xref: /openbmc/linux/arch/mips/kvm/emulate.c (revision eb9fe179)
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 
34 #include "trace.h"
35 
36 /*
37  * Compute the return address and do emulate branch simulation, if required.
38  * This function should be called only in branch delay slot active.
39  */
40 static int kvm_compute_return_epc(struct kvm_vcpu *vcpu, unsigned long instpc,
41 				  unsigned long *out)
42 {
43 	unsigned int dspcontrol;
44 	union mips_instruction insn;
45 	struct kvm_vcpu_arch *arch = &vcpu->arch;
46 	long epc = instpc;
47 	long nextpc;
48 	int err;
49 
50 	if (epc & 3) {
51 		kvm_err("%s: unaligned epc\n", __func__);
52 		return -EINVAL;
53 	}
54 
55 	/* Read the instruction */
56 	err = kvm_get_badinstrp((u32 *)epc, vcpu, &insn.word);
57 	if (err)
58 		return err;
59 
60 	switch (insn.i_format.opcode) {
61 		/* jr and jalr are in r_format format. */
62 	case spec_op:
63 		switch (insn.r_format.func) {
64 		case jalr_op:
65 			arch->gprs[insn.r_format.rd] = epc + 8;
66 			fallthrough;
67 		case jr_op:
68 			nextpc = arch->gprs[insn.r_format.rs];
69 			break;
70 		default:
71 			return -EINVAL;
72 		}
73 		break;
74 
75 		/*
76 		 * This group contains:
77 		 * bltz_op, bgez_op, bltzl_op, bgezl_op,
78 		 * bltzal_op, bgezal_op, bltzall_op, bgezall_op.
79 		 */
80 	case bcond_op:
81 		switch (insn.i_format.rt) {
82 		case bltz_op:
83 		case bltzl_op:
84 			if ((long)arch->gprs[insn.i_format.rs] < 0)
85 				epc = epc + 4 + (insn.i_format.simmediate << 2);
86 			else
87 				epc += 8;
88 			nextpc = epc;
89 			break;
90 
91 		case bgez_op:
92 		case bgezl_op:
93 			if ((long)arch->gprs[insn.i_format.rs] >= 0)
94 				epc = epc + 4 + (insn.i_format.simmediate << 2);
95 			else
96 				epc += 8;
97 			nextpc = epc;
98 			break;
99 
100 		case bltzal_op:
101 		case bltzall_op:
102 			arch->gprs[31] = epc + 8;
103 			if ((long)arch->gprs[insn.i_format.rs] < 0)
104 				epc = epc + 4 + (insn.i_format.simmediate << 2);
105 			else
106 				epc += 8;
107 			nextpc = epc;
108 			break;
109 
110 		case bgezal_op:
111 		case bgezall_op:
112 			arch->gprs[31] = epc + 8;
113 			if ((long)arch->gprs[insn.i_format.rs] >= 0)
114 				epc = epc + 4 + (insn.i_format.simmediate << 2);
115 			else
116 				epc += 8;
117 			nextpc = epc;
118 			break;
119 		case bposge32_op:
120 			if (!cpu_has_dsp) {
121 				kvm_err("%s: DSP branch but not DSP ASE\n",
122 					__func__);
123 				return -EINVAL;
124 			}
125 
126 			dspcontrol = rddsp(0x01);
127 
128 			if (dspcontrol >= 32)
129 				epc = epc + 4 + (insn.i_format.simmediate << 2);
130 			else
131 				epc += 8;
132 			nextpc = epc;
133 			break;
134 		default:
135 			return -EINVAL;
136 		}
137 		break;
138 
139 		/* These are unconditional and in j_format. */
140 	case jal_op:
141 		arch->gprs[31] = instpc + 8;
142 		fallthrough;
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 		WARN_ONCE(1, "CPU doesn't have BadInstr register\n");
279 		return -EINVAL;
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 		WARN_ONCE(1, "CPU doesn't have BadInstrp register\n");
301 		return -EINVAL;
302 	}
303 }
304 
305 /**
306  * kvm_mips_count_disabled() - Find whether the CP0_Count timer is disabled.
307  * @vcpu:	Virtual CPU.
308  *
309  * Returns:	1 if the CP0_Count timer is disabled by either the guest
310  *		CP0_Cause.DC bit or the count_ctl.DC bit.
311  *		0 otherwise (in which case CP0_Count timer is running).
312  */
313 int kvm_mips_count_disabled(struct kvm_vcpu *vcpu)
314 {
315 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
316 
317 	return	(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC) ||
318 		(kvm_read_c0_guest_cause(cop0) & CAUSEF_DC);
319 }
320 
321 /**
322  * kvm_mips_ktime_to_count() - Scale ktime_t to a 32-bit count.
323  *
324  * Caches the dynamic nanosecond bias in vcpu->arch.count_dyn_bias.
325  *
326  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
327  */
328 static u32 kvm_mips_ktime_to_count(struct kvm_vcpu *vcpu, ktime_t now)
329 {
330 	s64 now_ns, periods;
331 	u64 delta;
332 
333 	now_ns = ktime_to_ns(now);
334 	delta = now_ns + vcpu->arch.count_dyn_bias;
335 
336 	if (delta >= vcpu->arch.count_period) {
337 		/* If delta is out of safe range the bias needs adjusting */
338 		periods = div64_s64(now_ns, vcpu->arch.count_period);
339 		vcpu->arch.count_dyn_bias = -periods * vcpu->arch.count_period;
340 		/* Recalculate delta with new bias */
341 		delta = now_ns + vcpu->arch.count_dyn_bias;
342 	}
343 
344 	/*
345 	 * We've ensured that:
346 	 *   delta < count_period
347 	 *
348 	 * Therefore the intermediate delta*count_hz will never overflow since
349 	 * at the boundary condition:
350 	 *   delta = count_period
351 	 *   delta = NSEC_PER_SEC * 2^32 / count_hz
352 	 *   delta * count_hz = NSEC_PER_SEC * 2^32
353 	 */
354 	return div_u64(delta * vcpu->arch.count_hz, NSEC_PER_SEC);
355 }
356 
357 /**
358  * kvm_mips_count_time() - Get effective current time.
