GNU Linux-libre 5.10.217-gnu1
[releases.git] / arch / mips / kvm / emulate.c
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_vcpu *vcpu)
1266 {
1267         struct mips_coproc *cop0 = vcpu->arch.cop0;
1268         enum emulation_result er = EMULATE_DONE;
1269         u32 rt, rd, sel;
1270         unsigned long curr_pc;
1271
1272         /*
1273          * Update PC and hold onto current PC in case there is
1274          * an error and we want to rollback the PC
1275          */
1276         curr_pc = vcpu->arch.pc;
1277         er = update_pc(vcpu, cause);
1278         if (er == EMULATE_FAIL)
1279                 return er;
1280
1281         if (inst.co_format.co) {
1282                 switch (inst.co_format.func) {
1283                 case tlbr_op:   /*  Read indexed TLB entry  */
1284                         er = kvm_mips_emul_tlbr(vcpu);
1285                         break;
1286                 case tlbwi_op:  /*  Write indexed  */
1287                         er = kvm_mips_emul_tlbwi(vcpu);
1288                         break;
1289                 case tlbwr_op:  /*  Write random  */
1290                         er = kvm_mips_emul_tlbwr(vcpu);
1291                         break;
1292                 case tlbp_op:   /* TLB Probe */
1293                         er = kvm_mips_emul_tlbp(vcpu);
1294                         break;
1295                 case rfe_op:
1296                         kvm_err("!!!COP0_RFE!!!\n");
1297                         break;
1298                 case eret_op:
1299                         er = kvm_mips_emul_eret(vcpu);
1300                         goto dont_update_pc;
1301                 case wait_op:
1302                         er = kvm_mips_emul_wait(vcpu);
1303                         break;
1304                 case hypcall_op:
1305                         er = kvm_mips_emul_hypcall(vcpu, inst);
1306                         break;
1307                 }
1308         } else {
1309                 rt = inst.c0r_format.rt;
1310                 rd = inst.c0r_format.rd;
1311                 sel = inst.c0r_format.sel;
1312
1313                 switch (inst.c0r_format.rs) {
1314                 case mfc_op:
1315 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
1316                         cop0->stat[rd][sel]++;
1317 #endif
1318                         /* Get reg */
1319                         if ((rd == MIPS_CP0_COUNT) && (sel == 0)) {
1320                                 vcpu->arch.gprs[rt] =
1321                                     (s32)kvm_mips_read_count(vcpu);
1322                         } else if ((rd == MIPS_CP0_ERRCTL) && (sel == 0)) {
1323                                 vcpu->arch.gprs[rt] = 0x0;
1324 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1325                                 kvm_mips_trans_mfc0(inst, opc, vcpu);
1326 #endif
1327                         } else {
1328                                 vcpu->arch.gprs[rt] = (s32)cop0->reg[rd][sel];
1329
1330 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1331                                 kvm_mips_trans_mfc0(inst, opc, vcpu);
1332 #endif
1333                         }
1334
1335                         trace_kvm_hwr(vcpu, KVM_TRACE_MFC0,
1336                                       KVM_TRACE_COP0(rd, sel),
1337                                       vcpu->arch.gprs[rt]);
1338                         break;
1339
1340                 case dmfc_op:
1341                         vcpu->arch.gprs[rt] = cop0->reg[rd][sel];
1342
1343                         trace_kvm_hwr(vcpu, KVM_TRACE_DMFC0,
1344                                       KVM_TRACE_COP0(rd, sel),
1345                                       vcpu->arch.gprs[rt]);
1346                         break;
1347
1348                 case mtc_op:
1349 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
1350                         cop0->stat[rd][sel]++;
1351 #endif
1352                         trace_kvm_hwr(vcpu, KVM_TRACE_MTC0,
1353                                       KVM_TRACE_COP0(rd, sel),
1354                                       vcpu->arch.gprs[rt]);
1355
1356                         if ((rd == MIPS_CP0_TLB_INDEX)
1357                             && (vcpu->arch.gprs[rt] >=
1358                                 KVM_MIPS_GUEST_TLB_SIZE)) {
1359                                 kvm_err("Invalid TLB Index: %ld",
1360                                         vcpu->arch.gprs[rt]);
1361                                 er = EMULATE_FAIL;
1362                                 break;
1363                         }
1364                         if ((rd == MIPS_CP0_PRID) && (sel == 1)) {
1365                                 /*
1366                                  * Preserve core number, and keep the exception
1367                                  * base in guest KSeg0.
1368                                  */
1369                                 kvm_change_c0_guest_ebase(cop0, 0x1ffff000,
1370                                                           vcpu->arch.gprs[rt]);
1371                         } else if (rd == MIPS_CP0_TLB_HI && sel == 0) {
1372                                 kvm_mips_change_entryhi(vcpu,
1373                                                         vcpu->arch.gprs[rt]);
1374                         }
1375                         /* Are we writing to COUNT */
1376                         else if ((rd == MIPS_CP0_COUNT) && (sel == 0)) {
1377                                 kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
1378                                 goto done;
1379                         } else if ((rd == MIPS_CP0_COMPARE) && (sel == 0)) {
1380                                 /* If we are writing to COMPARE */
1381                                 /* Clear pending timer interrupt, if any */
1382                                 kvm_mips_write_compare(vcpu,
1383                                                        vcpu->arch.gprs[rt],
1384                                                        true);
1385                         } else if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1386                                 unsigned int old_val, val, change;
1387
1388                                 old_val = kvm_read_c0_guest_status(cop0);
1389                                 val = vcpu->arch.gprs[rt];
1390                                 change = val ^ old_val;
1391
1392                                 /* Make sure that the NMI bit is never set */
1393                                 val &= ~ST0_NMI;
1394
1395                                 /*
1396                                  * Don't allow CU1 or FR to be set unless FPU
1397                                  * capability enabled and exists in guest
1398                                  * configuration.
1399                                  */
1400                                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1401                                         val &= ~(ST0_CU1 | ST0_FR);
1402
1403                                 /*
1404                                  * Also don't allow FR to be set if host doesn't
1405                                  * support it.
1406                                  */
1407                                 if (!(current_cpu_data.fpu_id & MIPS_FPIR_F64))
1408                                         val &= ~ST0_FR;
1409
1410
1411                                 /* Handle changes in FPU mode */
1412                                 preempt_disable();
1413
1414                                 /*
1415                                  * FPU and Vector register state is made
1416                                  * UNPREDICTABLE by a change of FR, so don't
1417                                  * even bother saving it.
1418                                  */
1419                                 if (change & ST0_FR)
1420                                         kvm_drop_fpu(vcpu);
1421
1422                                 /*
1423                                  * If MSA state is already live, it is undefined
1424                                  * how it interacts with FR=0 FPU state, and we
1425                                  * don't want to hit reserved instruction
1426                                  * exceptions trying to save the MSA state later
1427                                  * when CU=1 && FR=1, so play it safe and save
1428                                  * it first.
1429                                  */
1430                                 if (change & ST0_CU1 && !(val & ST0_FR) &&
1431                                     vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1432                                         kvm_lose_fpu(vcpu);
1433
1434                                 /*
1435                                  * Propagate CU1 (FPU enable) changes
1436                                  * immediately if the FPU context is already
1437                                  * loaded. When disabling we leave the context
1438                                  * loaded so it can be quickly enabled again in
1439                                  * the near future.
1440                                  */
1441                                 if (change & ST0_CU1 &&
1442                                     vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1443                                         change_c0_status(ST0_CU1, val);
1444
1445                                 preempt_enable();
1446
1447                                 kvm_write_c0_guest_status(cop0, val);
1448
1449 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1450                                 /*
1451                                  * If FPU present, we need CU1/FR bits to take
1452                                  * effect fairly soon.
1453                                  */
1454                                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1455                                         kvm_mips_trans_mtc0(inst, opc, vcpu);
1456 #endif
1457                         } else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1458                                 unsigned int old_val, val, change, wrmask;
1459
1460                                 old_val = kvm_read_c0_guest_config5(cop0);
1461                                 val = vcpu->arch.gprs[rt];
1462
1463                                 /* Only a few bits are writable in Config5 */
1464                                 wrmask = kvm_mips_config5_wrmask(vcpu);
1465                                 change = (val ^ old_val) & wrmask;
1466                                 val = old_val ^ change;
1467
1468
1469                                 /* Handle changes in FPU/MSA modes */
1470                                 preempt_disable();
1471
1472                                 /*
1473                                  * Propagate FRE changes immediately if the FPU
1474                                  * context is already loaded.
1475                                  */
1476                                 if (change & MIPS_CONF5_FRE &&
1477                                     vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1478                                         change_c0_config5(MIPS_CONF5_FRE, val);
1479
1480                                 /*
1481                                  * Propagate MSAEn changes immediately if the
1482                                  * MSA context is already loaded. When disabling
1483                                  * we leave the context loaded so it can be
1484                                  * quickly enabled again in the near future.
