Invalid TLB Index: %ld",
1361 					vcpu->arch.gprs[rt]);
1362 				er = EMULATE_FAIL;
1363 				break;
1364 			}
1365 			if ((rd == MIPS_CP0_PRID) && (sel == 1)) {
1366 				/*
1367 				 * Preserve core number, and keep the exception
1368 				 * base in guest KSeg0.
1369 				 */
1370 				kvm_change_c0_guest_ebase(cop0, 0x1ffff000,
1371 							  vcpu->arch.gprs[rt]);
1372 			} else if (rd == MIPS_CP0_TLB_HI && sel == 0) {
1373 				kvm_mips_change_entryhi(vcpu,
1374 							vcpu->arch.gprs[rt]);
1375 			}
1376 			/* Are we writing to COUNT */
1377 			else if ((rd == MIPS_CP0_COUNT) && (sel == 0)) {
1378 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
1379 				goto done;
1380 			} else if ((rd == MIPS_CP0_COMPARE) && (sel == 0)) {
1381 				/* If we are writing to COMPARE */
1382 				/* Clear pending timer interrupt, if any */
1383 				kvm_mips_write_compare(vcpu,
1384 						       vcpu->arch.gprs[rt],
1385 						       true);
1386 			} else if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1387 				unsigned int old_val, val, change;
1388 
1389 				old_val = kvm_read_c0_guest_status(cop0);
1390 				val = vcpu->arch.gprs[rt];
1391 				change = val ^ old_val;
1392 
1393 				/* Make sure that the NMI bit is never set */
1394 				val &= ~ST0_NMI;
1395 
1396 				/*
1397 				 * Don't allow CU1 or FR to be set unless FPU
1398 				 * capability enabled and exists in guest
1399 				 * configuration.
1400 				 */
1401 				if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1402 					val &= ~(ST0_CU1 | ST0_FR);
1403 
1404 				/*
1405 				 * Also don't allow FR to be set if host doesn't
1406 				 * support it.
1407 				 */
1408 				if (!(current_cpu_data.fpu_id & MIPS_FPIR_F64))
1409 					val &= ~ST0_FR;
1410 
1411 
1412 				/* Handle changes in FPU mode */
1413 				preempt_disable();
1414 
1415 				/*
1416 				 * FPU and Vector register state is made
1417 				 * UNPREDICTABLE by a change of FR, so don't
1418 				 * even bother saving it.
1419 				 */
1420 				if (change & ST0_FR)
1421 					kvm_drop_fpu(vcpu);
1422 
1423 				/*
1424 				 * If MSA state is already live, it is undefined
1425 				 * how it interacts with FR=0 FPU state, and we
1426 				 * don't want to hit reserved instruction
1427 				 * exceptions trying to save the MSA state later
1428 				 * when CU=1 && FR=1, so play it safe and save
1429 				 * it first.
1430 				 */
1431 				if (change & ST0_CU1 && !(val & ST0_FR) &&
1432 				    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1433 					kvm_lose_fpu(vcpu);
1434 
1435 				/*
1436 				 * Propagate CU1 (FPU enable) changes
1437 				 * immediately if the FPU context is already
1438 				 * loaded. When disabling we leave the context
1439 				 * loaded so it can be quickly enabled again in
1440 				 * the near future.
1441 				 */
1442 				if (change & ST0_CU1 &&
1443 				    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1444 					change_c0_status(ST0_CU1, val);
1445 
1446 				preempt_enable();
1447 
1448 				kvm_write_c0_guest_status(cop0, val);
1449 
1450 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1451 				/*
1452 				 * If FPU present, we need CU1/FR bits to take
1453 				 * effect fairly soon.
1454 				 */
1455 				if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1456 					kvm_mips_trans_mtc0(inst, opc, vcpu);
1457 #endif
1458 			} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1459 				unsigned int old_val, val, change, wrmask;
1460 
1461 				old_val = kvm_read_c0_guest_config5(cop0);
1462 				val = vcpu->arch.gprs[rt];
1463 
1464 				/* Only a few bits are writable in Config5 */
1465 				wrmask = kvm_mips_config5_wrmask(vcpu);
1466 				change = (val ^ old_val) & wrmask;
1467 				val = old_val ^ change;
1468 
1469 
1470 				/* Handle changes in FPU/MSA modes */
1471 				preempt_disable();
1472 
1473 				/*
1474 				 * Propagate FRE changes immediately if the FPU
1475 				 * context is already loaded.
1476 				 */
1477 				if (change & MIPS_CONF5_FRE &&
1478 				    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1479 					change_c0_config5(MIPS_CONF5_FRE, val);
1480 
1481 				/*
1482 				 * Propagate MSAEn changes immediately if the
1483 				 * MSA context is already loaded. When disabling
1484 				 * we leave the context loaded so it can be
1485 				 * quickly enabled again in the near future.