359  * @vcpu:	Virtual CPU.
360  *
361  * Get effective monotonic ktime. This is usually a straightforward ktime_get(),
362  * except when the master disable bit is set in count_ctl, in which case it is
363  * count_resume, i.e. the time that the count was disabled.
364  *
365  * Returns:	Effective monotonic ktime for CP0_Count.
366  */
367 static inline ktime_t kvm_mips_count_time(struct kvm_vcpu *vcpu)
368 {
369 	if (unlikely(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC))
370 		return vcpu->arch.count_resume;
371 
372 	return ktime_get();
373 }
374 
375 /**
376  * kvm_mips_read_count_running() - Read the current count value as if running.
377  * @vcpu:	Virtual CPU.
378  * @now:	Kernel time to read CP0_Count at.
379  *
380  * Returns the current guest CP0_Count register at time @now and handles if the
381  * timer interrupt is pending and hasn't been handled yet.
382  *
383  * Returns:	The current value of the guest CP0_Count register.
384  */
385 static u32 kvm_mips_read_count_running(struct kvm_vcpu *vcpu, ktime_t now)
386 {
387 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
388 	ktime_t expires, threshold;
389 	u32 count, compare;
390 	int running;
391 
392 	/* Calculate the biased and scaled guest CP0_Count */
393 	count = vcpu->arch.count_bias + kvm_mips_ktime_to_count(vcpu, now);
394 	compare = kvm_read_c0_guest_compare(cop0);
395 
396 	/*
397 	 * Find whether CP0_Count has reached the closest timer interrupt. If
398 	 * not, we shouldn't inject it.
399 	 */
400 	if ((s32)(count - compare) < 0)
401 		return count;
402 
403 	/*
404 	 * The CP0_Count we're going to return has already reached the closest
405 	 * timer interrupt. Quickly check if it really is a new interrupt by
406 	 * looking at whether the interval until the hrtimer expiry time is
407 	 * less than 1/4 of the timer period.
408 	 */
409 	expires = hrtimer_get_expires(&vcpu->arch.comparecount_timer);
410 	threshold = ktime_add_ns(now, vcpu->arch.count_period / 4);
411 	if (ktime_before(expires, threshold)) {
412 		/*
413 		 * Cancel it while we handle it so there's no chance of
414 		 * interference with the timeout handler.
415 		 */
416 		running = hrtimer_cancel(&vcpu->arch.comparecount_timer);
417 
418 		/* Nothing should be waiting on the timeout */
419 		kvm_mips_callbacks->queue_timer_int(vcpu);
420 
421 		/*
422 		 * Restart the timer if it was running based on the expiry time
423 		 * we read, so that we don't push it back 2 periods.
424 		 */
425 		if (running) {
426 			expires = ktime_add_ns(expires,
427 					       vcpu->arch.count_period);
428 			hrtimer_start(&vcpu->arch.comparecount_timer, expires,
429 				      HRTIMER_MODE_ABS);
430 		}
431 	}
432 
433 	return count;
434 }
435 
436 /**
437  * kvm_mips_read_count() - Read the current count value.
438  * @vcpu:	Virtual CPU.
439  *
440  * Read the current guest CP0_Count value, taking into account whether the timer
441  * is stopped.
442  *
443  * Returns:	The current guest CP0_Count value.
444  */
445 u32 kvm_mips_read_count(struct kvm_vcpu *vcpu)
446 {
447 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
448 
449 	/* If count disabled just read static copy of count */
450 	if (kvm_mips_count_disabled(vcpu))
451 		return kvm_read_c0_guest_count(cop0);
452 
453 	return kvm_mips_read_count_running(vcpu, ktime_get());
454 }
455 
456 /**
457  * kvm_mips_freeze_hrtimer() - Safely stop the hrtimer.
458  * @vcpu:	Virtual CPU.
459  * @count:	Output pointer for CP0_Count value at point of freeze.
460  *
461  * Freeze the hrtimer safely and return both the ktime and the CP0_Count value
462  * at the point it was frozen. It is guaranteed that any pending interrupts at
463  * the point it was frozen are handled, and none after that point.
464  *
465  * This is useful where the time/CP0_Count is needed in the calculation of the
466  * new parameters.
467  *
468  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
469  *
470  * Returns:	The ktime at the point of freeze.
471  */
472 ktime_t kvm_mips_freeze_hrtimer(struct kvm_vcpu *vcpu, u32 *count)
473 {
474 	ktime_t now;
475 
476 	/* stop hrtimer before finding time */
477 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
478 	now = ktime_get();
479 
480 	/* find count at this point and handle pending hrtimer */
481 	*count = kvm_mips_read_count_running(vcpu, now);
482 
483 	return now;
484 }
485 
486 /**
487  * kvm_mips_resume_hrtimer() - Resume hrtimer, updating expiry.
488  * @vcpu:	Virtual CPU.
489  * @now:	ktime at point of resume.
490  * @count:	CP0_Count at point of resume.
491  *
492  * Resumes the timer and updates the timer expiry based on @now and @count.
493  * This can be used in conjunction with kvm_mips_freeze_timer() when timer
494  * parameters need to be changed.
495  *
496  * It is guaranteed that a timer interrupt immediately after resume will be
497  * handled, but not if CP_Compare is exactly at @count. That case is already
498  * handled by kvm_mips_freeze_timer().
499  *
500  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
501  */
502 static void kvm_mips_resume_hrtimer(struct kvm_vcpu *vcpu,
503 				    ktime_t now, u32 count)
504 {
505 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
506 	u32 compare;
507 	u64 delta;
508 	ktime_t expire;
509 
510 	/* Calculate timeout (wrap 0 to 2^32) */
511 	compare = kvm_read_c0_guest_compare(cop0);
512 	delta = (u64)(u32)(compare - count - 1) + 1;
513 	delta = div_u64(delta * NSEC_PER_SEC, vcpu->arch.count_hz);
514 	expire = ktime_add_ns(now, delta);
515 
516 	/* Update hrtimer to use new timeout */
517 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
518 	hrtimer_start(&vcpu->arch.comparecount_timer, expire, HRTIMER_MODE_ABS);
519 }
520 
521 /**
522  * kvm_mips_restore_hrtimer() - Restore hrtimer after a gap, updating expiry.
