]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - arch/mips/kvm/emulate.c
MIPS: KVM: Fix timer IRQ race when writing CP0_Compare
[karo-tx-linux.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/module.h>
17 #include <linux/vmalloc.h>
18 #include <linux/fs.h>
19 #include <linux/bootmem.h>
20 #include <linux/random.h>
21 #include <asm/page.h>
22 #include <asm/cacheflush.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 "opcode.h"
33 #include "interrupt.h"
34 #include "commpage.h"
35
36 #include "trace.h"
37
38 /*
39  * Compute the return address and do emulate branch simulation, if required.
40  * This function should be called only in branch delay slot active.
41  */
42 unsigned long kvm_compute_return_epc(struct kvm_vcpu *vcpu,
43         unsigned long instpc)
44 {
45         unsigned int dspcontrol;
46         union mips_instruction insn;
47         struct kvm_vcpu_arch *arch = &vcpu->arch;
48         long epc = instpc;
49         long nextpc = KVM_INVALID_INST;
50
51         if (epc & 3)
52                 goto unaligned;
53
54         /* Read the instruction */
55         insn.word = kvm_get_inst((uint32_t *) epc, vcpu);
56
57         if (insn.word == KVM_INVALID_INST)
58                 return KVM_INVALID_INST;
59
60         switch (insn.i_format.opcode) {
61                 /* jr and jalr are in r_format format. */
62         case spec_op:
63                 switch (insn.r_format.func) {
64                 case jalr_op:
65                         arch->gprs[insn.r_format.rd] = epc + 8;
66                         /* Fall through */
67                 case jr_op:
68                         nextpc = arch->gprs[insn.r_format.rs];
69                         break;
70                 }
71                 break;
72
73                 /*
74                  * This group contains:
75                  * bltz_op, bgez_op, bltzl_op, bgezl_op,
76                  * bltzal_op, bgezal_op, bltzall_op, bgezall_op.
77                  */
78         case bcond_op:
79                 switch (insn.i_format.rt) {
80                 case bltz_op:
81                 case bltzl_op:
82                         if ((long)arch->gprs[insn.i_format.rs] < 0)
83                                 epc = epc + 4 + (insn.i_format.simmediate << 2);
84                         else
85                                 epc += 8;
86                         nextpc = epc;
87                         break;
88
89                 case bgez_op:
90                 case bgezl_op:
91                         if ((long)arch->gprs[insn.i_format.rs] >= 0)
92                                 epc = epc + 4 + (insn.i_format.simmediate << 2);
93                         else
94                                 epc += 8;
95                         nextpc = epc;
96                         break;
97
98                 case bltzal_op:
99                 case bltzall_op:
100                         arch->gprs[31] = epc + 8;
101                         if ((long)arch->gprs[insn.i_format.rs] < 0)
102                                 epc = epc + 4 + (insn.i_format.simmediate << 2);
103                         else
104                                 epc += 8;
105                         nextpc = epc;
106                         break;
107
108                 case bgezal_op:
109                 case bgezall_op:
110                         arch->gprs[31] = epc + 8;
111                         if ((long)arch->gprs[insn.i_format.rs] >= 0)
112                                 epc = epc + 4 + (insn.i_format.simmediate << 2);
113                         else
114                                 epc += 8;
115                         nextpc = epc;
116                         break;
117                 case bposge32_op:
118                         if (!cpu_has_dsp)
119                                 goto sigill;
120
121                         dspcontrol = rddsp(0x01);
122
123                         if (dspcontrol >= 32)
124                                 epc = epc + 4 + (insn.i_format.simmediate << 2);
125                         else
126                                 epc += 8;
127                         nextpc = epc;
128                         break;
129                 }
130                 break;
131
132                 /* These are unconditional and in j_format. */
133         case jal_op:
134                 arch->gprs[31] = instpc + 8;
135         case j_op:
136                 epc += 4;
137                 epc >>= 28;
138                 epc <<= 28;
139                 epc |= (insn.j_format.target << 2);
140                 nextpc = epc;
141                 break;
142
143                 /* These are conditional and in i_format. */
144         case beq_op:
145         case beql_op:
146                 if (arch->gprs[insn.i_format.rs] ==