1485                                  */
1486                                 if (change & MIPS_CONF5_MSAEN &&
1487                                     vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1488                                         change_c0_config5(MIPS_CONF5_MSAEN,
1489                                                           val);
1490
1491                                 preempt_enable();
1492
1493                                 kvm_write_c0_guest_config5(cop0, val);
1494                         } else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1495                                 u32 old_cause, new_cause;
1496
1497                                 old_cause = kvm_read_c0_guest_cause(cop0);
1498                                 new_cause = vcpu->arch.gprs[rt];
1499                                 /* Update R/W bits */
1500                                 kvm_change_c0_guest_cause(cop0, 0x08800300,
1501                                                           new_cause);
1502                                 /* DC bit enabling/disabling timer? */
1503                                 if ((old_cause ^ new_cause) & CAUSEF_DC) {
1504                                         if (new_cause & CAUSEF_DC)
1505                                                 kvm_mips_count_disable_cause(vcpu);
1506                                         else
1507                                                 kvm_mips_count_enable_cause(vcpu);
1508                                 }
1509                         } else if ((rd == MIPS_CP0_HWRENA) && (sel == 0)) {
1510                                 u32 mask = MIPS_HWRENA_CPUNUM |
1511                                            MIPS_HWRENA_SYNCISTEP |
1512                                            MIPS_HWRENA_CC |
1513                                            MIPS_HWRENA_CCRES;
1514
1515                                 if (kvm_read_c0_guest_config3(cop0) &
1516                                     MIPS_CONF3_ULRI)
1517                                         mask |= MIPS_HWRENA_ULR;
1518                                 cop0->reg[rd][sel] = vcpu->arch.gprs[rt] & mask;
1519                         } else {
1520                                 cop0->reg[rd][sel] = vcpu->arch.gprs[rt];
1521 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1522                                 kvm_mips_trans_mtc0(inst, opc, vcpu);
1523 #endif
1524                         }
1525                         break;
1526
1527                 case dmtc_op:
1528                         kvm_err("!!!!!!![%#lx]dmtc_op: rt: %d, rd: %d, sel: %d!!!!!!\n",
1529                                 vcpu->arch.pc, rt, rd, sel);
1530                         trace_kvm_hwr(vcpu, KVM_TRACE_DMTC0,
1531                                       KVM_TRACE_COP0(rd, sel),
1532                                       vcpu->arch.gprs[rt]);
1533                         er = EMULATE_FAIL;
1534                         break;
1535
1536                 case mfmc0_op:
1537 #ifdef KVM_MIPS_DEBUG_COP0_COUNTERS
1538                         cop0->stat[MIPS_CP0_STATUS][0]++;
1539 #endif
1540                         if (rt != 0)
1541                                 vcpu->arch.gprs[rt] =
1542                                     kvm_read_c0_guest_status(cop0);
1543                         /* EI */
1544                         if (inst.mfmc0_format.sc) {
1545                                 kvm_debug("[%#lx] mfmc0_op: EI\n",
1546                                           vcpu->arch.pc);
1547                                 kvm_set_c0_guest_status(cop0, ST0_IE);
1548                         } else {
1549                                 kvm_debug("[%#lx] mfmc0_op: DI\n",
1550                                           vcpu->arch.pc);
1551                                 kvm_clear_c0_guest_status(cop0, ST0_IE);
1552                         }
1553
1554                         break;
1555
1556                 case wrpgpr_op:
1557                         {
1558                                 u32 css = cop0->reg[MIPS_CP0_STATUS][2] & 0xf;
1559                                 u32 pss =
1560                                     (cop0->reg[MIPS_CP0_STATUS][2] >> 6) & 0xf;
1561                                 /*
1562                                  * We don't support any shadow register sets, so
1563                                  * SRSCtl[PSS] == SRSCtl[CSS] = 0
1564                                  */
1565                                 if (css || pss) {
1566                                         er = EMULATE_FAIL;
1567                                         break;
1568                                 }
1569                                 kvm_debug("WRPGPR[%d][%d] = %#lx\n", pss, rd,
1570                                           vcpu->arch.gprs[rt]);
1571                                 vcpu->arch.gprs[rd] = vcpu->arch.gprs[rt];
1572                         }
1573                         break;
1574                 default:
1575                         kvm_err("[%#lx]MachEmulateCP0: unsupported COP0, copz: 0x%x\n",
1576                                 vcpu->arch.pc, inst.c0r_format.rs);
1577                         er = EMULATE_FAIL;
1578                         break;
1579                 }
1580         }
1581
1582 done:
1583         /* Rollback PC only if emulation was unsuccessful */
1584         if (er == EMULATE_FAIL)
1585                 vcpu->arch.pc = curr_pc;
1586
1587 dont_update_pc:
1588         /*
1589          * This is for special instructions whose emulation
1590          * updates the PC, so do not overwrite the PC under
1591          * any circumstances
1592          */
1593
1594         return er;
1595 }
1596
1597 enum emulation_result kvm_mips_emulate_store(union mips_instruction inst,
1598                                              u32 cause,
1599                                              struct kvm_vcpu *vcpu)
1600 {
1601         int r;
1602         enum emulation_result er;
1603         u32 rt;
1604         struct kvm_run *run = vcpu->run;
1605         void *data = run->mmio.data;
1606         unsigned int imme;
1607         unsigned long curr_pc;
1608
1609         /*
1610          * Update PC and hold onto current PC in case there is
1611          * an error and we want to rollback the PC
1612          */
1613         curr_pc = vcpu->arch.pc;
1614         er = update_pc(vcpu, cause);
1615         if (er == EMULATE_FAIL)
1616                 return er;
1617
1618         rt = inst.i_format.rt;
1619
1620         run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1621                                                 vcpu->arch.host_cp0_badvaddr);
1622         if (run->mmio.phys_addr == KVM_INVALID_ADDR)
1623                 goto out_fail;
1624
1625         switch (inst.i_format.opcode) {
1626 #if defined(CONFIG_64BIT) && defined(CONFIG_KVM_MIPS_VZ)
1627         case sd_op:
1628                 run->mmio.len = 8;
1629                 *(u64 *)data = vcpu->arch.gprs[rt];
1630
1631                 kvm_debug("[%#lx] OP_SD: eaddr: %#lx, gpr: %#lx, data: %#llx\n",
1632                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1633                           vcpu->arch.gprs[rt], *(u64 *)data);
1634                 break;
1635 #endif
1636
1637         case sw_op:
1638                 run->mmio.len = 4;
1639                 *(u32 *)data = vcpu->arch.gprs[rt];
1640
1641                 kvm_debug("[%#lx] OP_SW: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1642                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1643                           vcpu->arch.gprs[rt], *(u32 *)data);
1644                 break;
1645
1646         case sh_op:
1647                 run->mmio.len = 2;
1648                 *(u16 *)data = vcpu->arch.gprs[rt];
1649
1650                 kvm_debug("[%#lx] OP_SH: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1651                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1652                           vcpu->arch.gprs[rt], *(u16 *)data);
1653                 break;
1654
1655         case sb_op:
1656                 run->mmio.len = 1;
1657                 *(u8 *)data = vcpu->arch.gprs[rt];
1658
1659                 kvm_debug("[%#lx] OP_SB: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1660                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1661                           vcpu->arch.gprs[rt], *(u8 *)data);
1662                 break;
1663
1664         case swl_op:
1665                 run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1666                                         vcpu->arch.host_cp0_badvaddr) & (~0x3);
1667                 run->mmio.len = 4;
1668                 imme = vcpu->arch.host_cp0_badvaddr & 0x3;
1669                 switch (imme) {
1670                 case 0:
1671                         *(u32 *)data = ((*(u32 *)data) & 0xffffff00) |
1672                                         (vcpu->arch.gprs[rt] >> 24);
1673                         break;
1674                 case 1:
1675                         *(u32 *)data = ((*(u32 *)data) & 0xffff0000) |
1676                                         (vcpu->arch.gprs[rt] >> 16);
1677                         break;
1678                 case 2:
1679                         *(u32 *)data = ((*(u32 *)data) & 0xff000000) |
1680                                         (vcpu->arch.gprs[rt] >> 8);
1681                         break;
1682                 case 3:
1683                         *(u32 *)data = vcpu->arch.gprs[rt];
1684                         break;
1685                 default:
1686                         break;
1687                 }
1688
1689                 kvm_debug("[%#lx] OP_SWL: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1690                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1691                           vcpu->arch.gprs[rt], *(u32 *)data);
1692                 break;
1693
1694         case swr_op:
1695                 run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1696                                         vcpu->arch.host_cp0_badvaddr) & (~0x3);
1697                 run->mmio.len = 4;
1698                 imme = vcpu->arch.host_cp0_badvaddr & 0x3;