523  * @vcpu:	Virtual CPU.
524  * @before:	Time before Count was saved, lower bound of drift calculation.
525  * @count:	CP0_Count at point of restore.
526  * @min_drift:	Minimum amount of drift permitted before correction.
527  *		Must be <= 0.
528  *
529  * Restores the timer from a particular @count, accounting for drift. This can
530  * be used in conjunction with kvm_mips_freeze_timer() when a hardware timer is
531  * to be used for a period of time, but the exact ktime corresponding to the
532  * final Count that must be restored is not known.
533  *
534  * It is gauranteed that a timer interrupt immediately after restore will be
535  * handled, but not if CP0_Compare is exactly at @count. That case should
536  * already be handled when the hardware timer state is saved.
537  *
538  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is not
539  * stopped).
540  *
541  * Returns:	Amount of correction to count_bias due to drift.
542  */
543 int kvm_mips_restore_hrtimer(struct kvm_vcpu *vcpu, ktime_t before,
544 			     u32 count, int min_drift)
545 {
546 	ktime_t now, count_time;
547 	u32 now_count, before_count;
548 	u64 delta;
549 	int drift, ret = 0;
550 
551 	/* Calculate expected count at before */
552 	before_count = vcpu->arch.count_bias +
553 			kvm_mips_ktime_to_count(vcpu, before);
554 
555 	/*
556 	 * Detect significantly negative drift, where count is lower than
557 	 * expected. Some negative drift is expected when hardware counter is
558 	 * set after kvm_mips_freeze_timer(), and it is harmless to allow the
559 	 * time to jump forwards a little, within reason. If the drift is too
560 	 * significant, adjust the bias to avoid a big Guest.CP0_Count jump.
561 	 */
562 	drift = count - before_count;
563 	if (drift < min_drift) {
564 		count_time = before;
565 		vcpu->arch.count_bias += drift;
566 		ret = drift;
567 		goto resume;
568 	}
569 
570 	/* Calculate expected count right now */
571 	now = ktime_get();
572 	now_count = vcpu->arch.count_bias + kvm_mips_ktime_to_count(vcpu, now);
573 
574 	/*
575 	 * Detect positive drift, where count is higher than expected, and
576 	 * adjust the bias to avoid guest time going backwards.
577 	 */
578 	drift = count - now_count;
579 	if (drift > 0) {
580 		count_time = now;
581 		vcpu->arch.count_bias += drift;
582 		ret = drift;
583 		goto resume;
584 	}
585 
586 	/* Subtract nanosecond delta to find ktime when count was read */
587 	delta = (u64)(u32)(now_count - count);
588 	delta = div_u64(delta * NSEC_PER_SEC, vcpu->arch.count_hz);
589 	count_time = ktime_sub_ns(now, delta);
590 
591 resume:
592 	/* Resume using the calculated ktime */
593 	kvm_mips_resume_hrtimer(vcpu, count_time, count);
594 	return ret;
595 }
596 
597 /**
598  * kvm_mips_write_count() - Modify the count and update timer.
599  * @vcpu:	Virtual CPU.
600  * @count:	Guest CP0_Count value to set.
601  *
602  * Sets the CP0_Count value and updates the timer accordingly.
603  */
604 void kvm_mips_write_count(struct kvm_vcpu *vcpu, u32 count)
605 {
606 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
607 	ktime_t now;
608 
609 	/* Calculate bias */
610 	now = kvm_mips_count_time(vcpu);
611 	vcpu->arch.count_bias = count - kvm_mips_ktime_to_count(vcpu, now);
612 
613 	if (kvm_mips_count_disabled(vcpu))
614 		/* The timer's disabled, adjust the static count */
615 		kvm_write_c0_guest_count(cop0, count);
616 	else
617 		/* Update timeout */
618 		kvm_mips_resume_hrtimer(vcpu, now, count);
619 }
620 
621 /**
622  * kvm_mips_init_count() - Initialise timer.
623  * @vcpu:	Virtual CPU.
624  * @count_hz:	Frequency of timer.
625  *
626  * Initialise the timer to the specified frequency, zero it, and set it going if
627  * it's enabled.
628  */
629 void kvm_mips_init_count(struct kvm_vcpu *vcpu, unsigned long count_hz)
630 {
631 	vcpu->arch.count_hz = count_hz;
632 	vcpu->arch.count_period = div_u64((u64)NSEC_PER_SEC << 32, count_hz);
633 	vcpu->arch.count_dyn_bias = 0;
634 
635 	/* Starting at 0 */
636 	kvm_mips_write_count(vcpu, 0);
637 }
638 
639 /**
640  * kvm_mips_set_count_hz() - Update the frequency of the timer.
641  * @vcpu:	Virtual CPU.
642  * @count_hz:	Frequency of CP0_Count timer in Hz.
643  *
644  * Change the frequency of the CP0_Count timer. This is done atomically so that
645  * CP0_Count is continuous and no timer interrupt is lost.
646  *
647  * Returns:	-EINVAL if @count_hz is out of range.
648  *		0 on success.
649  */
650 int kvm_mips_set_count_hz(struct kvm_vcpu *vcpu, s64 count_hz)
651 {
652 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
653 	int dc;
654 	ktime_t now;
655 	u32 count;
656 
657 	/* ensure the frequency is in a sensible range... */
658 	if (count_hz <= 0 || count_hz > NSEC_PER_SEC)
659 		return -EINVAL;
660 	/* ... and has actually changed */
661 	if (vcpu->arch.count_hz == count_hz)
662 		return 0;
663 
664 	/* Safely freeze timer so we can keep it continuous */
665 	dc = kvm_mips_count_disabled(vcpu);
666 	if (dc) {
667 		now = kvm_mips_count_time(vcpu);
668 		count = kvm_read_c0_guest_count(cop0);
669 	} else {
670 		now = kvm_mips_freeze_hrtimer(vcpu, &count);
671 	}
672 
673 	/* Update the frequency */
674 	vcpu->arch.count_hz = count_hz;
675 	vcpu->arch.count_period = div_u64((u64)NSEC_PER_SEC << 32, count_hz);
676 	vcpu->arch.count_dyn_bias = 0;
677 
678 	/* Calculate adjusted bias so dynamic count is unchanged */
679 	vcpu->arch.count_bias = count - kvm_mips_ktime_to_count(vcpu, now);
680 
681 	/* Update and resume hrtimer */
682 	if (!dc)
683 		kvm_mips_resume_hrtimer(vcpu, now, count);
684 	return 0;
685 }
686 
687 /**
688  * kvm_mips_write_compare() - Modify compare and update timer.