147                     arch->gprs[insn.i_format.rt])
148                         epc = epc + 4 + (insn.i_format.simmediate << 2);
149                 else
150                         epc += 8;
151                 nextpc = epc;
152                 break;
153
154         case bne_op:
155         case bnel_op:
156                 if (arch->gprs[insn.i_format.rs] !=
157                     arch->gprs[insn.i_format.rt])
158                         epc = epc + 4 + (insn.i_format.simmediate << 2);
159                 else
160                         epc += 8;
161                 nextpc = epc;
162                 break;
163
164         case blez_op:           /* not really i_format */
165         case blezl_op:
166                 /* rt field assumed to be zero */
167                 if ((long)arch->gprs[insn.i_format.rs] <= 0)
168                         epc = epc + 4 + (insn.i_format.simmediate << 2);
169                 else
170                         epc += 8;
171                 nextpc = epc;
172                 break;
173
174         case bgtz_op:
175         case bgtzl_op:
176                 /* rt field assumed to be zero */
177                 if ((long)arch->gprs[insn.i_format.rs] > 0)
178                         epc = epc + 4 + (insn.i_format.simmediate << 2);
179                 else
180                         epc += 8;
181                 nextpc = epc;
182                 break;
183
184                 /* And now the FPA/cp1 branch instructions. */
185         case cop1_op:
186                 kvm_err("%s: unsupported cop1_op\n", __func__);
187                 break;
188         }
189
190         return nextpc;
191
192 unaligned:
193         kvm_err("%s: unaligned epc\n", __func__);
194         return nextpc;
195
196 sigill:
197         kvm_err("%s: DSP branch but not DSP ASE\n", __func__);
198         return nextpc;
199 }
200
201 enum emulation_result update_pc(struct kvm_vcpu *vcpu, uint32_t cause)
202 {
203         unsigned long branch_pc;
204         enum emulation_result er = EMULATE_DONE;
205
206         if (cause & CAUSEF_BD) {
207                 branch_pc = kvm_compute_return_epc(vcpu, vcpu->arch.pc);
208                 if (branch_pc == KVM_INVALID_INST) {
209                         er = EMULATE_FAIL;
210                 } else {
211                         vcpu->arch.pc = branch_pc;
212                         kvm_debug("BD update_pc(): New PC: %#lx\n",
213                                   vcpu->arch.pc);
214                 }
215         } else
216                 vcpu->arch.pc += 4;
217
218         kvm_debug("update_pc(): New PC: %#lx\n", vcpu->arch.pc);
219
220         return er;
221 }
222
223 /**
224  * kvm_mips_count_disabled() - Find whether the CP0_Count timer is disabled.
225  * @vcpu:       Virtual CPU.
226  *
227  * Returns:     1 if the CP0_Count timer is disabled by either the guest
228  *              CP0_Cause.DC bit or the count_ctl.DC bit.
229  *              0 otherwise (in which case CP0_Count timer is running).
230  */
231 static inline int kvm_mips_count_disabled(struct kvm_vcpu *vcpu)
232 {
233         struct mips_coproc *cop0 = vcpu->arch.cop0;
234
235         return  (vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC) ||
236                 (kvm_read_c0_guest_cause(cop0) & CAUSEF_DC);
237 }
238
239 /**
240  * kvm_mips_ktime_to_count() - Scale ktime_t to a 32-bit count.
241  *
242  * Caches the dynamic nanosecond bias in vcpu->arch.count_dyn_bias.
243  *
244  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
245  */
246 static uint32_t kvm_mips_ktime_to_count(struct kvm_vcpu *vcpu, ktime_t now)
247 {
248         s64 now_ns, periods;
249         u64 delta;
250
251         now_ns = ktime_to_ns(now);
252         delta = now_ns + vcpu->arch.count_dyn_bias;
253
254         if (delta >= vcpu->arch.count_period) {
255                 /* If delta is out of safe range the bias needs adjusting */
256                 periods = div64_s64(now_ns, vcpu->arch.count_period);
257                 vcpu->arch.count_dyn_bias = -periods * vcpu->arch.count_period;
258                 /* Recalculate delta with new bias */
259                 delta = now_ns + vcpu->arch.count_dyn_bias;
260         }
261
262         /*
263          * We've ensured that:
264          *   delta < count_period
265          *
266          * Therefore the intermediate delta*count_hz will never overflow since
267          * at the boundary condition:
268          *   delta = count_period
269          *   delta = NSEC_PER_SEC * 2^32 / count_hz
270          *   delta * count_hz = NSEC_PER_SEC * 2^32
271          */
272         return div_u64(delta * vcpu->arch.count_hz, NSEC_PER_SEC);
273 }
274
275 /**
276  * kvm_mips_count_time() - Get effective current time.