1699                 switch (imme) {
1700                 case 0:
1701                         *(u32 *)data = vcpu->arch.gprs[rt];
1702                         break;
1703                 case 1:
1704                         *(u32 *)data = ((*(u32 *)data) & 0xff) |
1705                                         (vcpu->arch.gprs[rt] << 8);
1706                         break;
1707                 case 2:
1708                         *(u32 *)data = ((*(u32 *)data) & 0xffff) |
1709                                         (vcpu->arch.gprs[rt] << 16);
1710                         break;
1711                 case 3:
1712                         *(u32 *)data = ((*(u32 *)data) & 0xffffff) |
1713                                         (vcpu->arch.gprs[rt] << 24);
1714                         break;
1715                 default:
1716                         break;
1717                 }
1718
1719                 kvm_debug("[%#lx] OP_SWR: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1720                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1721                           vcpu->arch.gprs[rt], *(u32 *)data);
1722                 break;
1723
1724 #if defined(CONFIG_64BIT) && defined(CONFIG_KVM_MIPS_VZ)
1725         case sdl_op:
1726                 run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1727                                         vcpu->arch.host_cp0_badvaddr) & (~0x7);
1728
1729                 run->mmio.len = 8;
1730                 imme = vcpu->arch.host_cp0_badvaddr & 0x7;
1731                 switch (imme) {
1732                 case 0:
1733                         *(u64 *)data = ((*(u64 *)data) & 0xffffffffffffff00) |
1734                                         ((vcpu->arch.gprs[rt] >> 56) & 0xff);
1735                         break;
1736                 case 1:
1737                         *(u64 *)data = ((*(u64 *)data) & 0xffffffffffff0000) |
1738                                         ((vcpu->arch.gprs[rt] >> 48) & 0xffff);
1739                         break;
1740                 case 2:
1741                         *(u64 *)data = ((*(u64 *)data) & 0xffffffffff000000) |
1742                                         ((vcpu->arch.gprs[rt] >> 40) & 0xffffff);
1743                         break;
1744                 case 3:
1745                         *(u64 *)data = ((*(u64 *)data) & 0xffffffff00000000) |
1746                                         ((vcpu->arch.gprs[rt] >> 32) & 0xffffffff);
1747                         break;
1748                 case 4:
1749                         *(u64 *)data = ((*(u64 *)data) & 0xffffff0000000000) |
1750                                         ((vcpu->arch.gprs[rt] >> 24) & 0xffffffffff);
1751                         break;
1752                 case 5:
1753                         *(u64 *)data = ((*(u64 *)data) & 0xffff000000000000) |
1754                                         ((vcpu->arch.gprs[rt] >> 16) & 0xffffffffffff);
1755                         break;
1756                 case 6:
1757                         *(u64 *)data = ((*(u64 *)data) & 0xff00000000000000) |
1758                                         ((vcpu->arch.gprs[rt] >> 8) & 0xffffffffffffff);
1759                         break;
1760                 case 7:
1761                         *(u64 *)data = vcpu->arch.gprs[rt];
1762                         break;
1763                 default:
1764                         break;
1765                 }
1766
1767                 kvm_debug("[%#lx] OP_SDL: eaddr: %#lx, gpr: %#lx, data: %llx\n",
1768                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1769                           vcpu->arch.gprs[rt], *(u64 *)data);
1770                 break;
1771
1772         case sdr_op:
1773                 run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1774                                         vcpu->arch.host_cp0_badvaddr) & (~0x7);
1775
1776                 run->mmio.len = 8;
1777                 imme = vcpu->arch.host_cp0_badvaddr & 0x7;
1778                 switch (imme) {
1779                 case 0:
1780                         *(u64 *)data = vcpu->arch.gprs[rt];
1781                         break;
1782                 case 1:
1783                         *(u64 *)data = ((*(u64 *)data) & 0xff) |
1784                                         (vcpu->arch.gprs[rt] << 8);
1785                         break;
1786                 case 2:
1787                         *(u64 *)data = ((*(u64 *)data) & 0xffff) |
1788                                         (vcpu->arch.gprs[rt] << 16);
1789                         break;
1790                 case 3:
1791                         *(u64 *)data = ((*(u64 *)data) & 0xffffff) |
1792                                         (vcpu->arch.gprs[rt] << 24);
1793                         break;
1794                 case 4:
1795                         *(u64 *)data = ((*(u64 *)data) & 0xffffffff) |
1796                                         (vcpu->arch.gprs[rt] << 32);
1797                         break;
1798                 case 5:
1799                         *(u64 *)data = ((*(u64 *)data) & 0xffffffffff) |
1800                                         (vcpu->arch.gprs[rt] << 40);
1801                         break;
1802                 case 6:
1803                         *(u64 *)data = ((*(u64 *)data) & 0xffffffffffff) |
1804                                         (vcpu->arch.gprs[rt] << 48);
1805                         break;
1806                 case 7:
1807                         *(u64 *)data = ((*(u64 *)data) & 0xffffffffffffff) |
1808                                         (vcpu->arch.gprs[rt] << 56);
1809                         break;
1810                 default:
1811                         break;
1812                 }
1813
1814                 kvm_debug("[%#lx] OP_SDR: eaddr: %#lx, gpr: %#lx, data: %llx\n",
1815                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1816                           vcpu->arch.gprs[rt], *(u64 *)data);
1817                 break;
1818 #endif
1819
1820 #ifdef CONFIG_CPU_LOONGSON64
1821         case sdc2_op:
1822                 rt = inst.loongson3_lsdc2_format.rt;
1823                 switch (inst.loongson3_lsdc2_format.opcode1) {
1824                 /*
1825                  * Loongson-3 overridden sdc2 instructions.
1826                  * opcode1              instruction
1827                  *   0x0          gssbx: store 1 bytes from GPR
1828                  *   0x1          gsshx: store 2 bytes from GPR
1829                  *   0x2          gsswx: store 4 bytes from GPR
1830                  *   0x3          gssdx: store 8 bytes from GPR
1831                  */
1832                 case 0x0:
1833                         run->mmio.len = 1;
1834                         *(u8 *)data = vcpu->arch.gprs[rt];
1835
1836                         kvm_debug("[%#lx] OP_GSSBX: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1837                                   vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1838                                   vcpu->arch.gprs[rt], *(u8 *)data);
1839                         break;
1840                 case 0x1:
1841                         run->mmio.len = 2;
1842                         *(u16 *)data = vcpu->arch.gprs[rt];
1843
1844                         kvm_debug("[%#lx] OP_GSSSHX: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1845                                   vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1846                                   vcpu->arch.gprs[rt], *(u16 *)data);
1847                         break;
1848                 case 0x2:
1849                         run->mmio.len = 4;
1850                         *(u32 *)data = vcpu->arch.gprs[rt];
1851
1852                         kvm_debug("[%#lx] OP_GSSWX: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1853                                   vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1854                                   vcpu->arch.gprs[rt], *(u32 *)data);
1855                         break;
1856                 case 0x3:
1857                         run->mmio.len = 8;
1858                         *(u64 *)data = vcpu->arch.gprs[rt];
1859
1860                         kvm_debug("[%#lx] OP_GSSDX: eaddr: %#lx, gpr: %#lx, data: %#llx\n",
1861                                   vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1862                                   vcpu->arch.gprs[rt], *(u64 *)data);
1863                         break;
1864                 default:
1865                         kvm_err("Godson Extended GS-Store not yet supported (inst=0x%08x)\n",
1866                                 inst.word);
1867                         break;
1868                 }
1869                 break;
1870 #endif
1871         default:
1872                 kvm_err("Store not yet supported (inst=0x%08x)\n",
1873                         inst.word);
1874                 goto out_fail;
1875         }
1876
1877         vcpu->mmio_needed = 1;
1878         run->mmio.is_write = 1;
1879         vcpu->mmio_is_write = 1;
1880
1881         r = kvm_io_bus_write(vcpu, KVM_MMIO_BUS,
1882                         run->mmio.phys_addr, run->mmio.len, data);
1883
1884         if (!r) {
1885                 vcpu->mmio_needed = 0;
1886                 return EMULATE_DONE;
1887         }
1888
1889         return EMULATE_DO_MMIO;
1890
1891 out_fail:
1892         /* Rollback PC if emulation was unsuccessful */
1893         vcpu->arch.pc = curr_pc;
1894         return EMULATE_FAIL;
1895 }
1896
1897 enum emulation_result kvm_mips_emulate_load(union mips_instruction inst,
1898                                             u32 cause, struct kvm_vcpu *vcpu)
1899 {
1900         struct kvm_run *run = vcpu->run;
1901         int r;
1902         enum emulation_result er;
1903         unsigned long curr_pc;
1904         u32 op, rt;
1905         unsigned int imme;
1906
1907         rt = inst.i_format.rt;
1908         op = inst.i_format.opcode;
1909
1910         /*
1911          * Find the resume PC now while we have safe and easy access to the
1912          * prior branch instruction, and save it for
1913          * kvm_mips_complete_mmio_load() to restore later.