689  * @vcpu:	Virtual CPU.
690  * @compare:	New CP0_Compare value.
691  * @ack:	Whether to acknowledge timer interrupt.
692  *
693  * Update CP0_Compare to a new value and update the timeout.
694  * If @ack, atomically acknowledge any pending timer interrupt, otherwise ensure
695  * any pending timer interrupt is preserved.
696  */
697 void kvm_mips_write_compare(struct kvm_vcpu *vcpu, u32 compare, bool ack)
698 {
699 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
700 	int dc;
701 	u32 old_compare = kvm_read_c0_guest_compare(cop0);
702 	s32 delta = compare - old_compare;
703 	u32 cause;
704 	ktime_t now = ktime_set(0, 0); /* silence bogus GCC warning */
705 	u32 count;
706 
707 	/* if unchanged, must just be an ack */
708 	if (old_compare == compare) {
709 		if (!ack)
710 			return;
711 		kvm_mips_callbacks->dequeue_timer_int(vcpu);
712 		kvm_write_c0_guest_compare(cop0, compare);
713 		return;
714 	}
715 
716 	/*
717 	 * If guest CP0_Compare moves forward, CP0_GTOffset should be adjusted
718 	 * too to prevent guest CP0_Count hitting guest CP0_Compare.
719 	 *
720 	 * The new GTOffset corresponds to the new value of CP0_Compare, and is
721 	 * set prior to it being written into the guest context. We disable
722 	 * preemption until the new value is written to prevent restore of a
723 	 * GTOffset corresponding to the old CP0_Compare value.
724 	 */
725 	if (delta > 0) {
726 		preempt_disable();
727 		write_c0_gtoffset(compare - read_c0_count());
728 		back_to_back_c0_hazard();
729 	}
730 
731 	/* freeze_hrtimer() takes care of timer interrupts <= count */
732 	dc = kvm_mips_count_disabled(vcpu);
733 	if (!dc)
734 		now = kvm_mips_freeze_hrtimer(vcpu, &count);
735 
736 	if (ack)
737 		kvm_mips_callbacks->dequeue_timer_int(vcpu);
738 	else
739 		/*
740 		 * With VZ, writing CP0_Compare acks (clears) CP0_Cause.TI, so
741 		 * preserve guest CP0_Cause.TI if we don't want to ack it.
742 		 */
743 		cause = kvm_read_c0_guest_cause(cop0);
744 
745 	kvm_write_c0_guest_compare(cop0, compare);
746 
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 	/* 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 (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_wait(struct kvm_vcpu *vcpu)
946 {
947 	kvm_debug("[%#lx] !!!WAIT!!! (%#lx)\n", vcpu->arch.pc,
948 		  vcpu->arch.pending_exceptions);
949 
950 	++vcpu->stat.wait_exits;
951 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_WAIT);
952 	if (!vcpu->arch.pending_exceptions) {
953 		kvm_vz_lose_htimer(vcpu);
954 		vcpu->arch.wait = 1;
955 		kvm_vcpu_halt(vcpu);
956 
957 		/*
958 		 * We are runnable, then definitely go off to user space to
959 		 * check if any I/O interrupts are pending.
960 		 */
961 		if (kvm_arch_vcpu_runnable(vcpu))
962 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
963 	}
964 
965 	return EMULATE_DONE;
966 }
967 
968 enum emulation_result kvm_mips_emulate_store(union mips_instruction inst,
969 					     u32 cause,
970 					     struct kvm_vcpu *vcpu)
971 {
972 	int r;
973 	enum emulation_result er;
974 	u32 rt;
975 	struct kvm_run *run = vcpu->run;
976 	void *data = run->mmio.data;
977 	unsigned int imme;
978 	unsigned long curr_pc;
979 
980 	/*
981 	 * Update PC and hold onto current PC in case there is
982 	 * an error and we want to rollback the PC
983 	 */
984 	curr_pc = vcpu->arch.pc;
985 	er = update_pc(vcpu, cause);
986 	if (er == EMULATE_FAIL)
987 		return er;
988 
989 	rt = inst.i_format.rt;
990 
991 	run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
992 						vcpu->arch.host_cp0_badvaddr);
993 	if (run->mmio.phys_addr == KVM_INVALID_ADDR)
994 		goto out_fail;
995 
996 	switch (inst.i_format.opcode) {
997 #if defined(CONFIG_64BIT)
998 	case sd_op:
999 		run->mmio.len = 8;
1000 		*(u64 *)data = vcpu->arch.gprs[rt];
1001 
1002 		kvm_debug("[%#lx] OP_SD: eaddr: %#lx, gpr: %#lx, data: %#llx\n",
1003 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1004 			  vcpu->arch.gprs[rt], *(u64 *)data);
1005 		break;
1006 #endif
1007 
1008 	case sw_op:
1009 		run->mmio.len = 4;
1010 		*(u32 *)data = vcpu->arch.gprs[rt];
1011 
1012 		kvm_debug("[%#lx] OP_SW: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1013 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1014 			  vcpu->arch.gprs[rt], *(u32 *)data);
1015 		break;
1016 
1017 	case sh_op:
1018 		run->mmio.len = 2;
1019 		*(u16 *)data = vcpu->arch.gprs[rt];
1020 
1021 		kvm_debug("[%#lx] OP_SH: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1022 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1023 			  vcpu->arch.gprs[rt], *(u16 *)data);
1024 		break;
1025 
1026 	case sb_op:
1027 		run->mmio.len = 1;
1028 		*(u8 *)data = vcpu->arch.gprs[rt];
1029 
1030 		kvm_debug("[%#lx] OP_SB: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1031 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1032 			  vcpu->arch.gprs[rt], *(u8 *)data);
1033 		break;
1034 
1035 	case swl_op:
1036 		run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1037 					vcpu->arch.host_cp0_badvaddr) & (~0x3);