277  * @vcpu:       Virtual CPU.
278  *
279  * Get effective monotonic ktime. This is usually a straightforward ktime_get(),
280  * except when the master disable bit is set in count_ctl, in which case it is
281  * count_resume, i.e. the time that the count was disabled.
282  *
283  * Returns:     Effective monotonic ktime for CP0_Count.
284  */
285 static inline ktime_t kvm_mips_count_time(struct kvm_vcpu *vcpu)
286 {
287         if (unlikely(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC))
288                 return vcpu->arch.count_resume;
289
290         return ktime_get();
291 }
292
293 /**
294  * kvm_mips_read_count_running() - Read the current count value as if running.
295  * @vcpu:       Virtual CPU.
296  * @now:        Kernel time to read CP0_Count at.
297  *
298  * Returns the current guest CP0_Count register at time @now and handles if the
299  * timer interrupt is pending and hasn't been handled yet.
300  *
301  * Returns:     The current value of the guest CP0_Count register.
302  */
303 static uint32_t kvm_mips_read_count_running(struct kvm_vcpu *vcpu, ktime_t now)
304 {
305         struct mips_coproc *cop0 = vcpu->arch.cop0;
306         ktime_t expires, threshold;
307         uint32_t count, compare;
308         int running;
309
310         /* Calculate the biased and scaled guest CP0_Count */
311         count = vcpu->arch.count_bias + kvm_mips_ktime_to_count(vcpu, now);
312         compare = kvm_read_c0_guest_compare(cop0);
313
314         /*
315          * Find whether CP0_Count has reached the closest timer interrupt. If
316          * not, we shouldn't inject it.
317          */
318         if ((int32_t)(count - compare) < 0)
319                 return count;
320
321         /*
322          * The CP0_Count we're going to return has already reached the closest
323          * timer interrupt. Quickly check if it really is a new interrupt by
324          * looking at whether the interval until the hrtimer expiry time is
325          * less than 1/4 of the timer period.
326          */
327         expires = hrtimer_get_expires(&vcpu->arch.comparecount_timer);
328         threshold = ktime_add_ns(now, vcpu->arch.count_period / 4);
329         if (ktime_before(expires, threshold)) {
330                 /*
331                  * Cancel it while we handle it so there's no chance of
332                  * interference with the timeout handler.
333                  */
334                 running = hrtimer_cancel(&vcpu->arch.comparecount_timer);
335
336                 /* Nothing should be waiting on the timeout */
337                 kvm_mips_callbacks->queue_timer_int(vcpu);
338
339                 /*
340                  * Restart the timer if it was running based on the expiry time
341                  * we read, so that we don't push it back 2 periods.
342                  */
343                 if (running) {
344                         expires = ktime_add_ns(expires,
345                                                vcpu->arch.count_period);
346                         hrtimer_start(&vcpu->arch.comparecount_timer, expires,
347                                       HRTIMER_MODE_ABS);
348                 }
349         }
350
351         return count;
352 }
353
354 /**
355  * kvm_mips_read_count() - Read the current count value.
356  * @vcpu:       Virtual CPU.
357  *
358  * Read the current guest CP0_Count value, taking into account whether the timer
359  * is stopped.
360  *
361  * Returns:     The current guest CP0_Count value.
362  */
363 uint32_t kvm_mips_read_count(struct kvm_vcpu *vcpu)
364 {
365         struct mips_coproc *cop0 = vcpu->arch.cop0;
366
367         /* If count disabled just read static copy of count */
368         if (kvm_mips_count_disabled(vcpu))
369                 return kvm_read_c0_guest_count(cop0);
370
371         return kvm_mips_read_count_running(vcpu, ktime_get());
372 }
373
374 /**
375  * kvm_mips_freeze_hrtimer() - Safely stop the hrtimer.
376  * @vcpu:       Virtual CPU.
377  * @count:      Output pointer for CP0_Count value at point of freeze.
378  *
379  * Freeze the hrtimer safely and return both the ktime and the CP0_Count value
380  * at the point it was frozen. It is guaranteed that any pending interrupts at
381  * the point it was frozen are handled, and none after that point.
382  *
383  * This is useful where the time/CP0_Count is needed in the calculation of the
384  * new parameters.
385  *
386  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
387  *
388  * Returns:     The ktime at the point of freeze.