1486 				 */
1487 				if (change & MIPS_CONF5_MSAEN &&
1488 				    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1489 					change_c0_config5(MIPS_CONF5_MSAEN,
1490 							  val);
1491 
1492 				preempt_enable();
1493 
1494 				kvm_write_c0_guest_config5(cop0, val);
1495 			} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1496 				u32 old_cause, new_cause;
1497 
1498 				old_cause = kvm_read_c0_guest_cause(cop0);
1499 				new_cause = vcpu->arch.gprs[rt];
1500 				/* Update R/W bits */
1501 				kvm_change_c0_guest_cause(cop0, 0x08800300,
1502 							  new_cause);
1503 				/* DC bit enabling/disabling timer? */
1504 				if ((old_cause ^ new_cause) & CAUSEF_DC) {
1505 					if (new_cause & CAUSEF_DC)
1506 						kvm_mips_count_disable_cause(vcpu);
1507 					else
1508 						kvm_mips_count_enable_cause(vcpu);
1509 				}
1510 			} else if ((rd == MIPS_CP0_HWRENA) && (sel == 0)) {
1511 				u32 mask = MIPS_HWRENA_CPUNUM |
1512 					   MIPS_HWRENA_SYNCISTEP |
1513 					   MIPS_HWRENA_CC |
1514 					   MIPS_HWRENA_CCRES;
1515 
1516 				if (kvm_read_c0_guest_config3(cop0) &
1517 				    MIPS_CONF3_ULRI)
1518 					mask |= MIPS_HWRENA_ULR;
1519 				cop0->reg[rd][sel] = vcpu->arch.gprs[rt] & mask;
1520 			} else {
1521 				cop0->reg[rd][sel] = vcpu->arch.gprs[rt];
1522 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1523 				kvm_mips_trans_mtc0(inst, opc, vcpu);
1524 #endif
1525 			}
1526 			break;
1527 
1528 		case dmtc_op:
1529 			kvm_err("!!!!!!![%#lx]dmtc_op: rt: %d, rd: %d, sel: %d!!!!!!\n",
1530 				vcpu->arch.pc, rt, rd, sel);
1531 			trace_kvm_hwr(vcpu, KVM_TRACE_DMTC0,
1532 				      KVM_TRACE_COP0(rd, sel),
1533 				      vcpu->arch.gprs[rt]);
1534 			er = EMULATE_FAIL;
1535 			break;
1536 
1537 		case mfmc0_op:
1538 #ifdef KVM_MIPS_DEBUG_COP0_COUNTERS
1539 			cop0->stat[MIPS_CP0_STATUS][0]++;
1540 #endif
1541 			if (rt != 0)
1542 				vcpu->arch.gprs[rt] =
1543 				    kvm_read_c0_guest_status(cop0);
1544 			/* EI */
1545 			if (inst.mfmc0_format.sc) {
1546 				kvm_debug("[%#lx] mfmc0_op: EI\n",
1547 					  vcpu->arch.pc);
1548 				kvm_set_c0_guest_status(cop0, ST0_IE);
1549 			} else {
1550 				kvm_debug("[%#lx] mfmc0_op: DI\n",
1551 					  vcpu->arch.pc);
1552 				kvm_clear_c0_guest_status(cop0, ST0_IE);
1553 			}
1554 
1555 			break;
1556 
1557 		case wrpgpr_op:
1558 			{
1559 				u32 css = cop0->reg[MIPS_CP0_STATUS][2] & 0xf;
1560 				u32 pss =
1561 				    (cop0->reg[MIPS_CP0_STATUS][2] >> 6) & 0xf;
1562 				/*
1563 				 * We don't support any shadow register sets, so
1564 				 * SRSCtl[PSS] == SRSCtl[CSS] = 0
1565 				 */
1566 				if (css || pss) {
1567 					er = EMULATE_FAIL;
1568 					break;
1569 				}
1570 				kvm_debug("WRPGPR[%d][%d] = %#lx\n", pss, rd,
1571 					  vcpu->arch.gprs[rt]);
1572 				vcpu->arch.gprs[rd] = vcpu->arch.gprs[rt];
1573 			}
1574 			break;
1575 		default:
1576 			kvm_err("[%#lx]MachEmulateCP0: unsupported COP0, copz: 0x%x\n",
1577 				vcpu->arch.pc, inst.c0r_format.rs);
1578 			er = EMULATE_FAIL;
1579 			break;
1580 		}
1581 	}
1582 
1583 done:
1584 	/* Rollback PC only if emulation was unsuccessful */
1585 	if (er == EMULATE_FAIL)
1586 		vcpu->arch.pc = curr_pc;
1587 
1588 dont_update_pc:
1589 	/*
1590 	 * This is for special instructions whose emulation
1591 	 * updates the PC, so do not overwrite the PC under
1592 	 * any circumstances
1593 	 */
1594 
1595 	return er;
1596 }
1597 
1598 enum emulation_result kvm_mips_emulate_store(union mips_instruction inst,
1599 					     u32 cause,
1600 					     struct kvm_run *run,
1601 					     struct kvm_vcpu *vcpu)
1602 {
1603 	enum emulation_result er;
1604 	u32 rt;
1605 	void *data = run->mmio.data;
1606 	unsigned long curr_pc;
1607 
1608 	/*
1609 	 * Update PC and hold onto current PC in case there is
1610 	 * an error and we want to rollback the PC
1611 	 */
1612 	curr_pc = vcpu->arch.pc;
1613 	er = update_pc(vcpu, cause);
1614 	if (er == EMULATE_FAIL)
1615 		return er;
1616 
1617 	rt = inst.i_format.rt;
1618 
1619 	run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1620 						vcpu->arch.host_cp0_badvaddr);
1621 	if (run->mmio.phys_addr == KVM_INVALID_ADDR)
1622 		goto out_fail;
1623 
1624 	switch (inst.i_format.opcode) {
1625 #if defined(CONFIG_64BIT) && defined(CONFIG_KVM_MIPS_VZ)
1626 	case sd_op:
1627 		run->mmio.len = 8;
1628 		*(u64 *)data = vcpu->arch.gprs[rt];
1629 
1630 		kvm_debug("[%#lx] OP_SD: eaddr: %#lx, gpr: %#lx, data: %#llx\n",
1631 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1632 			  vcpu->arch.gprs[rt], *(u64 *)data);
1633 		break;
1634 #endif
1635 
1636 	case sw_op:
1637 		run->mmio.len = 4;
1638 		*(u32 *)data = vcpu->arch.gprs[rt];
1639 
1640 		kvm_debug("[%#lx] OP_SW: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1641 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1642 			  vcpu->arch.gprs[rt], *(u32 *)data);
1643 		break;
1644 
1645 	case sh_op:
1646 		run->mmio.len = 2;
1647 		*(u16 *)data = vcpu->arch.gprs[rt];
1648 
1649 		kvm_debug("[%#lx] OP_SH: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1650 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1651 			  vcpu->arch.gprs[rt], *(u16 *)data);
1652 		break;
1653 
1654 	case sb_op:
1655 		run->mmio.len = 1;
1656 		*(u8 *)data = vcpu->arch.gprs[rt];
1657 
1658 		kvm_debug("[%#lx] OP_SB: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1659 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1660 			  vcpu->arch.gprs[rt], *(u8 *)data);
1661 		break;
1662 
1663 	default:
1664 		kvm_err("Store not yet supported (inst=0x%08x)\n",
1665 			inst.word);
1666 		goto out_fail;
1667 	}
1668 
1669 	run->mmio.is_write = 1;
1670 	vcpu->mmio_needed = 1;
1671 	vcpu->mmio_is_write = 1;
1672 	return EMULATE_DO_MMIO;
1673 
1674 out_fail:
1675 	/* Rollback PC if emulation was unsuccessful */
1676 	vcpu->arch.pc = curr_pc;
1677 	return EMULATE_FAIL;
1678 }
1679 
1680 enum emulation_result kvm_mips_emulate_load(union mips_instruction inst,
1681 					    u32 cause, struct kvm_run *run,
1682 					    struct kvm_vcpu *vcpu)
1683 {
1684 	enum emulation_result er;
1685 	unsigned long curr_pc;
1686 	u32 op, rt;
1687 
1688 	rt = inst.i_format.rt;
1689 	op = inst.i_format.opcode;
1690 
1691 	/*
1692 	 * Find the resume PC now while we have safe and easy access to the
1693 	 * prior branch instruction, and save it for
1694 	 * kvm_mips_complete_mmio_load() to restore later.