1914          */
1915         curr_pc = vcpu->arch.pc;
1916         er = update_pc(vcpu, cause);
1917         if (er == EMULATE_FAIL)
1918                 return er;
1919         vcpu->arch.io_pc = vcpu->arch.pc;
1920         vcpu->arch.pc = curr_pc;
1921
1922         vcpu->arch.io_gpr = rt;
1923
1924         run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1925                                                 vcpu->arch.host_cp0_badvaddr);
1926         if (run->mmio.phys_addr == KVM_INVALID_ADDR)
1927                 return EMULATE_FAIL;
1928
1929         vcpu->mmio_needed = 2;  /* signed */
1930         switch (op) {
1931 #if defined(CONFIG_64BIT) && defined(CONFIG_KVM_MIPS_VZ)
1932         case ld_op:
1933                 run->mmio.len = 8;
1934                 break;
1935
1936         case lwu_op:
1937                 vcpu->mmio_needed = 1;  /* unsigned */
1938                 fallthrough;
1939 #endif
1940         case lw_op:
1941                 run->mmio.len = 4;
1942                 break;
1943
1944         case lhu_op:
1945                 vcpu->mmio_needed = 1;  /* unsigned */
1946                 fallthrough;
1947         case lh_op:
1948                 run->mmio.len = 2;
1949                 break;
1950
1951         case lbu_op:
1952                 vcpu->mmio_needed = 1;  /* unsigned */
1953                 fallthrough;
1954         case lb_op:
1955                 run->mmio.len = 1;
1956                 break;
1957
1958         case lwl_op:
1959                 run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1960                                         vcpu->arch.host_cp0_badvaddr) & (~0x3);
1961
1962                 run->mmio.len = 4;
1963                 imme = vcpu->arch.host_cp0_badvaddr & 0x3;
1964                 switch (imme) {
1965                 case 0:
1966                         vcpu->mmio_needed = 3;  /* 1 byte */
1967                         break;
1968                 case 1:
1969                         vcpu->mmio_needed = 4;  /* 2 bytes */
1970                         break;
1971                 case 2:
1972                         vcpu->mmio_needed = 5;  /* 3 bytes */
1973                         break;
1974                 case 3:
1975                         vcpu->mmio_needed = 6;  /* 4 bytes */
1976                         break;
1977                 default:
1978                         break;
1979                 }
1980                 break;
1981
1982         case lwr_op:
1983                 run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
1984                                         vcpu->arch.host_cp0_badvaddr) & (~0x3);
1985
1986                 run->mmio.len = 4;
1987                 imme = vcpu->arch.host_cp0_badvaddr & 0x3;
1988                 switch (imme) {
1989                 case 0:
1990                         vcpu->mmio_needed = 7;  /* 4 bytes */
1991                         break;
1992                 case 1:
1993                         vcpu->mmio_needed = 8;  /* 3 bytes */
1994                         break;
1995                 case 2:
1996                         vcpu->mmio_needed = 9;  /* 2 bytes */
1997                         break;
1998                 case 3:
1999                         vcpu->mmio_needed = 10; /* 1 byte */
2000                         break;
2001                 default:
2002                         break;
2003                 }
2004                 break;
2005
2006 #if defined(CONFIG_64BIT) && defined(CONFIG_KVM_MIPS_VZ)
2007         case ldl_op:
2008                 run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
2009                                         vcpu->arch.host_cp0_badvaddr) & (~0x7);
2010
2011                 run->mmio.len = 8;
2012                 imme = vcpu->arch.host_cp0_badvaddr & 0x7;
2013                 switch (imme) {
2014                 case 0:
2015                         vcpu->mmio_needed = 11; /* 1 byte */
2016                         break;
2017                 case 1:
2018                         vcpu->mmio_needed = 12; /* 2 bytes */
2019                         break;
2020                 case 2:
2021                         vcpu->mmio_needed = 13; /* 3 bytes */
2022                         break;
2023                 case 3:
2024                         vcpu->mmio_needed = 14; /* 4 bytes */
2025                         break;
2026                 case 4:
2027                         vcpu->mmio_needed = 15; /* 5 bytes */
2028                         break;
2029                 case 5:
2030                         vcpu->mmio_needed = 16; /* 6 bytes */
2031                         break;
2032                 case 6:
2033                         vcpu->mmio_needed = 17; /* 7 bytes */
2034                         break;
2035                 case 7:
2036                         vcpu->mmio_needed = 18; /* 8 bytes */
2037                         break;
2038                 default:
2039                         break;
2040                 }
2041                 break;
2042
2043         case ldr_op:
2044                 run->mmio.phys_addr = kvm_mips_callbacks->gva_to_gpa(
2045                                         vcpu->arch.host_cp0_badvaddr) & (~0x7);
2046
2047                 run->mmio.len = 8;
2048                 imme = vcpu->arch.host_cp0_badvaddr & 0x7;
2049                 switch (imme) {
2050                 case 0:
2051                         vcpu->mmio_needed = 19; /* 8 bytes */
2052                         break;
2053                 case 1:
2054                         vcpu->mmio_needed = 20; /* 7 bytes */
2055                         break;
2056                 case 2:
2057                         vcpu->mmio_needed = 21; /* 6 bytes */
2058                         break;
2059                 case 3:
2060                         vcpu->mmio_needed = 22; /* 5 bytes */
2061                         break;
2062                 case 4:
2063                         vcpu->mmio_needed = 23; /* 4 bytes */
2064                         break;
2065                 case 5:
2066                         vcpu->mmio_needed = 24; /* 3 bytes */
2067                         break;
2068                 case 6:
2069                         vcpu->mmio_needed = 25; /* 2 bytes */
2070                         break;
2071                 case 7:
2072                         vcpu->mmio_needed = 26; /* 1 byte */
2073                         break;
2074                 default:
2075                         break;
2076                 }
2077                 break;
2078 #endif
2079
2080 #ifdef CONFIG_CPU_LOONGSON64
2081         case ldc2_op:
2082                 rt = inst.loongson3_lsdc2_format.rt;
2083                 switch (inst.loongson3_lsdc2_format.opcode1) {
2084                 /*
2085                  * Loongson-3 overridden ldc2 instructions.
2086                  * opcode1              instruction
2087                  *   0x0          gslbx: store 1 bytes from GPR
2088                  *   0x1          gslhx: store 2 bytes from GPR
2089                  *   0x2          gslwx: store 4 bytes from GPR
2090                  *   0x3          gsldx: store 8 bytes from GPR
2091                  */
2092                 case 0x0:
2093                         run->mmio.len = 1;
2094                         vcpu->mmio_needed = 27; /* signed */
2095                         break;
2096                 case 0x1:
2097                         run->mmio.len = 2;
2098                         vcpu->mmio_needed = 28; /* signed */
2099                         break;
2100                 case 0x2:
2101                         run->mmio.len = 4;
2102                         vcpu->mmio_needed = 29; /* signed */
2103                         break;
2104                 case 0x3:
2105                         run->mmio.len = 8;
2106                         vcpu->mmio_needed = 30; /* signed */
2107                         break;
2108                 default:
2109                         kvm_err("Godson Extended GS-Load for float not yet supported (inst=0x%08x)\n",
2110                                 inst.word);
2111                         break;
2112                 }
2113                 break;
2114 #endif
2115
2116         default:
2117                 kvm_err("Load not yet supported (inst=0x%08x)\n",
2118                         inst.word);
2119                 vcpu->mmio_needed = 0;
2120                 return EMULATE_FAIL;
2121         }
2122
2123         run->mmio.is_write = 0;
2124         vcpu->mmio_is_write = 0;
2125
2126         r = kvm_io_bus_read(vcpu, KVM_MMIO_BUS,
2127                         run->mmio.phys_addr, run->mmio.len, run->mmio.data);
2128
2129         if (!r) {
2130                 kvm_mips_complete_mmio_load(vcpu);
2131                 vcpu->mmio_needed = 0;
2132                 return EMULATE_DONE;
2133         }
2134
2135         return EMULATE_DO_MMIO;
2136 }
2137
2138 #ifndef CONFIG_KVM_MIPS_VZ
2139 static enum emulation_result kvm_mips_guest_cache_op(int (*fn)(unsigned long),
2140                                                      unsigned long curr_pc,
2141                                                      unsigned long addr,
2142                                                      struct kvm_vcpu *vcpu,
2143                                                      u32 cause)
2144 {
2145         int err;
2146
2147         for (;;) {
2148                 /* Carefully attempt the cache operation */
2149                 kvm_trap_emul_gva_lockless_begin(vcpu);
2150                 err = fn(addr);
2151                 kvm_trap_emul_gva_lockless_end(vcpu);
2152
2153                 if (likely(!err))
2154                         return EMULATE_DONE;
2155
2156                 /*
2157                  * Try to handle the fault and retry, maybe we just raced with a
2158                  * GVA invalidation.
2159                  */
2160                 switch (kvm_trap_emul_gva_fault(vcpu, addr, false)) {
2161                 case KVM_MIPS_GVA:
2162                 case KVM_MIPS_GPA:
2163                         /* bad virtual or physical address */
2164                         return EMULATE_FAIL;
2165                 case KVM_MIPS_TLB:
2166                         /* no matching guest TLB */
2167                         vcpu->arch.host_cp0_badvaddr = addr;
2168                         vcpu->arch.pc = curr_pc;
2169                         kvm_mips_emulate_tlbmiss_ld(cause, NULL, vcpu);
2170                         return EMULATE_EXCEPT;
2171                 case KVM_MIPS_TLBINV:
2172                         /* invalid matching guest TLB */
2173                         vcpu->arch.host_cp0_badvaddr = addr;
2174                         vcpu->arch.pc = curr_pc;
2175                         kvm_mips_emulate_tlbinv_ld(cause, NULL, vcpu);
2176                         return EMULATE_EXCEPT;
2177                 default:
2178                         break;
2179                 }
2180         }
2181 }
2182
2183 enum emulation_result kvm_mips_emulate_cache(union mips_instruction inst,
2184                                              u32 *opc, u32 cause,
2185                                              struct kvm_vcpu *vcpu)
2186 {
2187         enum emulation_result er = EMULATE_DONE;
2188         u32 cache, op_inst, op, base;
2189         s16 offset;
2190         struct kvm_vcpu_arch *arch = &vcpu->arch;
2191         unsigned long va;
2192         unsigned long curr_pc;
2193
2194         /*
2195          * Update PC and hold onto current PC in case there is
2196          * an error and we want to rollback the PC
2197          */
2198         curr_pc = vcpu->arch.pc;
2199         er = update_pc(vcpu, cause);
2200         if (er == EMULATE_FAIL)
2201                 return er;
2202
2203         base = inst.i_format.rs;
2204         op_inst = inst.i_format.rt;
2205         if (cpu_has_mips_r6)
2206                 offset = inst.spec3_format.simmediate;
2207         else
2208                 offset = inst.i_format.simmediate;
2209         cache = op_inst & CacheOp_Cache;
2210         op = op_inst & CacheOp_Op;
2211
2212         va = arch->gprs[base] + offset;
2213
2214         kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
2215                   cache, op, base, arch->gprs[base], offset);
2216
2217         /*
2218          * Treat INDEX_INV as a nop, basically issued by Linux on startup to
2219          * invalidate the caches entirely by stepping through all the
2220          * ways/indexes
2221          */
2222         if (op == Index_Writeback_Inv) {
2223                 kvm_debug("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
2224                           vcpu->arch.pc, vcpu->arch.gprs[31], cache, op, base,
2225                           arch->gprs[base], offset);
2226
2227                 if (cache == Cache_D) {
2228 #ifdef CONFIG_CPU_R4K_CACHE_TLB
2229                         r4k_blast_dcache();
2230 #else
2231                         switch (boot_cpu_type()) {
2232                         case CPU_CAVIUM_OCTEON3:
2233                                 /* locally flush icache */
2234                                 local_flush_icache_range(0, 0);
2235                                 break;
2236                         default:
2237                                 __flush_cache_all();
2238                                 break;
2239                         }
2240 #endif
2241                 } else if (cache == Cache_I) {
2242 #ifdef CONFIG_CPU_R4K_CACHE_TLB
2243                         r4k_blast_icache();
2244 #else
2245                         switch (boot_cpu_type()) {
2246                         case CPU_CAVIUM_OCTEON3:
2247                                 /* locally flush icache */
2248                                 local_flush_icache_range(0, 0);
2249                                 break;
2250                         default:
2251                                 flush_icache_all();
2252                                 break;
2253                         }
2254 #endif
2255                 } else {
2256                         kvm_err("%s: unsupported CACHE INDEX operation\n",
2257                                 __func__);
2258                         return EMULATE_FAIL;
2259                 }
2260
2261 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
2262                 kvm_mips_trans_cache_index(inst, opc, vcpu);
2263 #endif
2264                 goto done;
2265         }
2266
2267         /* XXXKYMA: Only a subset of cache ops are supported, used by Linux */
2268         if (op_inst == Hit_Writeback_Inv_D || op_inst == Hit_Invalidate_D) {
2269                 /*
2270                  * Perform the dcache part of icache synchronisation on the
2271                  * guest's behalf.