389  */
390 static ktime_t kvm_mips_freeze_hrtimer(struct kvm_vcpu *vcpu,
391                                        uint32_t *count)
392 {
393         ktime_t now;
394
395         /* stop hrtimer before finding time */
396         hrtimer_cancel(&vcpu->arch.comparecount_timer);
397         now = ktime_get();
398
399         /* find count at this point and handle pending hrtimer */
400         *count = kvm_mips_read_count_running(vcpu, now);
401
402         return now;
403 }
404
405 /**
406  * kvm_mips_resume_hrtimer() - Resume hrtimer, updating expiry.
407  * @vcpu:       Virtual CPU.
408  * @now:        ktime at point of resume.
409  * @count:      CP0_Count at point of resume.
410  *
411  * Resumes the timer and updates the timer expiry based on @now and @count.
412  * This can be used in conjunction with kvm_mips_freeze_timer() when timer
413  * parameters need to be changed.
414  *
415  * It is guaranteed that a timer interrupt immediately after resume will be
416  * handled, but not if CP_Compare is exactly at @count. That case is already
417  * handled by kvm_mips_freeze_timer().
418  *
419  * Assumes !kvm_mips_count_disabled(@vcpu) (guest CP0_Count timer is running).
420  */
421 static void kvm_mips_resume_hrtimer(struct kvm_vcpu *vcpu,
422                                     ktime_t now, uint32_t count)
423 {
424         struct mips_coproc *cop0 = vcpu->arch.cop0;
425         uint32_t compare;
426         u64 delta;
427         ktime_t expire;
428
429         /* Calculate timeout (wrap 0 to 2^32) */
430         compare = kvm_read_c0_guest_compare(cop0);
431         delta = (u64)(uint32_t)(compare - count - 1) + 1;
432         delta = div_u64(delta * NSEC_PER_SEC, vcpu->arch.count_hz);
433         expire = ktime_add_ns(now, delta);
434
435         /* Update hrtimer to use new timeout */
436         hrtimer_cancel(&vcpu->arch.comparecount_timer);
437         hrtimer_start(&vcpu->arch.comparecount_timer, expire, HRTIMER_MODE_ABS);
438 }
439
440 /**
441  * kvm_mips_write_count() - Modify the count and update timer.
442  * @vcpu:       Virtual CPU.
443  * @count:      Guest CP0_Count value to set.
444  *
445  * Sets the CP0_Count value and updates the timer accordingly.
446  */
447 void kvm_mips_write_count(struct kvm_vcpu *vcpu, uint32_t count)
448 {
449         struct mips_coproc *cop0 = vcpu->arch.cop0;
450         ktime_t now;
451
452         /* Calculate bias */
453         now = kvm_mips_count_time(vcpu);
454         vcpu->arch.count_bias = count - kvm_mips_ktime_to_count(vcpu, now);
455
456         if (kvm_mips_count_disabled(vcpu))
457                 /* The timer's disabled, adjust the static count */
458                 kvm_write_c0_guest_count(cop0, count);
459         else
460                 /* Update timeout */
461                 kvm_mips_resume_hrtimer(vcpu, now, count);
462 }
463
464 /**
465  * kvm_mips_init_count() - Initialise timer.
466  * @vcpu:       Virtual CPU.
467  *
468  * Initialise the timer to a sensible frequency, namely 100MHz, zero it, and set
469  * it going if it's enabled.
470  */
471 void kvm_mips_init_count(struct kvm_vcpu *vcpu)
472 {
473         /* 100 MHz */
474         vcpu->arch.count_hz = 100*1000*1000;
475         vcpu->arch.count_period = div_u64((u64)NSEC_PER_SEC << 32,
476                                           vcpu->arch.count_hz);
477         vcpu->arch.count_dyn_bias = 0;
478
479         /* Starting at 0 */
480         kvm_mips_write_count(vcpu, 0);
481 }
482
483 /**
484  * kvm_mips_set_count_hz() - Update the frequency of the timer.
485  * @vcpu:       Virtual CPU.
486  * @count_hz:   Frequency of CP0_Count timer in Hz.
487  *
488  * Change the frequency of the CP0_Count timer. This is done atomically so that
489  * CP0_Count is continuous and no timer interrupt is lost.
490  *
491  * Returns:     -EINVAL if @count_hz is out of range.
492  *              0 on success.