1695 	 */
1696 	curr_pc = vcpu->arch.pc;
1697 	er = update_pc(vcpu, cause);
1698 	if (er == EMULATE_FAIL)
1699 		return er;
1700 	vcpu->arch.io_pc = vcpu->arch.pc;
1701 	vcpu->arch.pc = curr_pc;
1702 
1703 	vcpu->arch.io_gpr = rt;
1704 
1705 	run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1706 						vcpu->arch.host_cp0_badvaddr);
1707 	if (run->mmio.phys_addr == KVM_INVALID_ADDR)
1708 		return EMULATE_FAIL;
1709 
1710 	vcpu->mmio_needed = 2;	/* signed */
1711 	switch (op) {
1712 #if defined(CONFIG_64BIT) && defined(CONFIG_KVM_MIPS_VZ)
1713 	case ld_op:
1714 		run->mmio.len = 8;
1715 		break;
1716 
1717 	case lwu_op:
1718 		vcpu->mmio_needed = 1;	/* unsigned */
1719 		/* fall through */
1720 #endif
1721 	case lw_op:
1722 		run->mmio.len = 4;
1723 		break;
1724 
1725 	case lhu_op:
1726 		vcpu->mmio_needed = 1;	/* unsigned */
1727 		fallthrough;
1728 	case lh_op:
1729 		run->mmio.len = 2;
1730 		break;
1731 
1732 	case lbu_op:
1733 		vcpu->mmio_needed = 1;	/* unsigned */
1734 		fallthrough;
1735 	case lb_op:
1736 		run->mmio.len = 1;
1737 		break;
1738 
1739 	default:
1740 		kvm_err("Load not yet supported (inst=0x%08x)\n",
1741 			inst.word);
1742 		vcpu->mmio_needed = 0;
1743 		return EMULATE_FAIL;
1744 	}
1745 
1746 	run->mmio.is_write = 0;
1747 	vcpu->mmio_is_write = 0;
1748 	return EMULATE_DO_MMIO;
1749 }
1750 
1751 #ifndef CONFIG_KVM_MIPS_VZ
1752 static enum emulation_result kvm_mips_guest_cache_op(int (*fn)(unsigned long),
1753 						     unsigned long curr_pc,
1754 						     unsigned long addr,
1755 						     struct kvm_run *run,
1756 						     struct kvm_vcpu *vcpu,
1757 						     u32 cause)
1758 {
1759 	int err;
1760 
1761 	for (;;) {
1762 		/* Carefully attempt the cache operation */
1763 		kvm_trap_emul_gva_lockless_begin(vcpu);
1764 		err = fn(addr);
1765 		kvm_trap_emul_gva_lockless_end(vcpu);
1766 
1767 		if (likely(!err))
1768 			return EMULATE_DONE;
1769 
1770 		/*
1771 		 * Try to handle the fault and retry, maybe we just raced with a
1772 		 * GVA invalidation.
1773 		 */
1774 		switch (kvm_trap_emul_gva_fault(vcpu, addr, false)) {
1775 		case KVM_MIPS_GVA:
1776 		case KVM_MIPS_GPA:
1777 			/* bad virtual or physical address */
1778 			return EMULATE_FAIL;
1779 		case KVM_MIPS_TLB:
1780 			/* no matching guest TLB */
1781 			vcpu->arch.host_cp0_badvaddr = addr;
1782 			vcpu->arch.pc = curr_pc;
1783 			kvm_mips_emulate_tlbmiss_ld(cause, NULL, run, vcpu);
1784 			return EMULATE_EXCEPT;
1785 		case KVM_MIPS_TLBINV:
1786 			/* invalid matching guest TLB */
1787 			vcpu->arch.host_cp0_badvaddr = addr;
1788 			vcpu->arch.pc = curr_pc;
1789 			kvm_mips_emulate_tlbinv_ld(cause, NULL, run, vcpu);
1790 			return EMULATE_EXCEPT;
1791 		default:
1792 			break;
1793 		}
1794 	}
1795 }
1796 
1797 enum emulation_result kvm_mips_emulate_cache(union mips_instruction inst,
1798 					     u32 *opc, u32 cause,
1799 					     struct kvm_run *run,
1800 					     struct kvm_vcpu *vcpu)
1801 {
1802 	enum emulation_result er = EMULATE_DONE;
1803 	u32 cache, op_inst, op, base;
1804 	s16 offset;
1805 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1806 	unsigned long va;
1807 	unsigned long curr_pc;
1808 
1809 	/*
1810 	 * Update PC and hold onto current PC in case there is
1811 	 * an error and we want to rollback the PC
1812 	 */
1813 	curr_pc = vcpu->arch.pc;
1814 	er = update_pc(vcpu, cause);
1815 	if (er == EMULATE_FAIL)
1816 		return er;
1817 
1818 	base = inst.i_format.rs;
1819 	op_inst = inst.i_format.rt;
1820 	if (cpu_has_mips_r6)
1821 		offset = inst.spec3_format.simmediate;
1822 	else
1823 		offset = inst.i_format.simmediate;
1824 	cache = op_inst & CacheOp_Cache;
1825 	op = op_inst & CacheOp_Op;
1826 
1827 	va = arch->gprs[base] + offset;
1828 
1829 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1830 		  cache, op, base, arch->gprs[base], offset);
1831 
1832 	/*
1833 	 * Treat INDEX_INV as a nop, basically issued by Linux on startup to
1834 	 * invalidate the caches entirely by stepping through all the
1835 	 * ways/indexes
1836 	 */
1837 	if (op == Index_Writeback_Inv) {
1838 		kvm_debug("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1839 			  vcpu->arch.pc, vcpu->arch.gprs[31], cache, op, base,
1840 			  arch->gprs[base], offset);
1841 
1842 		if (cache == Cache_D) {
1843 #ifdef CONFIG_CPU_R4K_CACHE_TLB
1844 			r4k_blast_dcache();
1845 #else
1846 			switch (boot_cpu_type()) {
1847 			case CPU_CAVIUM_OCTEON3:
1848 				/* locally flush icache */
1849 				local_flush_icache_range(0, 0);
1850 				break;
1851 			default:
1852 				__flush_cache_all();
1853 				break;
1854 			}
1855 #endif
1856 		} else if (cache == Cache_I) {
1857 #ifdef CONFIG_CPU_R4K_CACHE_TLB
1858 			r4k_blast_icache();
1859 #else
1860 			switch (boot_cpu_type()) {
1861 			case CPU_CAVIUM_OCTEON3:
1862 				/* locally flush icache */
1863 				local_flush_icache_range(0, 0);
1864 				break;
1865 			default:
1866 				flush_icache_all();
1867 				break;
1868 			}
1869 #endif
1870 		} else {
1871 			kvm_err("%s: unsupported CACHE INDEX operation\n",
1872 				__func__);
1873 			return EMULATE_FAIL;
1874 		}
1875 
1876 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1877 		kvm_mips_trans_cache_index(inst, opc, vcpu);
1878 #endif
1879 		goto done;
1880 	}
1881 
1882 	/* XXXKYMA: Only a subset of cache ops are supported, used by Linux */
1883 	if (op_inst == Hit_Writeback_Inv_D || op_inst == Hit_Invalidate_D) {
1884 		/*
1885 		 * Perform the dcache part of icache synchronisation on the
1886 		 * guest's behalf.