2272                  */
2273                 er = kvm_mips_guest_cache_op(protected_writeback_dcache_line,
2274                                              curr_pc, va, vcpu, cause);
2275                 if (er != EMULATE_DONE)
2276                         goto done;
2277 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
2278                 /*
2279                  * Replace the CACHE instruction, with a SYNCI, not the same,
2280                  * but avoids a trap
2281                  */
2282                 kvm_mips_trans_cache_va(inst, opc, vcpu);
2283 #endif
2284         } else if (op_inst == Hit_Invalidate_I) {
2285                 /* Perform the icache synchronisation on the guest's behalf */
2286                 er = kvm_mips_guest_cache_op(protected_writeback_dcache_line,
2287                                              curr_pc, va, vcpu, cause);
2288                 if (er != EMULATE_DONE)
2289                         goto done;
2290                 er = kvm_mips_guest_cache_op(protected_flush_icache_line,
2291                                              curr_pc, va, vcpu, cause);
2292                 if (er != EMULATE_DONE)
2293                         goto done;
2294
2295 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
2296                 /* Replace the CACHE instruction, with a SYNCI */
2297                 kvm_mips_trans_cache_va(inst, opc, vcpu);
2298 #endif
2299         } else {
2300                 kvm_err("NO-OP CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
2301                         cache, op, base, arch->gprs[base], offset);
2302                 er = EMULATE_FAIL;
2303         }
2304
2305 done:
2306         /* Rollback PC only if emulation was unsuccessful */
2307         if (er == EMULATE_FAIL)
2308                 vcpu->arch.pc = curr_pc;
2309         /* Guest exception needs guest to resume */
2310         if (er == EMULATE_EXCEPT)
2311                 er = EMULATE_DONE;
2312
2313         return er;
2314 }
2315
2316 enum emulation_result kvm_mips_emulate_inst(u32 cause, u32 *opc,
2317                                             struct kvm_vcpu *vcpu)
2318 {
2319         union mips_instruction inst;
2320         enum emulation_result er = EMULATE_DONE;
2321         int err;
2322
2323         /* Fetch the instruction. */
2324         if (cause & CAUSEF_BD)
2325                 opc += 1;
2326         err = kvm_get_badinstr(opc, vcpu, &inst.word);
2327         if (err)
2328                 return EMULATE_FAIL;
2329
2330         switch (inst.r_format.opcode) {
2331         case cop0_op:
2332                 er = kvm_mips_emulate_CP0(inst, opc, cause, vcpu);
2333                 break;
2334
2335 #ifndef CONFIG_CPU_MIPSR6
2336         case cache_op:
2337                 ++vcpu->stat.cache_exits;
2338                 trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
2339                 er = kvm_mips_emulate_cache(inst, opc, cause, vcpu);
2340                 break;
2341 #else
2342         case spec3_op:
2343                 switch (inst.spec3_format.func) {
2344                 case cache6_op:
2345                         ++vcpu->stat.cache_exits;
2346                         trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
2347                         er = kvm_mips_emulate_cache(inst, opc, cause,
2348                                                     vcpu);
2349                         break;
2350                 default:
2351                         goto unknown;
2352                 }
2353                 break;
2354 unknown:
2355 #endif
2356
2357         default:
2358                 kvm_err("Instruction emulation not supported (%p/%#x)\n", opc,
2359                         inst.word);
2360                 kvm_arch_vcpu_dump_regs(vcpu);
2361                 er = EMULATE_FAIL;
2362                 break;
2363         }
2364
2365         return er;
2366 }
2367 #endif /* CONFIG_KVM_MIPS_VZ */
2368
2369 /**
2370  * kvm_mips_guest_exception_base() - Find guest exception vector base address.
2371  *
2372  * Returns:     The base address of the current guest exception vector, taking
2373  *              both Guest.CP0_Status.BEV and Guest.CP0_EBase into account.
2374  */
2375 long kvm_mips_guest_exception_base(struct kvm_vcpu *vcpu)
2376 {
2377         struct mips_coproc *cop0 = vcpu->arch.cop0;
2378
2379         if (kvm_read_c0_guest_status(cop0) & ST0_BEV)
2380                 return KVM_GUEST_CKSEG1ADDR(0x1fc00200);
2381         else
2382                 return kvm_read_c0_guest_ebase(cop0) & MIPS_EBASE_BASE;
2383 }
2384
2385 enum emulation_result kvm_mips_emulate_syscall(u32 cause,
2386                                                u32 *opc,
2387                                                struct kvm_vcpu *vcpu)
2388 {
2389         struct mips_coproc *cop0 = vcpu->arch.cop0;
2390         struct kvm_vcpu_arch *arch = &vcpu->arch;
2391         enum emulation_result er = EMULATE_DONE;
2392
2393         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2394                 /* save old pc */
2395                 kvm_write_c0_guest_epc(cop0, arch->pc);
2396                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2397
2398                 if (cause & CAUSEF_BD)
2399                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2400                 else
2401                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2402
2403                 kvm_debug("Delivering SYSCALL @ pc %#lx\n", arch->pc);
2404
2405                 kvm_change_c0_guest_cause(cop0, (0xff),
2406                                           (EXCCODE_SYS << CAUSEB_EXCCODE));
2407
2408                 /* Set PC to the exception entry point */
2409                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2410
2411         } else {
2412                 kvm_err("Trying to deliver SYSCALL when EXL is already set\n");
2413                 er = EMULATE_FAIL;
2414         }
2415
2416         return er;
2417 }
2418
2419 enum emulation_result kvm_mips_emulate_tlbmiss_ld(u32 cause,
2420                                                   u32 *opc,
2421                                                   struct kvm_vcpu *vcpu)
2422 {
2423         struct mips_coproc *cop0 = vcpu->arch.cop0;
2424         struct kvm_vcpu_arch *arch = &vcpu->arch;
2425         unsigned long entryhi = (vcpu->arch.  host_cp0_badvaddr & VPN2_MASK) |
2426                         (kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2427
2428         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2429                 /* save old pc */
2430                 kvm_write_c0_guest_epc(cop0, arch->pc);
2431                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2432
2433                 if (cause & CAUSEF_BD)
2434                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2435                 else
2436                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2437
2438                 kvm_debug("[EXL == 0] delivering TLB MISS @ pc %#lx\n",
2439                           arch->pc);
2440
2441                 /* set pc to the exception entry point */
2442                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x0;
2443
2444         } else {
2445                 kvm_debug("[EXL == 1] delivering TLB MISS @ pc %#lx\n",
2446                           arch->pc);
2447
2448                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2449         }
2450
2451         kvm_change_c0_guest_cause(cop0, (0xff),
2452                                   (EXCCODE_TLBL << CAUSEB_EXCCODE));
2453
2454         /* setup badvaddr, context and entryhi registers for the guest */
2455         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2456         /* XXXKYMA: is the context register used by linux??? */
2457         kvm_write_c0_guest_entryhi(cop0, entryhi);
2458
2459         return EMULATE_DONE;
2460 }
2461
2462 enum emulation_result kvm_mips_emulate_tlbinv_ld(u32 cause,
2463                                                  u32 *opc,
2464                                                  struct kvm_vcpu *vcpu)
2465 {
2466         struct mips_coproc *cop0 = vcpu->arch.cop0;
2467         struct kvm_vcpu_arch *arch = &vcpu->arch;
2468         unsigned long entryhi =
2469                 (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2470                 (kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2471
2472         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2473                 /* save old pc */
2474                 kvm_write_c0_guest_epc(cop0, arch->pc);
2475                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2476
2477                 if (cause & CAUSEF_BD)
2478                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2479                 else
2480                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2481
2482                 kvm_debug("[EXL == 0] delivering TLB INV @ pc %#lx\n",
2483                           arch->pc);
2484         } else {
2485                 kvm_debug("[EXL == 1] delivering TLB MISS @ pc %#lx\n",
2486                           arch->pc);
2487         }
2488
2489         /* set pc to the exception entry point */
2490         arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2491
2492         kvm_change_c0_guest_cause(cop0, (0xff),
2493                                   (EXCCODE_TLBL << CAUSEB_EXCCODE));
2494
2495         /* setup badvaddr, context and entryhi registers for the guest */
2496         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2497         /* XXXKYMA: is the context register used by linux??? */
2498         kvm_write_c0_guest_entryhi(cop0, entryhi);
2499
2500         return EMULATE_DONE;
2501 }
2502
2503 enum emulation_result kvm_mips_emulate_tlbmiss_st(u32 cause,
2504                                                   u32 *opc,
2505                                                   struct kvm_vcpu *vcpu)
2506 {
2507         struct mips_coproc *cop0 = vcpu->arch.cop0;
2508         struct kvm_vcpu_arch *arch = &vcpu->arch;
2509         unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2510                         (kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2511
2512         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2513                 /* save old pc */
2514                 kvm_write_c0_guest_epc(cop0, arch->pc);
2515                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2516
2517                 if (cause & CAUSEF_BD)
2518                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2519                 else
2520                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2521
2522                 kvm_debug("[EXL == 0] Delivering TLB MISS @ pc %#lx\n",
2523                           arch->pc);
2524
2525                 /* Set PC to the exception entry point */
2526                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x0;
2527         } else {
2528                 kvm_debug("[EXL == 1] Delivering TLB MISS @ pc %#lx\n",
2529                           arch->pc);
2530                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2531         }
2532
2533         kvm_change_c0_guest_cause(cop0, (0xff),
2534                                   (EXCCODE_TLBS << CAUSEB_EXCCODE));
2535
2536         /* setup badvaddr, context and entryhi registers for the guest */