493  */
494 int kvm_mips_set_count_hz(struct kvm_vcpu *vcpu, s64 count_hz)
495 {
496         struct mips_coproc *cop0 = vcpu->arch.cop0;
497         int dc;
498         ktime_t now;
499         u32 count;
500
501         /* ensure the frequency is in a sensible range... */
502         if (count_hz <= 0 || count_hz > NSEC_PER_SEC)
503                 return -EINVAL;
504         /* ... and has actually changed */
505         if (vcpu->arch.count_hz == count_hz)
506                 return 0;
507
508         /* Safely freeze timer so we can keep it continuous */
509         dc = kvm_mips_count_disabled(vcpu);
510         if (dc) {
511                 now = kvm_mips_count_time(vcpu);
512                 count = kvm_read_c0_guest_count(cop0);
513         } else {
514                 now = kvm_mips_freeze_hrtimer(vcpu, &count);
515         }
516
517         /* Update the frequency */
518         vcpu->arch.count_hz = count_hz;
519         vcpu->arch.count_period = div_u64((u64)NSEC_PER_SEC << 32, count_hz);
520         vcpu->arch.count_dyn_bias = 0;
521
522         /* Calculate adjusted bias so dynamic count is unchanged */
523         vcpu->arch.count_bias = count - kvm_mips_ktime_to_count(vcpu, now);
524
525         /* Update and resume hrtimer */
526         if (!dc)
527                 kvm_mips_resume_hrtimer(vcpu, now, count);
528         return 0;
529 }
530
531 /**
532  * kvm_mips_write_compare() - Modify compare and update timer.
533  * @vcpu:       Virtual CPU.
534  * @compare:    New CP0_Compare value.
535  * @ack:        Whether to acknowledge timer interrupt.
536  *
537  * Update CP0_Compare to a new value and update the timeout.
538  * If @ack, atomically acknowledge any pending timer interrupt, otherwise ensure
539  * any pending timer interrupt is preserved.
540  */
541 void kvm_mips_write_compare(struct kvm_vcpu *vcpu, uint32_t compare, bool ack)
542 {
543         struct mips_coproc *cop0 = vcpu->arch.cop0;
544         int dc;
545         u32 old_compare = kvm_read_c0_guest_compare(cop0);
546         ktime_t now;
547         uint32_t count;
548
549         /* if unchanged, must just be an ack */
550         if (old_compare == compare) {
551                 if (!ack)
552                         return;
553                 kvm_mips_callbacks->dequeue_timer_int(vcpu);
554                 kvm_write_c0_guest_compare(cop0, compare);
555                 return;
556         }
557
558         /* freeze_hrtimer() takes care of timer interrupts <= count */
559         dc = kvm_mips_count_disabled(vcpu);
560         if (!dc)
561                 now = kvm_mips_freeze_hrtimer(vcpu, &count);
562
563         if (ack)
564                 kvm_mips_callbacks->dequeue_timer_int(vcpu);
565
566         kvm_write_c0_guest_compare(cop0, compare);
567
568         /* resume_hrtimer() takes care of timer interrupts > count */
569         if (!dc)
570                 kvm_mips_resume_hrtimer(vcpu, now, count);
571 }
572
573 /**
574  * kvm_mips_count_disable() - Disable count.
575  * @vcpu:       Virtual CPU.
576  *
577  * Disable the CP0_Count timer. A timer interrupt on or before the final stop
578  * time will be handled but not after.
579  *
580  * Assumes CP0_Count was previously enabled but now Guest.CP0_Cause.DC or
581  * count_ctl.DC has been set (count disabled).
582  *
583  * Returns:     The time that the timer was stopped.
584  */
585 static ktime_t kvm_mips_count_disable(struct kvm_vcpu *vcpu)
586 {
587         struct mips_coproc *cop0 = vcpu->arch.cop0;
588         uint32_t count;
589         ktime_t now;
590
591         /* Stop hrtimer */
592         hrtimer_cancel(&vcpu->arch.comparecount_timer);
593
594         /* Set the static count from the dynamic count, handling pending TI */
595         now = ktime_get();
596         count = kvm_mips_read_count_running(vcpu, now);
597         kvm_write_c0_guest_count(cop0, count);
598
599         return now;
600 }
601
602 /**
603  * kvm_mips_count_disable_cause() - Disable count using CP0_Cause.DC.
604  * @vcpu:       Virtual CPU.
605  *
606  * Disable the CP0_Count timer and set CP0_Cause.DC. A timer interrupt on or
607  * before the final stop time will be handled if the timer isn't disabled by
608  * count_ctl.DC, but not after.
609  *
610  * Assumes CP0_Cause.DC is clear (count enabled).