1038 		run->mmio.len = 4;
1039 		imme = vcpu->arch.host_cp0_badvaddr & 0x3;
1040 		switch (imme) {
1041 		case 0:
1042 			*(u32 *)data = ((*(u32 *)data) & 0xffffff00) |
1043 					(vcpu->arch.gprs[rt] >> 24);
1044 			break;
1045 		case 1:
1046 			*(u32 *)data = ((*(u32 *)data) & 0xffff0000) |
1047 					(vcpu->arch.gprs[rt] >> 16);
1048 			break;
1049 		case 2:
1050 			*(u32 *)data = ((*(u32 *)data) & 0xff000000) |
1051 					(vcpu->arch.gprs[rt] >> 8);
1052 			break;
1053 		case 3:
1054 			*(u32 *)data = vcpu->arch.gprs[rt];
1055 			break;
1056 		default:
1057 			break;
1058 		}
1059 
1060 		kvm_debug("[%#lx] OP_SWL: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1061 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1062 			  vcpu->arch.gprs[rt], *(u32 *)data);
1063 		break;
1064 
1065 	case swr_op:
1066 		run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1067 					vcpu->arch.host_cp0_badvaddr) & (~0x3);
1068 		run->mmio.len = 4;
1069 		imme = vcpu->arch.host_cp0_badvaddr & 0x3;
1070 		switch (imme) {
1071 		case 0:
1072 			*(u32 *)data = vcpu->arch.gprs[rt];
1073 			break;
1074 		case 1:
1075 			*(u32 *)data = ((*(u32 *)data) & 0xff) |
1076 					(vcpu->arch.gprs[rt] << 8);
1077 			break;
1078 		case 2:
1079 			*(u32 *)data = ((*(u32 *)data) & 0xffff) |
1080 					(vcpu->arch.gprs[rt] << 16);
1081 			break;
1082 		case 3:
1083 			*(u32 *)data = ((*(u32 *)data) & 0xffffff) |
1084 					(vcpu->arch.gprs[rt] << 24);
1085 			break;
1086 		default:
1087 			break;
1088 		}
1089 
1090 		kvm_debug("[%#lx] OP_SWR: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1091 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1092 			  vcpu->arch.gprs[rt], *(u32 *)data);
1093 		break;
1094 
1095 #if defined(CONFIG_64BIT)
1096 	case sdl_op:
1097 		run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1098 					vcpu->arch.host_cp0_badvaddr) & (~0x7);
1099 
1100 		run->mmio.len = 8;
1101 		imme = vcpu->arch.host_cp0_badvaddr & 0x7;
1102 		switch (imme) {
1103 		case 0:
1104 			*(u64 *)data = ((*(u64 *)data) & 0xffffffffffffff00) |
1105 					((vcpu->arch.gprs[rt] >> 56) & 0xff);
1106 			break;
1107 		case 1:
1108 			*(u64 *)data = ((*(u64 *)data) & 0xffffffffffff0000) |
1109 					((vcpu->arch.gprs[rt] >> 48) & 0xffff);
1110 			break;
1111 		case 2:
1112 			*(u64 *)data = ((*(u64 *)data) & 0xffffffffff000000) |
1113 					((vcpu->arch.gprs[rt] >> 40) & 0xffffff);
1114 			break;
1115 		case 3:
1116 			*(u64 *)data = ((*(u64 *)data) & 0xffffffff00000000) |
1117 					((vcpu->arch.gprs[rt] >> 32) & 0xffffffff);
1118 			break;
1119 		case 4:
1120 			*(u64 *)data = ((*(u64 *)data) & 0xffffff0000000000) |
1121 					((vcpu->arch.gprs[rt] >> 24) & 0xffffffffff);
1122 			break;
1123 		case 5:
1124 			*(u64 *)data = ((*(u64 *)data) & 0xffff000000000000) |
1125 					((vcpu->arch.gprs[rt] >> 16) & 0xffffffffffff);
1126 			break;
1127 		case 6:
1128 			*(u64 *)data = ((*(u64 *)data) & 0xff00000000000000) |
1129 					((vcpu->arch.gprs[rt] >> 8) & 0xffffffffffffff);
1130 			break;
1131 		case 7:
1132 			*(u64 *)data = vcpu->arch.gprs[rt];
1133 			break;
1134 		default:
1135 			break;
1136 		}
1137 
1138 		kvm_debug("[%#lx] OP_SDL: eaddr: %#lx, gpr: %#lx, data: %llx\n",
1139 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1140 			  vcpu->arch.gprs[rt], *(u64 *)data);
1141 		break;
1142 
1143 	case sdr_op:
1144 		run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1145 					vcpu->arch.host_cp0_badvaddr) & (~0x7);
1146 
1147 		run->mmio.len = 8;
1148 		imme = vcpu->arch.host_cp0_badvaddr & 0x7;
1149 		switch (imme) {
1150 		case 0:
1151 			*(u64 *)data = vcpu->arch.gprs[rt];
1152 			break;
1153 		case 1:
1154 			*(u64 *)data = ((*(u64 *)data) & 0xff) |
1155 					(vcpu->arch.gprs[rt] << 8);
1156 			break;
1157 		case 2:
1158 			*(u64 *)data = ((*(u64 *)data) & 0xffff) |
1159 					(vcpu->arch.gprs[rt] << 16);
1160 			break;
1161 		case 3:
1162 			*(u64 *)data = ((*(u64 *)data) & 0xffffff) |
1163 					(vcpu->arch.gprs[rt] << 24);
1164 			break;
1165 		case 4:
1166 			*(u64 *)data = ((*(u64 *)data) & 0xffffffff) |
1167 					(vcpu->arch.gprs[rt] << 32);
1168 			break;
1169 		case 5:
1170 			*(u64 *)data = ((*(u64 *)data) & 0xffffffffff) |
1171 					(vcpu->arch.gprs[rt] << 40);
1172 			break;
1173 		case 6:
1174 			*(u64 *)data = ((*(u64 *)data) & 0xffffffffffff) |
1175 					(vcpu->arch.gprs[rt] << 48);
1176 			break;
1177 		case 7:
1178 			*(u64 *)data = ((*(u64 *)data) & 0xffffffffffffff) |
1179 					(vcpu->arch.gprs[rt] << 56);
1180 			break;
1181 		default:
1182 			break;
1183 		}
1184 
1185 		kvm_debug("[%#lx] OP_SDR: eaddr: %#lx, gpr: %#lx, data: %llx\n",
1186 			  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1187 			  vcpu->arch.gprs[rt], *(u64 *)data);
1188 		break;
1189 #endif
1190 
1191 #ifdef CONFIG_CPU_LOONGSON64
1192 	case sdc2_op:
1193 		rt = inst.loongson3_lsdc2_format.rt;
1194 		switch (inst.loongson3_lsdc2_format.opcode1) {
1195 		/*
1196 		 * Loongson-3 overridden sdc2 instructions.