2537         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2538         /* XXXKYMA: is the context register used by linux??? */
2539         kvm_write_c0_guest_entryhi(cop0, entryhi);
2540
2541         return EMULATE_DONE;
2542 }
2543
2544 enum emulation_result kvm_mips_emulate_tlbinv_st(u32 cause,
2545                                                  u32 *opc,
2546                                                  struct kvm_vcpu *vcpu)
2547 {
2548         struct mips_coproc *cop0 = vcpu->arch.cop0;
2549         struct kvm_vcpu_arch *arch = &vcpu->arch;
2550         unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2551                 (kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2552
2553         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2554                 /* save old pc */
2555                 kvm_write_c0_guest_epc(cop0, arch->pc);
2556                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2557
2558                 if (cause & CAUSEF_BD)
2559                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2560                 else
2561                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2562
2563                 kvm_debug("[EXL == 0] Delivering TLB MISS @ pc %#lx\n",
2564                           arch->pc);
2565         } else {
2566                 kvm_debug("[EXL == 1] Delivering TLB MISS @ pc %#lx\n",
2567                           arch->pc);
2568         }
2569
2570         /* Set PC to the exception entry point */
2571         arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2572
2573         kvm_change_c0_guest_cause(cop0, (0xff),
2574                                   (EXCCODE_TLBS << CAUSEB_EXCCODE));
2575
2576         /* setup badvaddr, context and entryhi registers for the guest */
2577         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2578         /* XXXKYMA: is the context register used by linux??? */
2579         kvm_write_c0_guest_entryhi(cop0, entryhi);
2580
2581         return EMULATE_DONE;
2582 }
2583
2584 enum emulation_result kvm_mips_emulate_tlbmod(u32 cause,
2585                                               u32 *opc,
2586                                               struct kvm_vcpu *vcpu)
2587 {
2588         struct mips_coproc *cop0 = vcpu->arch.cop0;
2589         unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2590                         (kvm_read_c0_guest_entryhi(cop0) & KVM_ENTRYHI_ASID);
2591         struct kvm_vcpu_arch *arch = &vcpu->arch;
2592
2593         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2594                 /* save old pc */
2595                 kvm_write_c0_guest_epc(cop0, arch->pc);
2596                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2597
2598                 if (cause & CAUSEF_BD)
2599                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2600                 else
2601                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2602
2603                 kvm_debug("[EXL == 0] Delivering TLB MOD @ pc %#lx\n",
2604                           arch->pc);
2605         } else {
2606                 kvm_debug("[EXL == 1] Delivering TLB MOD @ pc %#lx\n",
2607                           arch->pc);
2608         }
2609
2610         arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2611
2612         kvm_change_c0_guest_cause(cop0, (0xff),
2613                                   (EXCCODE_MOD << CAUSEB_EXCCODE));
2614
2615         /* setup badvaddr, context and entryhi registers for the guest */
2616         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2617         /* XXXKYMA: is the context register used by linux??? */
2618         kvm_write_c0_guest_entryhi(cop0, entryhi);
2619
2620         return EMULATE_DONE;
2621 }
2622
2623 enum emulation_result kvm_mips_emulate_fpu_exc(u32 cause,
2624                                                u32 *opc,
2625                                                struct kvm_vcpu *vcpu)
2626 {
2627         struct mips_coproc *cop0 = vcpu->arch.cop0;
2628         struct kvm_vcpu_arch *arch = &vcpu->arch;
2629
2630         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2631                 /* save old pc */
2632                 kvm_write_c0_guest_epc(cop0, arch->pc);
2633                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2634
2635                 if (cause & CAUSEF_BD)
2636                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2637                 else
2638                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2639
2640         }
2641
2642         arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2643
2644         kvm_change_c0_guest_cause(cop0, (0xff),
2645                                   (EXCCODE_CPU << CAUSEB_EXCCODE));
2646         kvm_change_c0_guest_cause(cop0, (CAUSEF_CE), (0x1 << CAUSEB_CE));
2647
2648         return EMULATE_DONE;
2649 }
2650
2651 enum emulation_result kvm_mips_emulate_ri_exc(u32 cause,
2652                                               u32 *opc,
2653                                               struct kvm_vcpu *vcpu)
2654 {
2655         struct mips_coproc *cop0 = vcpu->arch.cop0;
2656         struct kvm_vcpu_arch *arch = &vcpu->arch;
2657         enum emulation_result er = EMULATE_DONE;
2658
2659         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2660                 /* save old pc */
2661                 kvm_write_c0_guest_epc(cop0, arch->pc);
2662                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2663
2664                 if (cause & CAUSEF_BD)
2665                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2666                 else
2667                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2668
2669                 kvm_debug("Delivering RI @ pc %#lx\n", arch->pc);
2670
2671                 kvm_change_c0_guest_cause(cop0, (0xff),
2672                                           (EXCCODE_RI << CAUSEB_EXCCODE));
2673
2674                 /* Set PC to the exception entry point */
2675                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2676
2677         } else {
2678                 kvm_err("Trying to deliver RI when EXL is already set\n");
2679                 er = EMULATE_FAIL;
2680         }
2681
2682         return er;
2683 }
2684
2685 enum emulation_result kvm_mips_emulate_bp_exc(u32 cause,
2686                                               u32 *opc,
2687                                               struct kvm_vcpu *vcpu)
2688 {
2689         struct mips_coproc *cop0 = vcpu->arch.cop0;
2690         struct kvm_vcpu_arch *arch = &vcpu->arch;
2691         enum emulation_result er = EMULATE_DONE;
2692
2693         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2694                 /* save old pc */
2695                 kvm_write_c0_guest_epc(cop0, arch->pc);
2696                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2697
2698                 if (cause & CAUSEF_BD)
2699                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2700                 else
2701                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2702
2703                 kvm_debug("Delivering BP @ pc %#lx\n", arch->pc);
2704
2705                 kvm_change_c0_guest_cause(cop0, (0xff),
2706                                           (EXCCODE_BP << CAUSEB_EXCCODE));
2707
2708                 /* Set PC to the exception entry point */
2709                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2710
2711         } else {
2712                 kvm_err("Trying to deliver BP when EXL is already set\n");
2713                 er = EMULATE_FAIL;
2714         }
2715
2716         return er;
2717 }
2718
2719 enum emulation_result kvm_mips_emulate_trap_exc(u32 cause,
2720                                                 u32 *opc,
2721                                                 struct kvm_vcpu *vcpu)
2722 {
2723         struct mips_coproc *cop0 = vcpu->arch.cop0;
2724         struct kvm_vcpu_arch *arch = &vcpu->arch;
2725         enum emulation_result er = EMULATE_DONE;
2726
2727         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2728                 /* save old pc */
2729                 kvm_write_c0_guest_epc(cop0, arch->pc);
2730                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2731
2732                 if (cause & CAUSEF_BD)
2733                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2734                 else
2735                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2736
2737                 kvm_debug("Delivering TRAP @ pc %#lx\n", arch->pc);
2738
2739                 kvm_change_c0_guest_cause(cop0, (0xff),
2740                                           (EXCCODE_TR << CAUSEB_EXCCODE));
2741
2742                 /* Set PC to the exception entry point */
2743                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2744
2745         } else {
2746                 kvm_err("Trying to deliver TRAP when EXL is already set\n");
2747                 er = EMULATE_FAIL;
2748         }
2749
2750         return er;
2751 }
2752
2753 enum emulation_result kvm_mips_emulate_msafpe_exc(u32 cause,
2754                                                   u32 *opc,
2755                                                   struct kvm_vcpu *vcpu)
2756 {
2757         struct mips_coproc *cop0 = vcpu->arch.cop0;
2758         struct kvm_vcpu_arch *arch = &vcpu->arch;
2759         enum emulation_result er = EMULATE_DONE;
2760
2761         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2762                 /* save old pc */
2763                 kvm_write_c0_guest_epc(cop0, arch->pc);
2764                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2765
2766                 if (cause & CAUSEF_BD)
2767                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2768                 else
2769                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2770
2771                 kvm_debug("Delivering MSAFPE @ pc %#lx\n", arch->pc);
2772
2773                 kvm_change_c0_guest_cause(cop0, (0xff),
2774                                           (EXCCODE_MSAFPE << CAUSEB_EXCCODE));
2775
2776                 /* Set PC to the exception entry point */
2777                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2778
2779         } else {
2780                 kvm_err("Trying to deliver MSAFPE when EXL is already set\n");
2781                 er = EMULATE_FAIL;
2782         }
2783
2784         return er;
2785 }
2786
2787 enum emulation_result kvm_mips_emulate_fpe_exc(u32 cause,
2788                                                u32 *opc,
2789                                                struct kvm_vcpu *vcpu)
2790 {
2791         struct mips_coproc *cop0 = vcpu->arch.cop0;
2792         struct kvm_vcpu_arch *arch = &vcpu->arch;
2793         enum emulation_result er = EMULATE_DONE;
2794
2795         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2796                 /* save old pc */
2797                 kvm_write_c0_guest_epc(cop0, arch->pc);
2798                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2799
2800                 if (cause & CAUSEF_BD)
2801                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2802                 else
2803                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2804