611  */
612 void kvm_mips_count_disable_cause(struct kvm_vcpu *vcpu)
613 {
614         struct mips_coproc *cop0 = vcpu->arch.cop0;
615
616         kvm_set_c0_guest_cause(cop0, CAUSEF_DC);
617         if (!(vcpu->arch.count_ctl & KVM_REG_MIPS_COUNT_CTL_DC))
618                 kvm_mips_count_disable(vcpu);
619 }
620
621 /**
622  * kvm_mips_count_enable_cause() - Enable count using CP0_Cause.DC.
623  * @vcpu:       Virtual CPU.
624  *
625  * Enable the CP0_Count timer and clear CP0_Cause.DC. A timer interrupt after
626  * the start time will be handled if the timer isn't disabled by count_ctl.DC,
627  * potentially before even returning, so the caller should be careful with
628  * ordering of CP0_Cause modifications so as not to lose it.
629  *
630  * Assumes CP0_Cause.DC is set (count disabled).
631  */
632 void kvm_mips_count_enable_cause(struct kvm_vcpu *vcpu)
633 {
634         struct mips_coproc *cop0 = vcpu->arch.cop0;
635         uint32_t count;
636
637         kvm_clear_c0_guest_cause(cop0, CAUSEF_DC);
638
639         /*
640          * Set the dynamic count to match the static count.
641          * This starts the hrtimer if count_ctl.DC allows it.
642          * Otherwise it conveniently updates the biases.
643          */
644         count = kvm_read_c0_guest_count(cop0);
645         kvm_mips_write_count(vcpu, count);
646 }
647
648 /**
649  * kvm_mips_set_count_ctl() - Update the count control KVM register.
650  * @vcpu:       Virtual CPU.
651  * @count_ctl:  Count control register new value.
652  *
653  * Set the count control KVM register. The timer is updated accordingly.
654  *
655  * Returns:     -EINVAL if reserved bits are set.
656  *              0 on success.
657  */
658 int kvm_mips_set_count_ctl(struct kvm_vcpu *vcpu, s64 count_ctl)
659 {
660         struct mips_coproc *cop0 = vcpu->arch.cop0;
661         s64 changed = count_ctl ^ vcpu->arch.count_ctl;
662         s64 delta;
663         ktime_t expire, now;
664         uint32_t count, compare;
665
666         /* Only allow defined bits to be changed */
667         if (changed & ~(s64)(KVM_REG_MIPS_COUNT_CTL_DC))
668                 return -EINVAL;
669
670         /* Apply new value */
671         vcpu->arch.count_ctl = count_ctl;
672
673         /* Master CP0_Count disable */
674         if (changed & KVM_REG_MIPS_COUNT_CTL_DC) {
675                 /* Is CP0_Cause.DC already disabling CP0_Count? */
676                 if (kvm_read_c0_guest_cause(cop0) & CAUSEF_DC) {
677                         if (count_ctl & KVM_REG_MIPS_COUNT_CTL_DC)
678                                 /* Just record the current time */
679                                 vcpu->arch.count_resume = ktime_get();
680                 } else if (count_ctl & KVM_REG_MIPS_COUNT_CTL_DC) {
681                         /* disable timer and record current time */
682                         vcpu->arch.count_resume = kvm_mips_count_disable(vcpu);
683                 } else {
684                         /*
685                          * Calculate timeout relative to static count at resume
686                          * time (wrap 0 to 2^32).
687                          */
688                         count = kvm_read_c0_guest_count(cop0);
689                         compare = kvm_read_c0_guest_compare(cop0);
690                         delta = (u64)(uint32_t)(compare - count - 1) + 1;
691                         delta = div_u64(delta * NSEC_PER_SEC,
692                                         vcpu->arch.count_hz);
693                         expire = ktime_add_ns(vcpu->arch.count_resume, delta);
694
695                         /* Handle pending interrupt */
696                         now = ktime_get();
697                         if (ktime_compare(now, expire) >= 0)
698                                 /* Nothing should be waiting on the timeout */
699                                 kvm_mips_callbacks->queue_timer_int(vcpu);
700
701                         /* Resume hrtimer without changing bias */
702                         count = kvm_mips_read_count_running(vcpu, now);
703                         kvm_mips_resume_hrtimer(vcpu, now, count);
704                 }
705         }
706
707         return 0;
708 }
709
710 /**
711  * kvm_mips_set_count_resume() - Update the count resume KVM register.
712  * @vcpu:               Virtual CPU.
713  * @count_resume:       Count resume register new value.
714  *
715  * Set the count resume KVM register.