1197 		 * opcode1              instruction
1198 		 *   0x0          gssbx: store 1 bytes from GPR
1199 		 *   0x1          gsshx: store 2 bytes from GPR
1200 		 *   0x2          gsswx: store 4 bytes from GPR
1201 		 *   0x3          gssdx: store 8 bytes from GPR
1202 		 */
1203 		case 0x0:
1204 			run->mmio.len = 1;
1205 			*(u8 *)data = vcpu->arch.gprs[rt];
1206 
1207 			kvm_debug("[%#lx] OP_GSSBX: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1208 				  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1209 				  vcpu->arch.gprs[rt], *(u8 *)data);
1210 			break;
1211 		case 0x1:
1212 			run->mmio.len = 2;
1213 			*(u16 *)data = vcpu->arch.gprs[rt];
1214 
1215 			kvm_debug("[%#lx] OP_GSSSHX: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1216 				  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1217 				  vcpu->arch.gprs[rt], *(u16 *)data);
1218 			break;
1219 		case 0x2:
1220 			run->mmio.len = 4;
1221 			*(u32 *)data = vcpu->arch.gprs[rt];
1222 
1223 			kvm_debug("[%#lx] OP_GSSWX: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1224 				  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1225 				  vcpu->arch.gprs[rt], *(u32 *)data);
1226 			break;
1227 		case 0x3:
1228 			run->mmio.len = 8;
1229 			*(u64 *)data = vcpu->arch.gprs[rt];
1230 
1231 			kvm_debug("[%#lx] OP_GSSDX: eaddr: %#lx, gpr: %#lx, data: %#llx\n",
1232 				  vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1233 				  vcpu->arch.gprs[rt], *(u64 *)data);
1234 			break;
1235 		default:
1236 			kvm_err("Godson Extended GS-Store not yet supported (inst=0x%08x)\n",
1237 				inst.word);
1238 			break;
1239 		}
1240 		break;
1241 #endif
1242 	default:
1243 		kvm_err("Store not yet supported (inst=0x%08x)\n",
1244 			inst.word);
1245 		goto out_fail;
1246 	}
1247 
1248 	vcpu->mmio_needed = 1;
1249 	run->mmio.is_write = 1;
1250 	vcpu->mmio_is_write = 1;
1251 
1252 	r = kvm_io_bus_write(vcpu, KVM_MMIO_BUS,
1253 			run->mmio.phys_addr, run->mmio.len, data);
1254 
1255 	if (!r) {
1256 		vcpu->mmio_needed = 0;
1257 		return EMULATE_DONE;
1258 	}
1259 
1260 	return EMULATE_DO_MMIO;
1261 
1262 out_fail:
1263 	/* Rollback PC if emulation was unsuccessful */
1264 	vcpu->arch.pc = curr_pc;
1265 	return EMULATE_FAIL;
1266 }
1267 
1268 enum emulation_result kvm_mips_emulate_load(union mips_instruction inst,
1269 					    u32 cause, struct kvm_vcpu *vcpu)
1270 {
1271 	struct kvm_run *run = vcpu->run;
1272 	int r;
1273 	enum emulation_result er;
1274 	unsigned long curr_pc;
1275 	u32 op, rt;
1276 	unsigned int imme;
1277 
1278 	rt = inst.i_format.rt;
1279 	op = inst.i_format.opcode;
1280 
1281 	/*
1282 	 * Find the resume PC now while we have safe and easy access to the
1283 	 * prior branch instruction, and save it for
1284 	 * kvm_mips_complete_mmio_load() to restore later.