2805                 kvm_debug("Delivering FPE @ pc %#lx\n", arch->pc);
2806
2807                 kvm_change_c0_guest_cause(cop0, (0xff),
2808                                           (EXCCODE_FPE << CAUSEB_EXCCODE));
2809
2810                 /* Set PC to the exception entry point */
2811                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2812
2813         } else {
2814                 kvm_err("Trying to deliver FPE when EXL is already set\n");
2815                 er = EMULATE_FAIL;
2816         }
2817
2818         return er;
2819 }
2820
2821 enum emulation_result kvm_mips_emulate_msadis_exc(u32 cause,
2822                                                   u32 *opc,
2823                                                   struct kvm_vcpu *vcpu)
2824 {
2825         struct mips_coproc *cop0 = vcpu->arch.cop0;
2826         struct kvm_vcpu_arch *arch = &vcpu->arch;
2827         enum emulation_result er = EMULATE_DONE;
2828
2829         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2830                 /* save old pc */
2831                 kvm_write_c0_guest_epc(cop0, arch->pc);
2832                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2833
2834                 if (cause & CAUSEF_BD)
2835                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2836                 else
2837                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2838
2839                 kvm_debug("Delivering MSADIS @ pc %#lx\n", arch->pc);
2840
2841                 kvm_change_c0_guest_cause(cop0, (0xff),
2842                                           (EXCCODE_MSADIS << CAUSEB_EXCCODE));
2843
2844                 /* Set PC to the exception entry point */
2845                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
2846
2847         } else {
2848                 kvm_err("Trying to deliver MSADIS when EXL is already set\n");
2849                 er = EMULATE_FAIL;
2850         }
2851
2852         return er;
2853 }
2854
2855 enum emulation_result kvm_mips_handle_ri(u32 cause, u32 *opc,
2856                                          struct kvm_vcpu *vcpu)
2857 {
2858         struct mips_coproc *cop0 = vcpu->arch.cop0;
2859         struct kvm_vcpu_arch *arch = &vcpu->arch;
2860         enum emulation_result er = EMULATE_DONE;
2861         unsigned long curr_pc;
2862         union mips_instruction inst;
2863         int err;
2864
2865         /*
2866          * Update PC and hold onto current PC in case there is
2867          * an error and we want to rollback the PC
2868          */
2869         curr_pc = vcpu->arch.pc;
2870         er = update_pc(vcpu, cause);
2871         if (er == EMULATE_FAIL)
2872                 return er;
2873
2874         /* Fetch the instruction. */
2875         if (cause & CAUSEF_BD)
2876                 opc += 1;
2877         err = kvm_get_badinstr(opc, vcpu, &inst.word);
2878         if (err) {
2879                 kvm_err("%s: Cannot get inst @ %p (%d)\n", __func__, opc, err);
2880                 return EMULATE_FAIL;
2881         }
2882
2883         if (inst.r_format.opcode == spec3_op &&
2884             inst.r_format.func == rdhwr_op &&
2885             inst.r_format.rs == 0 &&
2886             (inst.r_format.re >> 3) == 0) {
2887                 int usermode = !KVM_GUEST_KERNEL_MODE(vcpu);
2888                 int rd = inst.r_format.rd;
2889                 int rt = inst.r_format.rt;
2890                 int sel = inst.r_format.re & 0x7;
2891
2892                 /* If usermode, check RDHWR rd is allowed by guest HWREna */
2893                 if (usermode && !(kvm_read_c0_guest_hwrena(cop0) & BIT(rd))) {
2894                         kvm_debug("RDHWR %#x disallowed by HWREna @ %p\n",
2895                                   rd, opc);
2896                         goto emulate_ri;
2897                 }
2898                 switch (rd) {
2899                 case MIPS_HWR_CPUNUM:           /* CPU number */
2900                         arch->gprs[rt] = vcpu->vcpu_id;
2901                         break;
2902                 case MIPS_HWR_SYNCISTEP:        /* SYNCI length */
2903                         arch->gprs[rt] = min(current_cpu_data.dcache.linesz,
2904                                              current_cpu_data.icache.linesz);
2905                         break;
2906                 case MIPS_HWR_CC:               /* Read count register */
2907                         arch->gprs[rt] = (s32)kvm_mips_read_count(vcpu);
2908                         break;
2909                 case MIPS_HWR_CCRES:            /* Count register resolution */
2910                         switch (current_cpu_data.cputype) {
2911                         case CPU_20KC:
2912                         case CPU_25KF:
2913                                 arch->gprs[rt] = 1;
2914                                 break;
2915                         default:
2916                                 arch->gprs[rt] = 2;
2917                         }
2918                         break;
2919                 case MIPS_HWR_ULR:              /* Read UserLocal register */
2920                         arch->gprs[rt] = kvm_read_c0_guest_userlocal(cop0);
2921                         break;
2922
2923                 default:
2924                         kvm_debug("RDHWR %#x not supported @ %p\n", rd, opc);
2925                         goto emulate_ri;
2926                 }
2927
2928                 trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR, KVM_TRACE_HWR(rd, sel),
2929                               vcpu->arch.gprs[rt]);
2930         } else {
2931                 kvm_debug("Emulate RI not supported @ %p: %#x\n",
2932                           opc, inst.word);
2933                 goto emulate_ri;
2934         }
2935
2936         return EMULATE_DONE;
2937
2938 emulate_ri:
2939         /*
2940          * Rollback PC (if in branch delay slot then the PC already points to
2941          * branch target), and pass the RI exception to the guest OS.
2942          */
2943         vcpu->arch.pc = curr_pc;
2944         return kvm_mips_emulate_ri_exc(cause, opc, vcpu);
2945 }
2946
2947 enum emulation_result kvm_mips_complete_mmio_load(struct kvm_vcpu *vcpu)
2948 {
2949         struct kvm_run *run = vcpu->run;
2950         unsigned long *gpr = &vcpu->arch.gprs[vcpu->arch.io_gpr];
2951         enum emulation_result er = EMULATE_DONE;
2952
2953         if (run->mmio.len > sizeof(*gpr)) {
2954                 kvm_err("Bad MMIO length: %d", run->mmio.len);
2955                 er = EMULATE_FAIL;
2956                 goto done;
2957         }
2958
2959         /* Restore saved resume PC */
2960         vcpu->arch.pc = vcpu->arch.io_pc;
2961
2962         switch (run->mmio.len) {
2963         case 8:
2964                 switch (vcpu->mmio_needed) {
2965                 case 11:
2966                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffffffff) |
2967                                 (((*(s64 *)run->mmio.data) & 0xff) << 56);
2968                         break;
2969                 case 12:
2970                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffffff) |
2971                                 (((*(s64 *)run->mmio.data) & 0xffff) << 48);
2972                         break;
2973                 case 13:
2974                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffff) |
2975                                 (((*(s64 *)run->mmio.data) & 0xffffff) << 40);
2976                         break;
2977                 case 14:
2978                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffff) |
2979                                 (((*(s64 *)run->mmio.data) & 0xffffffff) << 32);
2980                         break;
2981                 case 15:
2982                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffff) |
2983                                 (((*(s64 *)run->mmio.data) & 0xffffffffff) << 24);
2984                         break;
2985                 case 16:
2986                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffff) |
2987                                 (((*(s64 *)run->mmio.data) & 0xffffffffffff) << 16);
2988                         break;
2989                 case 17:
2990                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xff) |
2991                                 (((*(s64 *)run->mmio.data) & 0xffffffffffffff) << 8);
2992                         break;
2993                 case 18:
2994                 case 19:
2995                         *gpr = *(s64 *)run->mmio.data;
2996                         break;
2997                 case 20:
2998                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xff00000000000000) |
2999                                 ((((*(s64 *)run->mmio.data)) >> 8) & 0xffffffffffffff);
3000                         break;
3001                 case 21:
3002                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffff000000000000) |
3003                                 ((((*(s64 *)run->mmio.data)) >> 16) & 0xffffffffffff);
3004                         break;
3005                 case 22:
3006                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffff0000000000) |
3007                                 ((((*(s64 *)run->mmio.data)) >> 24) & 0xffffffffff);
3008                         break;
3009                 case 23:
3010                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffff00000000) |
3011                                 ((((*(s64 *)run->mmio.data)) >> 32) & 0xffffffff);
3012                         break;
3013                 case 24:
3014                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffff000000) |
3015                                 ((((*(s64 *)run->mmio.data)) >> 40) & 0xffffff);
3016                         break;
3017                 case 25:
3018                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffffff0000) |
3019                                 ((((*(s64 *)run->mmio.data)) >> 48) & 0xffff);
3020                         break;
3021                 case 26:
3022                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffffffffffff00) |
3023                                 ((((*(s64 *)run->mmio.data)) >> 56) & 0xff);
3024                         break;
3025                 default:
3026                         *gpr = *(s64 *)run->mmio.data;
3027                 }
3028                 break;
3029
3030         case 4:
3031                 switch (vcpu->mmio_needed) {
3032                 case 1:
3033                         *gpr = *(u32 *)run->mmio.data;
3034                         break;
3035                 case 2:
3036                         *gpr = *(s32 *)run->mmio.data;
3037                         break;
3038                 case 3:
3039                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffff) |
3040                                 (((*(s32 *)run->mmio.data) & 0xff) << 24);
3041                         break;
3042                 case 4:
3043                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffff) |
3044                                 (((*(s32 *)run->mmio.data) & 0xffff) << 16);
3045                         break;
3046                 case 5:
3047                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xff) |
3048                                 (((*(s32 *)run->mmio.data) & 0xffffff) << 8);
3049                         break;