1887 		 */
1888 		er = kvm_mips_guest_cache_op(protected_writeback_dcache_line,
1889 					     curr_pc, va, run, vcpu, cause);
1890 		if (er != EMULATE_DONE)
1891 			goto done;
1892 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1893 		/*
1894 		 * Replace the CACHE instruction, with a SYNCI, not the same,
1895 		 * but avoids a trap
1896 		 */
1897 		kvm_mips_trans_cache_va(inst, opc, vcpu);
1898 #endif
1899 	} else if (op_inst == Hit_Invalidate_I) {
1900 		/* Perform the icache synchronisation on the guest's behalf */
1901 		er = kvm_mips_guest_cache_op(protected_writeback_dcache_line,
1902 					     curr_pc, va, run, vcpu, cause);
1903 		if (er != EMULATE_DONE)
1904 			goto done;
1905 		er = kvm_mips_guest_cache_op(protected_flush_icache_line,
1906 					     curr_pc, va, run, vcpu, cause);
1907 		if (er != EMULATE_DONE)
1908 			goto done;
1909 
1910 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1911 		/* Replace the CACHE instruction, with a SYNCI */
1912 		kvm_mips_trans_cache_va(inst, opc, vcpu);
1913 #endif
1914 	} else {
1915 		kvm_err("NO-OP CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1916 			cache, op, base, arch->gprs[base], offset);
1917 		er = EMULATE_FAIL;
1918 	}
1919 
1920 done:
1921 	/* Rollback PC only if emulation was unsuccessful */
1922 	if (er == EMULATE_FAIL)
1923 		vcpu->arch.pc = curr_pc;
1924 	/* Guest exception needs guest to resume */
1925 	if (er == EMULATE_EXCEPT)
1926 		er = EMULATE_DONE;
1927 
1928 	return er;
1929 }
1930 
1931 enum emulation_result kvm_mips_emulate_inst(u32 cause, u32 *opc,
1932 					    struct kvm_run *run,
1933 					    struct kvm_vcpu *vcpu)
1934 {
1935 	union mips_instruction inst;
1936 	enum emulation_result er = EMULATE_DONE;
1937 	int err;
1938 
1939 	/* Fetch the instruction. */
1940 	if (cause & CAUSEF_BD)
1941 		opc += 1;
1942 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1943 	if (err)
1944 		return EMULATE_FAIL;
1945 
1946 	switch (inst.r_format.opcode) {
1947 	case cop0_op:
1948 		er = kvm_mips_emulate_CP0(inst, opc, cause, run, vcpu);
1949 		break;
1950 
1951 #ifndef CONFIG_CPU_MIPSR6
1952 	case cache_op:
1953 		++vcpu->stat.cache_exits;
1954 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1955 		er = kvm_mips_emulate_cache(inst, opc, cause, run, vcpu);
1956 		break;
1957 #else
1958 	case spec3_op:
1959 		switch (inst.spec3_format.func) {
1960 		case cache6_op:
1961 			++vcpu->stat.cache_exits;
1962 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1963 			er = kvm_mips_emulate_cache(inst, opc, cause, run,
1964 						    vcpu);
1965 			break;
1966 		default:
1967 			goto unknown;
1968 		}
1969 		break;
1970 unknown:
1971 #endif
1972 
1973 	default:
1974 		kvm_err("Instruction emulation not supported (%p/%#x)\n", opc,
1975 			inst.word);
1976 		kvm_arch_vcpu_dump_regs(vcpu);
1977 		er = EMULATE_FAIL;
1978 		break;
1979 	}
1980 
1981 	return er;
1982 }
1983 #endif /* CONFIG_KVM_MIPS_VZ */
1984 
1985 /**
1986  * kvm_mips_guest_exception_base() - Find guest exception vector base address.
1987  *
1988  * Returns:	The base address of the current guest exception vector, taking
1989  *		both Guest.CP0_Status.BEV and Guest.CP0_EBase into account.
1990  */
1991 long kvm_mips_guest_exception_base(struct kvm_vcpu *vcpu)
1992 {
1993 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1994 
1995 	if (kvm_read_c0_guest_status(cop0) & ST0_BEV)
1996 		return KVM_GUEST_CKSEG1ADDR(0x1fc00200);
1997 	else
1998 		return kvm_read_c0_guest_ebase(cop0) & MIPS_EBASE_BASE;
1999 }
2000 
2001 enum emulation_result kvm_mips_emulate_syscall(u32 cause,
2002 					       u32 *opc,
2003 					       struct kvm_run *run,
2004 					       struct kvm_vcpu *vcpu)
2005 {
2006 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2007 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2008 	enum emulation_result er = EMULATE_DONE;
2009 
2010 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2011 		/* save old pc */
2012 		kvm_write_c0_guest_epc(cop0, arch->pc);
2013 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2014 
2015 		if (cause & CAUSEF_BD)
2016 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2017 		else
2018 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2019 
2020 		kvm_debug("Delivering SYSCALL @ pc %#lx\n", arch->pc);
2021 
2022 		kvm_change_c0_guest_cause(cop0, (0xff),
2023 					  (EXCCODE_SYS << CAUSEB_EXCCODE));
2024 
2025 		/* Set PC to the exception entry point */
2026 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2027 
2028 	} else {
2029 		kvm_err("Trying to deliver SYSCALL when EXL is already set\n");
2030 		er = EMULATE_FAIL;
2031 	}
2032 
2033 	return er;
2034 }
2035 
2036 enum emulation_result kvm_mips_emulate_tlbmiss_ld(u32 cause,
2037 						  u32 *opc,
2038 						  struct kvm_run *run,
2039 						  struct kvm_vcpu *vcpu)
2040 {
2041 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2042 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2043 	unsigned long entryhi = (vcpu->arch.  host_cp0_badvaddr & VPN2_MASK) |
2044 			(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2045 
2046 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2047 		/* save old pc */
2048 		kvm_write_c0_guest_epc(cop0, arch->pc);
2049 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2050 
2051 		if (cause & CAUSEF_BD)
2052 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2053 		else
2054 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2055 
2056 		kvm_debug("[EXL == 0] delivering TLB MISS @ pc %#lx\n",
2057 			  arch->pc);
2058 
2059 		/* set pc to the exception entry point */
2060 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x0;
2061 
2062 	} else {
2063 		kvm_debug("[EXL == 1] delivering TLB MISS @ pc %#lx\n",
2064 			  arch->pc);
2065 
2066 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2067 	}
2068 
2069 	kvm_change_c0_guest_cause(cop0, (0xff),
2070 				  (EXCCODE_TLBL << CAUSEB_EXCCODE));
2071 
2072 	/* setup badvaddr, context and entryhi registers for the guest */
2073 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2074 	/* XXXKYMA: is the context register used by linux??? */
2075 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2076 