716  *
717  * Returns:     -EINVAL if out of valid range (0..now).
718  *              0 on success.
719  */
720 int kvm_mips_set_count_resume(struct kvm_vcpu *vcpu, s64 count_resume)
721 {
722         /*
723          * It doesn't make sense for the resume time to be in the future, as it
724          * would be possible for the next interrupt to be more than a full
725          * period in the future.
726          */
727         if (count_resume < 0 || count_resume > ktime_to_ns(ktime_get()))
728                 return -EINVAL;
729
730         vcpu->arch.count_resume = ns_to_ktime(count_resume);
731         return 0;
732 }
733
734 /**
735  * kvm_mips_count_timeout() - Push timer forward on timeout.
736  * @vcpu:       Virtual CPU.
737  *
738  * Handle an hrtimer event by push the hrtimer forward a period.
739  *
740  * Returns:     The hrtimer_restart value to return to the hrtimer subsystem.
741  */
742 enum hrtimer_restart kvm_mips_count_timeout(struct kvm_vcpu *vcpu)
743 {
744         /* Add the Count period to the current expiry time */
745         hrtimer_add_expires_ns(&vcpu->arch.comparecount_timer,
746                                vcpu->arch.count_period);
747         return HRTIMER_RESTART;
748 }
749
750 enum emulation_result kvm_mips_emul_eret(struct kvm_vcpu *vcpu)
751 {
752         struct mips_coproc *cop0 = vcpu->arch.cop0;
753         enum emulation_result er = EMULATE_DONE;
754
755         if (kvm_read_c0_guest_status(cop0) & ST0_EXL) {
756                 kvm_debug("[%#lx] ERET to %#lx\n", vcpu->arch.pc,
757                           kvm_read_c0_guest_epc(cop0));
758                 kvm_clear_c0_guest_status(cop0, ST0_EXL);
759                 vcpu->arch.pc = kvm_read_c0_guest_epc(cop0);
760
761         } else if (kvm_read_c0_guest_status(cop0) & ST0_ERL) {
762                 kvm_clear_c0_guest_status(cop0, ST0_ERL);
763                 vcpu->arch.pc = kvm_read_c0_guest_errorepc(cop0);
764         } else {
765                 kvm_err("[%#lx] ERET when MIPS_SR_EXL|MIPS_SR_ERL == 0\n",
766                         vcpu->arch.pc);
767                 er = EMULATE_FAIL;
768         }
769
770         return er;
771 }
772
773 enum emulation_result kvm_mips_emul_wait(struct kvm_vcpu *vcpu)
774 {
775         kvm_debug("[%#lx] !!!WAIT!!! (%#lx)\n", vcpu->arch.pc,
776                   vcpu->arch.pending_exceptions);
777
778         ++vcpu->stat.wait_exits;
779         trace_kvm_exit(vcpu, WAIT_EXITS);
780         if (!vcpu->arch.pending_exceptions) {
781                 vcpu->arch.wait = 1;
782                 kvm_vcpu_block(vcpu);
783
784                 /*
785                  * We we are runnable, then definitely go off to user space to
786                  * check if any I/O interrupts are pending.
787                  */
788                 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
789                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
790                         vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
791                 }
792         }
793
794         return EMULATE_DONE;
795 }
796
797 /*
798  * XXXKYMA: Linux doesn't seem to use TLBR, return EMULATE_FAIL for now so that
799  * we can catch this, if things ever change
800  */
801 enum emulation_result kvm_mips_emul_tlbr(struct kvm_vcpu *vcpu)
802 {
803         struct mips_coproc *cop0 = vcpu->arch.cop0;
804         uint32_t pc = vcpu->arch.pc;
805
806         kvm_err("[%#x] COP0_TLBR [%ld]\n", pc, kvm_read_c0_guest_index(cop0));
807         return EMULATE_FAIL;
808 }
809
810 /* Write Guest TLB Entry @ Index */
811 enum emulation_result kvm_mips_emul_tlbwi(struct kvm_vcpu *vcpu)
812 {
813         struct mips_coproc *cop0 = vcpu->arch.cop0;
814         int index = kvm_read_c0_guest_index(cop0);
815         struct kvm_mips_tlb *tlb = NULL;
816         uint32_t pc = vcpu->arch.pc;
817