1285 	 */
1286 	curr_pc = vcpu->arch.pc;
1287 	er = update_pc(vcpu, cause);
1288 	if (er == EMULATE_FAIL)
1289 		return er;
1290 	vcpu->arch.io_pc = vcpu->arch.pc;
1291 	vcpu->arch.pc = curr_pc;
1292 
1293 	vcpu->arch.io_gpr = rt;
1294 
1295 	run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1296 						vcpu->arch.host_cp0_badvaddr);
1297 	if (run->mmio.phys_addr == KVM_INVALID_ADDR)
1298 		return EMULATE_FAIL;
1299 
1300 	vcpu->mmio_needed = 2;	/* signed */
1301 	switch (op) {
1302 #if defined(CONFIG_64BIT)
1303 	case ld_op:
1304 		run->mmio.len = 8;
1305 		break;
1306 
1307 	case lwu_op:
1308 		vcpu->mmio_needed = 1;	/* unsigned */
1309 		fallthrough;
1310 #endif
1311 	case lw_op:
1312 		run->mmio.len = 4;
1313 		break;
1314 
1315 	case lhu_op:
1316 		vcpu->mmio_needed = 1;	/* unsigned */
1317 		fallthrough;
1318 	case lh_op:
1319 		run->mmio.len = 2;
1320 		break;
1321 
1322 	case lbu_op:
1323 		vcpu->mmio_needed = 1;	/* unsigned */
1324 		fallthrough;
1325 	case lb_op:
1326 		run->mmio.len = 1;
1327 		break;
1328 
1329 	case lwl_op:
1330 		run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1331 					vcpu->arch.host_cp0_badvaddr) & (~0x3);
1332 
1333 		run->mmio.len = 4;
1334 		imme = vcpu->arch.host_cp0_badvaddr & 0x3;
1335 		switch (imme) {
1336 		case 0:
1337 			vcpu->mmio_needed = 3;	/* 1 byte */
1338 			break;
1339 		case 1:
1340 			vcpu->mmio_needed = 4;	/* 2 bytes */
1341 			break;
1342 		case 2:
1343 			vcpu->mmio_needed = 5;	/* 3 bytes */
1344 			break;
1345 		case 3:
1346 			vcpu->mmio_needed = 6;	/* 4 bytes */
1347 			break;
1348 		default:
1349 			break;
1350 		}
1351 		break;
1352 
1353 	case lwr_op:
1354 		run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1355 					vcpu->arch.host_cp0_badvaddr) & (~0x3);
1356 
1357 		run->mmio.len = 4;
1358 		imme = vcpu->arch.host_cp0_badvaddr & 0x3;
1359 		switch (imme) {
1360 		case 0:
1361 			vcpu->mmio_needed = 7;	/* 4 bytes */
1362 			break;
1363 		case 1:
1364 			vcpu->mmio_needed = 8;	/* 3 bytes */
1365 			break;
1366 		case 2:
1367 			vcpu->mmio_needed = 9;	/* 2 bytes */
1368 			break;
1369 		case 3:
1370 			vcpu->mmio_needed = 10;	/* 1 byte */
1371 			break;
1372 		default:
1373 			break;
1374 		}
1375 		break;
1376 
1377 #if defined(CONFIG_64BIT)
1378 	case ldl_op:
1379 		run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1380 					vcpu->arch.host_cp0_badvaddr) & (~0x7);
1381 
1382 		run->mmio.len = 8;
1383 		imme = vcpu->arch.host_cp0_badvaddr & 0x7;
1384 		switch (imme) {
1385 		case 0:
1386 			vcpu->mmio_needed = 11;	/* 1 byte */
1387 			break;
1388 		case 1:
1389 			vcpu->mmio_needed = 12;	/* 2 bytes */
1390 			break;
1391 		case 2:
1392 			vcpu->mmio_needed = 13;	/* 3 bytes */
1393 			break;
1394 		case 3:
1395 			vcpu->mmio_needed = 14;	/* 4 bytes */
1396 			break;
1397 		case 4:
1398 			vcpu->mmio_needed = 15;	/* 5 bytes */
1399 			break;
1400 		case 5:
1401 			vcpu->mmio_needed = 16;	/* 6 bytes */
1402 			break;
1403 		case 6:
1404 			vcpu->mmio_needed = 17;	/* 7 bytes */
1405 			break;
1406 		case 7:
1407 			vcpu->mmio_needed = 18;	/* 8 bytes */
1408 			break;
1409 		default:
1410 			break;
1411 		}
1412 		break;
1413 
1414 	case ldr_op:
1415 		run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1416 					vcpu->arch.host_cp0_badvaddr) & (~0x7);
1417 
1418 		run->mmio.len = 8;
1419 		imme = vcpu->arch.host_cp0_badvaddr & 0x7;
1420 		switch (imme) {
1421 		case 0:
1422 			vcpu->mmio_needed = 19;	/* 8 bytes */
1423 			break;
1424 		case 1:
1425 			vcpu->mmio_needed = 20;	/* 7 bytes */
1426 			break;
1427 		case 2:
1428 			vcpu->mmio_needed = 21;	/* 6 bytes */
1429 			break;
1430 		case 3:
1431 			vcpu->mmio_needed = 22;	/* 5 bytes */
1432 			break;
1433 		case 4:
1434 			vcpu->mmio_needed = 23;	/* 4 bytes */
1435 			break;
1436 		case 5:
1437 			vcpu->mmio_needed = 24;	/* 3 bytes */
1438 			break;
1439 		case 6:
1440 			vcpu->mmio_needed = 25;	/* 2 bytes */
1441 			break;
1442 		case 7:
1443 			vcpu->mmio_needed = 26;	/* 1 byte */
1444 			break;
1445 		default:
1446 			break;
1447 		}
1448 		break;
1449 #endif
1450 
1451 #ifdef CONFIG_CPU_LOONGSON64
1452 	case ldc2_op:
1453 		rt = inst.loongson3_lsdc2_format.rt;
1454 		switch (inst.loongson3_lsdc2_format.opcode1) {
1455 		/*
1456 		 * Loongson-3 overridden ldc2 instructions.