2077 	return EMULATE_DONE;
2078 }
2079 
2080 enum emulation_result kvm_mips_emulate_tlbinv_ld(u32 cause,
2081 						 u32 *opc,
2082 						 struct kvm_run *run,
2083 						 struct kvm_vcpu *vcpu)
2084 {
2085 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2086 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2087 	unsigned long entryhi =
2088 		(vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2089 		(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2090 
2091 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2092 		/* save old pc */
2093 		kvm_write_c0_guest_epc(cop0, arch->pc);
2094 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2095 
2096 		if (cause & CAUSEF_BD)
2097 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2098 		else
2099 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2100 
2101 		kvm_debug("[EXL == 0] delivering TLB INV @ pc %#lx\n",
2102 			  arch->pc);
2103 	} else {
2104 		kvm_debug("[EXL == 1] delivering TLB MISS @ pc %#lx\n",
2105 			  arch->pc);
2106 	}
2107 
2108 	/* set pc to the exception entry point */
2109 	arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2110 
2111 	kvm_change_c0_guest_cause(cop0, (0xff),
2112 				  (EXCCODE_TLBL << CAUSEB_EXCCODE));
2113 
2114 	/* setup badvaddr, context and entryhi registers for the guest */
2115 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2116 	/* XXXKYMA: is the context register used by linux??? */
2117 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2118 
2119 	return EMULATE_DONE;
2120 }
2121 
2122 enum emulation_result kvm_mips_emulate_tlbmiss_st(u32 cause,
2123 						  u32 *opc,
2124 						  struct kvm_run *run,
2125 						  struct kvm_vcpu *vcpu)
2126 {
2127 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2128 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2129 	unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2130 			(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2131 
2132 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2133 		/* save old pc */
2134 		kvm_write_c0_guest_epc(cop0, arch->pc);
2135 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2136 
2137 		if (cause & CAUSEF_BD)
2138 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2139 		else
2140 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2141 
2142 		kvm_debug("[EXL == 0] Delivering TLB MISS @ pc %#lx\n",
2143 			  arch->pc);
2144 
2145 		/* Set PC to the exception entry point */
2146 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x0;
2147 	} else {
2148 		kvm_debug("[EXL == 1] Delivering TLB MISS @ pc %#lx\n",
2149 			  arch->pc);
2150 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2151 	}
2152 
2153 	kvm_change_c0_guest_cause(cop0, (0xff),
2154 				  (EXCCODE_TLBS << CAUSEB_EXCCODE));
2155 
2156 	/* setup badvaddr, context and entryhi registers for the guest */
2157 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2158 	/* XXXKYMA: is the context register used by linux??? */
2159 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2160 
2161 	return EMULATE_DONE;
2162 }
2163 
2164 enum emulation_result kvm_mips_emulate_tlbinv_st(u32 cause,
2165 						 u32 *opc,
2166 						 struct kvm_run *run,
2167 						 struct kvm_vcpu *vcpu)
2168 {
2169 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2170 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2171 	unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2172 		(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2173 
2174 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2175 		/* save old pc */
2176 		kvm_write_c0_guest_epc(cop0, arch->pc);
2177 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2178 
2179 		if (cause & CAUSEF_BD)
2180 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2181 		else
2182 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2183 
2184 		kvm_debug("[EXL == 0] Delivering TLB MISS @ pc %#lx\n",
2185 			  arch->pc);
2186 	} else {
2187 		kvm_debug("[EXL == 1] Delivering TLB MISS @ pc %#lx\n",
2188 			  arch->pc);
2189 	}
2190 
2191 	/* Set PC to the exception entry point */
2192 	arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2193 
2194 	kvm_change_c0_guest_cause(cop0, (0xff),
2195 				  (EXCCODE_TLBS << CAUSEB_EXCCODE));
2196 
2197 	/* setup badvaddr, context and entryhi registers for the guest */
2198 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2199 	/* XXXKYMA: is the context register used by linux??? */
2200 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2201 
2202 	return EMULATE_DONE;
2203 }
2204 
2205 enum emulation_result kvm_mips_emulate_tlbmod(u32 cause,
2206 					      u32 *opc,
2207 					      struct kvm_run *run,
2208 					      struct kvm_vcpu *vcpu)
2209 {
2210 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2211 	unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2212 			(kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2213 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2214 
2215 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2216 		/* save old pc */
2217 		kvm_write_c0_guest_epc(cop0, arch->pc);
2218 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2219 
2220 		if (cause & CAUSEF_BD)
2221 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2222 		else
2223 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2224 
2225 		kvm_debug("[EXL == 0] Delivering TLB MOD @ pc %#lx\n",
2226 			  arch->pc);
2227 	} else {
2228 		kvm_debug("[EXL == 1] Delivering TLB MOD @ pc %#lx\n",
2229 			  arch->pc);
2230 	}
2231 
2232 	arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2233 
2234 	kvm_change_c0_guest_cause(cop0, (0xff),
2235 				  (EXCCODE_MOD << CAUSEB_EXCCODE));
2236 
2237 	/* setup badvaddr, context and entryhi registers for the guest */
2238 	kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2239 	/* XXXKYMA: is the context register used by linux??? */
2240 	kvm_write_c0_guest_entryhi(cop0, entryhi);
2241 
2242 	return EMULATE_DONE;
2243 }
2244 
2245 enum emulation_result kvm_mips_emulate_fpu_exc(u32 cause,
2246 					       u32 *opc,
2247 					       struct kvm_run *run,
2248 					       struct kvm_vcpu *vcpu)
2249 {