3050                 case 6:
3051                 case 7:
3052                         *gpr = *(s32 *)run->mmio.data;
3053                         break;
3054                 case 8:
3055                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xff000000) |
3056                                 ((((*(s32 *)run->mmio.data)) >> 8) & 0xffffff);
3057                         break;
3058                 case 9:
3059                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffff0000) |
3060                                 ((((*(s32 *)run->mmio.data)) >> 16) & 0xffff);
3061                         break;
3062                 case 10:
3063                         *gpr = (vcpu->arch.gprs[vcpu->arch.io_gpr] & 0xffffff00) |
3064                                 ((((*(s32 *)run->mmio.data)) >> 24) & 0xff);
3065                         break;
3066                 default:
3067                         *gpr = *(s32 *)run->mmio.data;
3068                 }
3069                 break;
3070
3071         case 2:
3072                 if (vcpu->mmio_needed == 1)
3073                         *gpr = *(u16 *)run->mmio.data;
3074                 else
3075                         *gpr = *(s16 *)run->mmio.data;
3076
3077                 break;
3078         case 1:
3079                 if (vcpu->mmio_needed == 1)
3080                         *gpr = *(u8 *)run->mmio.data;
3081                 else
3082                         *gpr = *(s8 *)run->mmio.data;
3083                 break;
3084         }
3085
3086 done:
3087         return er;
3088 }
3089
3090 static enum emulation_result kvm_mips_emulate_exc(u32 cause,
3091                                                   u32 *opc,
3092                                                   struct kvm_vcpu *vcpu)
3093 {
3094         u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
3095         struct mips_coproc *cop0 = vcpu->arch.cop0;
3096         struct kvm_vcpu_arch *arch = &vcpu->arch;
3097         enum emulation_result er = EMULATE_DONE;
3098
3099         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
3100                 /* save old pc */
3101                 kvm_write_c0_guest_epc(cop0, arch->pc);
3102                 kvm_set_c0_guest_status(cop0, ST0_EXL);
3103
3104                 if (cause & CAUSEF_BD)
3105                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
3106                 else
3107                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
3108
3109                 kvm_change_c0_guest_cause(cop0, (0xff),
3110                                           (exccode << CAUSEB_EXCCODE));
3111
3112                 /* Set PC to the exception entry point */
3113                 arch->pc = kvm_mips_guest_exception_base(vcpu) + 0x180;
3114                 kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
3115
3116                 kvm_debug("Delivering EXC %d @ pc %#lx, badVaddr: %#lx\n",
3117                           exccode, kvm_read_c0_guest_epc(cop0),
3118                           kvm_read_c0_guest_badvaddr(cop0));
3119         } else {
3120                 kvm_err("Trying to deliver EXC when EXL is already set\n");
3121                 er = EMULATE_FAIL;
3122         }
3123
3124         return er;
3125 }
3126
3127 enum emulation_result kvm_mips_check_privilege(u32 cause,
3128                                                u32 *opc,
3129                                                struct kvm_vcpu *vcpu)
3130 {
3131         enum emulation_result er = EMULATE_DONE;
3132         u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
3133         unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
3134
3135         int usermode = !KVM_GUEST_KERNEL_MODE(vcpu);
3136
3137         if (usermode) {
3138                 switch (exccode) {
3139                 case EXCCODE_INT:
3140                 case EXCCODE_SYS:
3141                 case EXCCODE_BP:
3142                 case EXCCODE_RI:
3143                 case EXCCODE_TR:
3144                 case EXCCODE_MSAFPE:
3145                 case EXCCODE_FPE:
3146                 case EXCCODE_MSADIS:
3147                         break;
3148
3149                 case EXCCODE_CPU:
3150                         if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 0)
3151                                 er = EMULATE_PRIV_FAIL;
3152                         break;
3153
3154                 case EXCCODE_MOD:
3155                         break;
3156
3157                 case EXCCODE_TLBL:
3158                         /*
3159                          * We we are accessing Guest kernel space, then send an
3160                          * address error exception to the guest
3161                          */
3162                         if (badvaddr >= (unsigned long) KVM_GUEST_KSEG0) {
3163                                 kvm_debug("%s: LD MISS @ %#lx\n", __func__,
3164                                           badvaddr);
3165                                 cause &= ~0xff;
3166                                 cause |= (EXCCODE_ADEL << CAUSEB_EXCCODE);
3167                                 er = EMULATE_PRIV_FAIL;
3168                         }
3169                         break;
3170
3171                 case EXCCODE_TLBS:
3172                         /*
3173                          * We we are accessing Guest kernel space, then send an
3174                          * address error exception to the guest
3175                          */
3176                         if (badvaddr >= (unsigned long) KVM_GUEST_KSEG0) {
3177                                 kvm_debug("%s: ST MISS @ %#lx\n", __func__,
3178                                           badvaddr);
3179                                 cause &= ~0xff;
3180                                 cause |= (EXCCODE_ADES << CAUSEB_EXCCODE);
3181                                 er = EMULATE_PRIV_FAIL;
3182                         }
3183                         break;
3184
3185                 case EXCCODE_ADES:
3186                         kvm_debug("%s: address error ST @ %#lx\n", __func__,
3187                                   badvaddr);
3188                         if ((badvaddr & PAGE_MASK) == KVM_GUEST_COMMPAGE_ADDR) {
3189                                 cause &= ~0xff;
3190                                 cause |= (EXCCODE_TLBS << CAUSEB_EXCCODE);
3191                         }
3192                         er = EMULATE_PRIV_FAIL;
3193                         break;
3194                 case EXCCODE_ADEL:
3195                         kvm_debug("%s: address error LD @ %#lx\n", __func__,
3196                                   badvaddr);
3197                         if ((badvaddr & PAGE_MASK) == KVM_GUEST_COMMPAGE_ADDR) {
3198                                 cause &= ~0xff;
3199                                 cause |= (EXCCODE_TLBL << CAUSEB_EXCCODE);
3200                         }
3201                         er = EMULATE_PRIV_FAIL;
3202                         break;
3203                 default:
3204                         er = EMULATE_PRIV_FAIL;
3205                         break;
3206                 }
3207         }
3208
3209         if (er == EMULATE_PRIV_FAIL)
3210                 kvm_mips_emulate_exc(cause, opc, vcpu);
3211
3212         return er;
3213 }
3214
3215 /*
3216  * User Address (UA) fault, this could happen if
3217  * (1) TLB entry not present/valid in both Guest and shadow host TLBs, in this
3218  *     case we pass on the fault to the guest kernel and let it handle it.
3219  * (2) TLB entry is present in the Guest TLB but not in the shadow, in this
3220  *     case we inject the TLB from the Guest TLB into the shadow host TLB
3221  */
3222 enum emulation_result kvm_mips_handle_tlbmiss(u32 cause,
3223                                               u32 *opc,
3224                                               struct kvm_vcpu *vcpu,
3225                                               bool write_fault)
3226 {
3227         enum emulation_result er = EMULATE_DONE;
3228         u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
3229         unsigned long va = vcpu->arch.host_cp0_badvaddr;
3230         int index;
3231
3232         kvm_debug("kvm_mips_handle_tlbmiss: badvaddr: %#lx\n",
3233                   vcpu->arch.host_cp0_badvaddr);
3234
3235         /*
3236          * KVM would not have got the exception if this entry was valid in the
3237          * shadow host TLB. Check the Guest TLB, if the entry is not there then
3238          * send the guest an exception. The guest exc handler should then inject
3239          * an entry into the guest TLB.
3240          */
3241         index = kvm_mips_guest_tlb_lookup(vcpu,
3242                       (va & VPN2_MASK) |
3243                       (kvm_read_c0_guest_entryhi(vcpu->arch.cop0) &
3244                        KVM_ENTRYHI_ASID));
3245         if (index < 0) {
3246                 if (exccode == EXCCODE_TLBL) {
3247                         er = kvm_mips_emulate_tlbmiss_ld(cause, opc, vcpu);
3248                 } else if (exccode == EXCCODE_TLBS) {
3249                         er = kvm_mips_emulate_tlbmiss_st(cause, opc, vcpu);
3250                 } else {
3251                         kvm_err("%s: invalid exc code: %d\n", __func__,
3252                                 exccode);
3253                         er = EMULATE_FAIL;
3254                 }
3255         } else {
3256                 struct kvm_mips_tlb *tlb = &vcpu->arch.guest_tlb[index];
3257
3258                 /*
3259                  * Check if the entry is valid, if not then setup a TLB invalid
3260                  * exception to the guest
3261                  */
3262                 if (!TLB_IS_VALID(*tlb, va)) {
3263                         if (exccode == EXCCODE_TLBL) {
3264                                 er = kvm_mips_emulate_tlbinv_ld(cause, opc,
3265                                                                 vcpu);
3266                         } else if (exccode == EXCCODE_TLBS) {
3267                                 er = kvm_mips_emulate_tlbinv_st(cause, opc,
3268                                                                 vcpu);
3269                         } else {
3270                                 kvm_err("%s: invalid exc code: %d\n", __func__,
3271                                         exccode);
3272                                 er = EMULATE_FAIL;
3273                         }
3274                 } else {
3275                         kvm_debug("Injecting hi: %#lx, lo0: %#lx, lo1: %#lx into shadow host TLB\n",
3276                                   tlb->tlb_hi, tlb->tlb_lo[0], tlb->tlb_lo[1]);
3277                         /*
3278                          * OK we have a Guest TLB entry, now inject it into the
3279                          * shadow host TLB
3280                          */
3281                         if (kvm_mips_handle_mapped_seg_tlb_fault(vcpu, tlb, va,
3282                                                                  write_fault)) {
3283                                 kvm_err("%s: handling mapped seg tlb fault for %lx, index: %u, vcpu: %p, ASID: %#lx\n",
3284                                         __func__, va, index, vcpu,
3285                                         read_c0_entryhi());
3286                                 er = EMULATE_FAIL;
3287                         }
3288                 }
3289         }
3290
3291         return er;
3292 }