1457 		 * opcode1              instruction
1458 		 *   0x0          gslbx: store 1 bytes from GPR
1459 		 *   0x1          gslhx: store 2 bytes from GPR
1460 		 *   0x2          gslwx: store 4 bytes from GPR
1461 		 *   0x3          gsldx: store 8 bytes from GPR
1462 		 */
1463 		case 0x0:
1464 			run->mmio.len = 1;
1465 			vcpu->mmio_needed = 27;	/* signed */
1466 			break;
1467 		case 0x1:
1468 			run->mmio.len = 2;
1469 			vcpu->mmio_needed = 28;	/* signed */
1470 			break;
1471 		case 0x2:
1472 			run->mmio.len = 4;
1473 			vcpu->mmio_needed = 29;	/* signed */
1474 			break;
1475 		case 0x3:
1476 			run->mmio.len = 8;
1477 			vcpu->mmio_needed = 30;	/* signed */
1478 			break;
1479 		default:
1480 			kvm_err("Godson Extended GS-Load for float not yet supported (inst=0x%08x)\n",
1481 				inst.word);
1482 			break;
1483 		}
1484 		break;
1485 #endif
1486 
1487 	default:
1488 		kvm_err("Load not yet supported (inst=0x%08x)\n",
1489 			inst.word);
1490 		vcpu->mmio_needed = 0;
1491 		return EMULATE_FAIL;
1492 	}
1493 
1494 	run->mmio.is_write = 0;
1495 	vcpu->mmio_is_write = 0;
1496 
1497 	r = kvm_io_bus_read(vcpu, KVM_MMIO_BUS,
1498 			run->mmio.phys_addr, run->mmio.len, run->mmio.data);
1499 
1500 	if (!r) {
1501 		kvm_mips_complete_mmio_load(vcpu);
1502 		vcpu->mmio_needed = 0;
1503 		return EMULATE_DONE;
1504 	}
1505 
1506 	return EMULATE_DO_MMIO;
1507 }
1508 
1509 enum emulation_result kvm_mips_complete_mmio_load(struct kvm_vcpu *vcpu)
1510 {
1511 	struct kvm_run *run = vcpu->run;
1512 	unsigned long *gpr = &vcpu->arch.gprs[vcpu->arch.io_gpr];
1513 	enum emulation_result er = EMULATE_DONE;
1514 
1515 	if (run->mmio.len > sizeof(*gpr)) {
1516 		kvm_err("Bad MMIO length: %d", run->mmio.len);
1517 		er = EMULATE_FAIL;
1518 		goto done;
1519 	}
1520 
1521 	/* Restore saved resume PC */
1522 	vcpu->arch.pc = vcpu->arch.io_pc;
1523 
1524 	switch (run->mmio.len) {
1525 	case 8:
1526 		switch (vcpu->mmio_needed) {
1527 		case 11:
1528 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffffffff) |
1529 				(((*(s64 *)run->mmio.data) & 0xff) << 56);
1530 			break;
1531 		case 12:
1532 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffffff) |
1533 				(((*(s64 *)run->mmio.data) & 0xffff) << 48);
1534 			break;
1535 		case 13:
1536 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffff) |
1537 				(((*(s64 *)run->mmio.data) & 0xffffff) << 40);
1538 			break;
1539 		case 14:
1540 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffff) |
1541 				(((*(s64 *)run->mmio.data) & 0xffffffff) << 32);
1542 			break;
1543 		case 15:
1544 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffff) |
1545 				(((*(s64 *)run->mmio.data) & 0xffffffffff) << 24);
1546 			break;
1547 		case 16:
1548 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffff) |
1549 				(((*(s64 *)run->mmio.data) & 0xffffffffffff) << 16);
1550 			break;
1551 		case 17:
1552 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xff) |
1553 				(((*(s64 *)run->mmio.data) & 0xffffffffffffff) << 8);
1554 			break;
1555 		case 18:
1556 		case 19:
1557 			*gpr = *(s64 *)run->mmio.data;
1558 			break;
1559 		case 20:
1560 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xff00000000000000) |
1561 				((((*(s64 *)run->mmio.data)) >> 8) & 0xffffffffffffff);
1562 			break;
1563 		case 21:
1564 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffff000000000000) |
1565 				((((*(s64 *)run->mmio.data)) >> 16) & 0xffffffffffff);
1566 			break;
1567 		case 22:
1568 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffff0000000000) |
1569 				((((*(s64 *)run->mmio.data)) >> 24) & 0xffffffffff);
1570 			break;
1571 		case 23:
1572 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffff00000000) |
1573 				((((*(s64 *)run->mmio.data)) >> 32) & 0xffffffff);
1574 			break;
1575 		case 24:
1576 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffff000000) |
1577 				((((*(s64 *)run->mmio.data)) >> 40) & 0xffffff);
1578 			break;
1579 		case 25:
1580 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffffff0000) |
1581 				((((*(s64 *)run->mmio.data)) >> 48) & 0xffff);
1582 			break;
1583 		case 26:
1584 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffffffff00) |
1585 				((((*(s64 *)run->mmio.data)) >> 56) & 0xff);
1586 			break;
1587 		default:
1588 			*gpr = *(s64 *)run->mmio.data;
1589 		}
1590 		break;
1591 
1592 	case 4:
1593 		switch (vcpu->mmio_needed) {
1594 		case 1:
1595 			*gpr = *(u32 *)run->mmio.data;
1596 			break;
1597 		case 2:
1598 			*gpr = *(s32 *)run->mmio.data;
1599 			break;
1600 		case 3:
1601 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffff) |
1602 				(((*(s32 *)run->mmio.data) & 0xff) << 24);
1603 			break;
1604 		case 4:
1605 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffff) |
1606 				(((*(s32 *)run->mmio.data) & 0xffff) << 16);
1607 			break;
1608 		case 5:
1609 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xff) |
1610 				(((*(s32 *)run->mmio.data) & 0xffffff) << 8);
1611 			break;
1612 		case 6:
1613 		case 7:
1614 			*gpr = *(s32 *)run->mmio.data;
1615 			break;
1616 		case 8:
1617 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xff000000) |
1618 				((((*(s32 *)run->mmio.data)) >> 8) & 0xffffff);
1619 			break;
1620 		case 9:
1621 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffff0000) |
1622 				((((*(s32 *)run->mmio.data)) >> 16) & 0xffff);
1623 			break;
1624 		case 10:
1625 			*gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffff00) |
1626 				((((*(s32 *)run->mmio.data)) >> 24) & 0xff);
1627 			break;
1628 		default:
1629 			*gpr = *(s32 *)run->mmio.data;
1630 		}
1631 		break;
1632 
1633 	case 2:
1634 		if (vcpu->mmio_needed == 1)
1635 			*gpr = *(u16 *)run->mmio.data;
1636 		else
1637 			*gpr = *(s16 *)run->mmio.data;
1638 
1639 		break;
1640 	case 1:
1641 		if (vcpu->mmio_needed == 1)
1642 			*gpr = *(u8 *)run->mmio.data;
1643 		else
1644 			*gpr = *(s8 *)run->mmio.data;
1645 		break;
1646 	}
1647 
1648 done:
1649 	return er;
1650 }
1651