2250 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2251 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2252 
2253 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2254 		/* save old pc */
2255 		kvm_write_c0_guest_epc(cop0, arch->pc);
2256 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2257 
2258 		if (cause & CAUSEF_BD)
2259 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2260 		else
2261 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2262 
2263 	}
2264 
2265 	arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2266 
2267 	kvm_change_c0_guest_cause(cop0, (0xff),
2268 				  (EXCCODE_CPU << CAUSEB_EXCCODE));
2269 	kvm_change_c0_guest_cause(cop0, (CAUSEF_CE), (0x1 << CAUSEB_CE));
2270 
2271 	return EMULATE_DONE;
2272 }
2273 
2274 enum emulation_result kvm_mips_emulate_ri_exc(u32 cause,
2275 					      u32 *opc,
2276 					      struct kvm_run *run,
2277 					      struct kvm_vcpu *vcpu)
2278 {
2279 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2280 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2281 	enum emulation_result er = EMULATE_DONE;
2282 
2283 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2284 		/* save old pc */
2285 		kvm_write_c0_guest_epc(cop0, arch->pc);
2286 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2287 
2288 		if (cause & CAUSEF_BD)
2289 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2290 		else
2291 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2292 
2293 		kvm_debug("Delivering RI @ pc %#lx\n", arch->pc);
2294 
2295 		kvm_change_c0_guest_cause(cop0, (0xff),
2296 					  (EXCCODE_RI << CAUSEB_EXCCODE));
2297 
2298 		/* Set PC to the exception entry point */
2299 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2300 
2301 	} else {
2302 		kvm_err("Trying to deliver RI when EXL is already set\n");
2303 		er = EMULATE_FAIL;
2304 	}
2305 
2306 	return er;
2307 }
2308 
2309 enum emulation_result kvm_mips_emulate_bp_exc(u32 cause,
2310 					      u32 *opc,
2311 					      struct kvm_run *run,
2312 					      struct kvm_vcpu *vcpu)
2313 {
2314 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2315 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2316 	enum emulation_result er = EMULATE_DONE;
2317 
2318 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2319 		/* save old pc */
2320 		kvm_write_c0_guest_epc(cop0, arch->pc);
2321 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2322 
2323 		if (cause & CAUSEF_BD)
2324 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2325 		else
2326 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2327 
2328 		kvm_debug("Delivering BP @ pc %#lx\n", arch->pc);
2329 
2330 		kvm_change_c0_guest_cause(cop0, (0xff),
2331 					  (EXCCODE_BP << CAUSEB_EXCCODE));
2332 
2333 		/* Set PC to the exception entry point */
2334 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2335 
2336 	} else {
2337 		kvm_err("Trying to deliver BP when EXL is already set\n");
2338 		er = EMULATE_FAIL;
2339 	}
2340 
2341 	return er;
2342 }
2343 
2344 enum emulation_result kvm_mips_emulate_trap_exc(u32 cause,
2345 						u32 *opc,
2346 						struct kvm_run *run,
2347 						struct kvm_vcpu *vcpu)
2348 {
2349 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2350 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2351 	enum emulation_result er = EMULATE_DONE;
2352 
2353 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2354 		/* save old pc */
2355 		kvm_write_c0_guest_epc(cop0, arch->pc);
2356 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2357 
2358 		if (cause & CAUSEF_BD)
2359 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2360 		else
2361 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2362 
2363 		kvm_debug("Delivering TRAP @ pc %#lx\n", arch->pc);
2364 
2365 		kvm_change_c0_guest_cause(cop0, (0xff),
2366 					  (EXCCODE_TR << CAUSEB_EXCCODE));
2367 
2368 		/* Set PC to the exception entry point */
2369 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2370 
2371 	} else {
2372 		kvm_err("Trying to deliver TRAP when EXL is already set\n");
2373 		er = EMULATE_FAIL;
2374 	}
2375 
2376 	return er;
2377 }
2378 
2379 enum emulation_result kvm_mips_emulate_msafpe_exc(u32 cause,
2380 						  u32 *opc,
2381 						  struct kvm_run *run,
2382 						  struct kvm_vcpu *vcpu)
2383 {
2384 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2385 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2386 	enum emulation_result er = EMULATE_DONE;
2387 
2388 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2389 		/* save old pc */
2390 		kvm_write_c0_guest_epc(cop0, arch->pc);
2391 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2392 
2393 		if (cause & CAUSEF_BD)
2394 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2395 		else
2396 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2397 
2398 		kvm_debug("Delivering MSAFPE @ pc %#lx\n", arch->pc);
2399 
2400 		kvm_change_c0_guest_cause(cop0, (0xff),
2401 					  (EXCCODE_MSAFPE << CAUSEB_EXCCODE));
2402 
2403 		/* Set PC to the exception entry point */
2404 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2405 
2406 	} else {
2407 		kvm_err("Trying to deliver MSAFPE when EXL is already set\n");
2408 		er = EMULATE_FAIL;
2409 	}
2410 
2411 	return er;
2412 }
2413 
2414 enum emulation_result kvm_mips_emulate_fpe_exc(u32 cause,
2415 					       u32 *opc,
2416 					       struct kvm_run *run,
2417 					       struct kvm_vcpu *vcpu)
2418 {
2419 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2420 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2421 	enum emulation_result er = EMULATE_DONE;
2422 
2423 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2424 		/* save old pc */
2425 		kvm_write_c0_guest_epc(cop0, arch->pc);
2426 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2427 
2428 		if (cause & CAUSEF_BD)
2429 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2430 		else
2431 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2432 
2433 		kvm_debug("Delivering FPE @ pc %#lx\n", arch->pc);
2434 
2435 		kvm_change_c0_guest_cause(cop0, (0xff),
2436 					  (EXCCODE_FPE << CAUSEB_EXCCODE));
2437 
2438 		/* Set PC to the exception entry point */
2439 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2440 
2441 	} else {
2442 		kvm_err("Trying to deliver FPE when EXL is already set\n");
2443 		er = EMULATE_FAIL;
2444 	}
2445 
2446 	return er;
2447 }
2448 
2449 enum emulation_result kvm_mips_emulate_msadis_exc(u32 cause,
2450 						  u32 *opc,
2451 						  struct kvm_run *run,
2452 						  struct kvm_vcpu *vcpu)
2453 {
2454 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2455 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2456 	enum emulation_result er = EMULATE_DONE;
2457 
2458 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2459 		/* save old pc */
2460 		kvm_write_c0_guest_epc(cop0, arch->pc);
2461 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2462 
2463 		if (cause & CAUSEF_BD)
2464 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2465 		else
2466 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2467 
2468 		kvm_debug("Delivering MSADIS @ pc %#lx\n", arch->pc);
2469 
2470 		kvm_change_c0_guest_cause(cop0, (0xff),
2471 					  (EXCCODE_MSADIS << CAUSEB_EXCCODE));
2472 
2473 		/* Set PC to the exception entry point */
2474 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2475 
2476 	} else {
2477 		kvm_err("Trying to deliver MSADIS when EXL is already set\n");
2478 		er = EMULATE_FAIL;
2479 	}
2480 
2481 	return er;
2482 }
2483 
2484 enum emulation_result kvm_mips_handle_ri(u32 cause, u32 *opc,
2485 					 struct kvm_run *run,
2486 					 struct kvm_vcpu *vcpu)
2487 {
2488 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2489 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2490 	enum emulation_result er = EMULATE_DONE;
2491 	unsigned long curr_pc;
2492 	union mips_instruction inst;
2493 	int err;
2494 
2495 	/*
2496 	 * Update PC and hold onto current PC in case there is
2497 	 * an error and we want to rollback the PC
2498 	 */
2499 	curr_pc = vcpu->arch.pc;
2500 	er = update_pc(vcpu, cause);
2501 	if (er == EMULATE_FAIL)
2502 		return er;
2503 
2504 	/* Fetch the instruction. */
2505 	if (cause & CAUSEF_BD)
2506 		opc += 1;
2507 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
2508 	if (err) {
2509 		kvm_err("%s: Cannot get inst @ %p (%d)\n", __func__, opc, err);
2510 		return EMULATE_FAIL;
2511 	}
2512 
2513 	if (inst.r_format.opcode == spec3_op &&
2514 	    inst.r_format.func == rdhwr_op &&
2515 	    inst.r_format.rs == 0 &&
2516 	    (inst.r_format.re >> 3) == 0) {
2517 		int usermode = !KVM_GUEST_KERNEL_MODE(vcpu);
2518 		int rd = inst.r_format.rd;
2519 		int rt = inst.r_format.rt;
2520 		int sel = inst.r_format.re & 0x7;
2521 
2522 		/* If usermode, check RDHWR rd is allowed by guest HWREna */
2523 		if (usermode && !(kvm_read_c0_guest_hwrena(cop0) & BIT(rd))) {
2524 			kvm_debug("RDHWR %#x disallowed by HWREna @ %p\n",
2525 				  rd, opc);
2526 			goto emulate_ri;
2527 		}
2528 		switch (rd) {
2529 		case MIPS_HWR_CPUNUM:		/* CPU number */
2530 			arch->gprs[rt] = vcpu->vcpu_id;
2531 			break;
2532 		case MIPS_HWR_SYNCISTEP:	/* SYNCI length */
2533 			arch->gprs[rt] = min(current_cpu_data.dcache.linesz,
2534 					     current_cpu_data.icache.linesz);
2535 			break;
2536 		case MIPS_HWR_CC:		/* Read count register */
2537 			arch->gprs[rt] = (s32)kvm_mips_read_count(vcpu);
2538 			break;
2539 		case MIPS_HWR_CCRES:		/* Count register resolution */
2540 			switch (current_cpu_data.cputype) {
2541 			case CPU_20KC:
2542 			case CPU_25KF:
2543 				arch->gprs[rt] = 1;
2544 				break;
2545 			default:
2546 				arch->gprs[rt] = 2;
2547 			}
2548 			break;
2549 		case MIPS_HWR_ULR:		/* Read UserLocal register */
2550 			arch->gprs[rt] = kvm_read_c0_guest_userlocal(cop0);
2551 			break;
2552 
2553 		default:
2554 			kvm_debug("RDHWR %#x not supported @ %p\n", rd, opc);
2555 			goto emulate_ri;
2556 		}
2557 
2558 		trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR, KVM_TRACE_HWR(rd, sel),
2559 			      vcpu->arch.gprs[rt]);
2560 	} else {
2561 		kvm_debug("Emulate RI not supported @ %p: %#x\n",
2562 			  opc, inst.word);
2563 		goto emulate_ri;
2564 	}
2565 
2566 	return EMULATE_DONE;
2567 
2568 emulate_ri:
2569 	/*
2570 	 * Rollback PC (if in branch delay slot then the PC already points to
2571 	 * branch target), and pass the RI exception to the guest OS.
2572 	 */
2573 	vcpu->arch.pc = curr_pc;
2574 	return kvm_mips_emulate_ri_exc(cause, opc, run, vcpu);
2575 }
2576 
2577 enum emulation_result kvm_mips_complete_mmio_load(struct kvm_vcpu *vcpu,
2578 						  struct kvm_run *run)
2579 {
2580 	unsigned long *gpr = &vcpu->arch.gprs[vcpu->arch.io_gpr];
2581 	enum emulation_result er = EMULATE_DONE;
2582 
2583 	if (run->mmio.len > sizeof(*gpr)) {
2584 		kvm_err("Bad MMIO length: %d", run->mmio.len);
2585 		er = EMULATE_FAIL;
2586 		goto done;
2587 	}
2588 
2589 	/* Restore saved resume PC */
2590 	vcpu->arch.pc = vcpu->arch.io_pc;
2591 
2592 	switch (run->mmio.len) {
2593 	case 8:
2594 		*gpr = *(s64 *)run->mmio.data;
2595 		break;
2596 
2597 	case 4:
2598 		if (vcpu->mmio_needed == 2)
2599 			*gpr = *(s32 *)run->mmio.data;
2600 		else
2601 			*gpr = *(u32 *)run->mmio.data;
2602 		break;
2603 
2604 	case 2:
2605 		if (vcpu->mmio_needed == 2)
2606 			*gpr = *(s16 *) run->mmio.data;
2607 		else
2608 			*gpr = *(u16 *)run->mmio.data;
2609 
2610 		break;
2611 	case 1:
2612 		if (vcpu->mmio_needed == 2)
2613 			*gpr = *(s8 *) run->mmio.data;
2614 		else
2615 			*gpr = *(u8 *) run->mmio.data;
2616 		break;
2617 	}
2618 
2619 done:
2620 	return er;
2621 }
2622 
2623 static enum emulation_result kvm_mips_emulate_exc(u32 cause,
2624 						  u32 *opc,
2625 						  struct kvm_run *run,
2626 						  struct kvm_vcpu *vcpu)
2627 {
2628 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2629 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2630 	struct kvm_vcpu_arch *arch = &vcpu->arch;
2631 	enum emulation_result er = EMULATE_DONE;
2632 
2633 	if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2634 		/* save old pc */
2635 		kvm_write_c0_guest_epc(cop0, arch->pc);
2636 		kvm_set_c0_guest_status(cop0, ST0_EXL);
2637 
2638 		if (cause & CAUSEF_BD)
2639 			kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2640 		else
2641 			kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2642 
2643 		kvm_change_c0_guest_cause(cop0, (0xff),
2644 					  (exccode << CAUSEB_EXCCODE));
2645 
2646 		/* Set PC to the exception entry point */
2647 		arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2648 		kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2649 
2650 		kvm_debug("Delivering EXC %d @ pc %#lx, badVaddr: %#lx\n",
2651 			  exccode, kvm_read_c0_guest_epc(cop0),
2652 			  kvm_read_c0_guest_badvaddr(cop0));
2653 	} else {
2654 		kvm_err("Trying to deliver EXC when EXL is already set\n");
2655 		er = EMULATE_FAIL;
2656 	}
2657 
2658 	return er;
2659 }
2660 
2661 enum emulation_result kvm_mips_check_privilege(u32 cause,
2662 					       u32 *opc,
2663 					       struct kvm_run *run,
2664 					       struct kvm_vcpu *vcpu)
2665 {
2666 	enum emulation_result er = EMULATE_DONE;
2667 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2668 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
2669 
2670 	int usermode = !KVM_GUEST_KERNEL_MODE(vcpu);
2671 
2672 	if (usermode) {
2673 		switch (exccode) {
2674 		case EXCCODE_INT:
2675 		case EXCCODE_SYS:
2676 		case EXCCODE_BP:
2677 		case EXCCODE_RI:
2678 		case EXCCODE_TR:
2679 		case EXCCODE_MSAFPE:
2680 		case EXCCODE_FPE:
2681 		case EXCCODE_MSADIS:
2682 			break;
2683 
2684 		case EXCCODE_CPU:
2685 			if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 0)
2686 				er = EMULATE_PRIV_FAIL;
2687 			break;
2688 
2689 		case EXCCODE_MOD:
2690 			break;
2691 
2692 		case EXCCODE_TLBL:
2693 			/*
2694 			 * We we are accessing Guest kernel space, then send an
2695 			 * address error exception to the guest
2696 			 */
2697 			if (badvaddr >= (unsigned long) KVM_GUEST_KSEG0) {
2698 				kvm_debug("%s: LD MISS @ %#lx\n", __func__,
2699 					  badvaddr);
2700 				cause &= ~0xff;
2701 				cause |= (EXCCODE_ADEL << CAUSEB_EXCCODE);
2702 				er = EMULATE_PRIV_FAIL;
2703 			}
2704 			break;
2705 
2706 		case EXCCODE_TLBS:
2707 			/*
2708 			 * We we are accessing Guest kernel space, then send an
2709 			 * address error exception to the guest
2710 			 */
2711 			if (badvaddr >= (unsigned long) KVM_GUEST_KSEG0) {
2712 				kvm_debug("%s: ST MISS @ %#lx\n", __func__,
2713 					  badvaddr);
2714 				cause &= ~0xff;
2715 				cause |= (EXCCODE_ADES << CAUSEB_EXCCODE);
2716 				er = EMULATE_PRIV_FAIL;
2717 			}
2718 			break;
2719 
2720 		case EXCCODE_ADES:
2721 			kvm_debug("%s: address error ST @ %#lx\n", __func__,
2722 				  badvaddr);
2723 			if ((badvaddr & PAGE_MASK) == KVM_GUEST_COMMPAGE_ADDR) {
2724 				cause &= ~0xff;
2725 				cause |= (EXCCODE_TLBS << CAUSEB_EXCCODE);
2726 			}
2727 			er = EMULATE_PRIV_FAIL;
2728 			break;
2729 		case EXCCODE_ADEL:
2730 			kvm_debug("%s: address error LD @ %#lx\n", __func__,
2731 				  badvaddr);
2732 			if ((badvaddr & PAGE_MASK) == KVM_GUEST_COMMPAGE_ADDR) {
2733 				cause &= ~0xff;
2734 				cause |= (EXCCODE_TLBL << CAUSEB_EXCCODE);
2735 			}
2736 			er = EMULATE_PRIV_FAIL;
2737 			break;
2738 		default:
2739 			er = EMULATE_PRIV_FAIL;
2740 			break;
2741 		}
2742 	}
2743 
2744 	if (er == EMULATE_PRIV_FAIL)
2745 		kvm_mips_emulate_exc(cause, opc, run, vcpu);
2746 
2747 	return er;
2748 }
2749 
2750 /*
2751  * User Address (UA) fault, this could happen if
2752  * (1) TLB entry not present/valid in both Guest and shadow host TLBs, in this
2753  *     case we pass on the fault to the guest kernel and let it handle it.
2754  * (2) TLB entry is present in the Guest TLB but not in the shadow, in this
2755  *     case we inject the TLB from the Guest TLB into the shadow host TLB
2756  */
2757 enum emulation_result kvm_mips_handle_tlbmiss(u32 cause,
2758 					      u32 *opc,
2759 					      struct kvm_run *run,
2760 					      struct kvm_vcpu *vcpu,
2761 					      bool write_fault)
2762 {
2763 	enum emulation_result er = EMULATE_DONE;
2764 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2765 	unsigned long va = vcpu->arch.host_cp0_badvaddr;
2766 	int index;
2767 
2768 	kvm_debug("kvm_mips_handle_tlbmiss: badvaddr: %#lx\n",
2769 		  vcpu->arch.host_cp0_badvaddr);
2770 
2771 	/*
2772 	 * KVM would not have got the exception if this entry was valid in the
2773 	 * shadow host TLB. Check the Guest TLB, if the entry is not there then
2774 	 * send the guest an exception. The guest exc handler should then inject
2775 	 * an entry into the guest TLB.
2776 	 */
2777 	index = kvm_mips_guest_tlb_lookup(vcpu,
2778 		      (va & VPN2_MASK) |
2779 		      (kvm_read_c0_guest_entryhi(vcpu->arch.cop0) &
2780 		       KVM_ENTRYHI_ASID));
2781 	if (index < 0) {
2782 		if (exccode == EXCCODE_TLBL) {
2783 			er = kvm_mips_emulate_tlbmiss_ld(cause, opc, run, vcpu);
2784 		} else if (exccode == EXCCODE_TLBS) {
2785 			er = kvm_mips_emulate_tlbmiss_st(cause, opc, run, vcpu);
2786 		} else {
2787 			kvm_err("%s: invalid exc code: %d\n", __func__,
2788 				exccode);
2789 			er = EMULATE_FAIL;
2790 		}
2791 	} else {
2792 		struct kvm_mips_tlb *tlb = &vcpu->arch.guest_tlb[index];
2793 
2794 		/*
2795 		 * Check if the entry is valid, if not then setup a TLB invalid
2796 		 * exception to the guest
2797 		 */
2798 		if (!TLB_IS_VALID(*tlb, va)) {
2799 			if (exccode == EXCCODE_TLBL) {
2800 				er = kvm_mips_emulate_tlbinv_ld(cause, opc, run,
2801 								vcpu);
2802 			} else if (exccode == EXCCODE_TLBS) {
2803 				er = kvm_mips_emulate_tlbinv_st(cause, opc, run,
2804 								vcpu);
2805 			} else {
2806 				kvm_err("%s: invalid exc code: %d\n", __func__,
2807 					exccode);
2808 				er = EMULATE_FAIL;
2809 			}
2810 		} else {
2811 			kvm_debug("Injecting hi: %#lx, lo0: %#lx, lo1: %#lx into shadow host TLB\n",
2812 				  tlb->tlb_hi, tlb->tlb_lo[0], tlb->tlb_lo[1]);
2813 			/*
2814 			 * OK we have a Guest TLB entry, now inject it into the
2815 			 * shadow host TLB
2816 			 */
2817 			if (kvm_mips_handle_mapped_seg_tlb_fault(vcpu, tlb, va,
2818 								 write_fault)) {
2819 				kvm_err("%s: handling mapped seg tlb fault for %lx, index: %u, vcpu: %p, ASID: %#lx\n",
2820 					__func__, va, index, vcpu,
2821 					read_c0_entryhi());
2822 				er = EMULATE_FAIL;
2823 			}
2824 		}
2825 	}
2826 
2827 	return er;
2828 }
2829