818         if (index < 0 || index >= KVM_MIPS_GUEST_TLB_SIZE) {
819                 kvm_debug("%s: illegal index: %d\n", __func__, index);
820                 kvm_debug("[%#x] COP0_TLBWI [%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx, mask: %#lx)\n",
821                           pc, index, kvm_read_c0_guest_entryhi(cop0),
822                           kvm_read_c0_guest_entrylo0(cop0),
823                           kvm_read_c0_guest_entrylo1(cop0),
824                           kvm_read_c0_guest_pagemask(cop0));
825                 index = (index & ~0x80000000) % KVM_MIPS_GUEST_TLB_SIZE;
826         }
827
828         tlb = &vcpu->arch.guest_tlb[index];
829         /*
830          * Probe the shadow host TLB for the entry being overwritten, if one
831          * matches, invalidate it
832          */
833         kvm_mips_host_tlb_inv(vcpu, tlb->tlb_hi);
834
835         tlb->tlb_mask = kvm_read_c0_guest_pagemask(cop0);
836         tlb->tlb_hi = kvm_read_c0_guest_entryhi(cop0);
837         tlb->tlb_lo0 = kvm_read_c0_guest_entrylo0(cop0);
838         tlb->tlb_lo1 = kvm_read_c0_guest_entrylo1(cop0);
839
840         kvm_debug("[%#x] COP0_TLBWI [%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx, mask: %#lx)\n",
841                   pc, index, kvm_read_c0_guest_entryhi(cop0),
842                   kvm_read_c0_guest_entrylo0(cop0),
843                   kvm_read_c0_guest_entrylo1(cop0),
844                   kvm_read_c0_guest_pagemask(cop0));
845
846         return EMULATE_DONE;
847 }
848
849 /* Write Guest TLB Entry @ Random Index */
850 enum emulation_result kvm_mips_emul_tlbwr(struct kvm_vcpu *vcpu)
851 {
852         struct mips_coproc *cop0 = vcpu->arch.cop0;
853         struct kvm_mips_tlb *tlb = NULL;
854         uint32_t pc = vcpu->arch.pc;
855         int index;
856
857         get_random_bytes(&index, sizeof(index));
858         index &= (KVM_MIPS_GUEST_TLB_SIZE - 1);
859
860         tlb = &vcpu->arch.guest_tlb[index];
861
862         /*
863          * Probe the shadow host TLB for the entry being overwritten, if one
864          * matches, invalidate it
865          */
866         kvm_mips_host_tlb_inv(vcpu, tlb->tlb_hi);
867
868         tlb->tlb_mask = kvm_read_c0_guest_pagemask(cop0);
869         tlb->tlb_hi = kvm_read_c0_guest_entryhi(cop0);
870         tlb->tlb_lo0 = kvm_read_c0_guest_entrylo0(cop0);
871         tlb->tlb_lo1 = kvm_read_c0_guest_entrylo1(cop0);
872
873         kvm_debug("[%#x] COP0_TLBWR[%d] (entryhi: %#lx, entrylo0: %#lx entrylo1: %#lx)\n",
874                   pc, index, kvm_read_c0_guest_entryhi(cop0),
875                   kvm_read_c0_guest_entrylo0(cop0),
876                   kvm_read_c0_guest_entrylo1(cop0));
877
878         return EMULATE_DONE;
879 }
880
881 enum emulation_result kvm_mips_emul_tlbp(struct kvm_vcpu *vcpu)
882 {
883         struct mips_coproc *cop0 = vcpu->arch.cop0;
884         long entryhi = kvm_read_c0_guest_entryhi(cop0);
885         uint32_t pc = vcpu->arch.pc;
886         int index = -1;
887
888         index = kvm_mips_guest_tlb_lookup(vcpu, entryhi);
889
890         kvm_write_c0_guest_index(cop0, index);
891
892         kvm_debug("[%#x] COP0_TLBP (entryhi: %#lx), index: %d\n", pc, entryhi,
893                   index);
894
895         return EMULATE_DONE;
896 }
897
898 /**
899  * kvm_mips_config1_wrmask() - Find mask of writable bits in guest Config1
900  * @vcpu:       Virtual CPU.
901  *
902  * Finds the mask of bits which are writable in the guest's Config1 CP0
903  * register, by userland (currently read-only to the guest).
904  */
905 unsigned int kvm_mips_config1_wrmask(struct kvm_vcpu *vcpu)
906 {
907         unsigned int mask = 0;
908
909         /* Permit FPU to be present if FPU is supported */
910         if (kvm_mips_guest_can_have_fpu(&vcpu->arch))
911                 mask |= MIPS_CONF1_FP;
912
913         return mask;
914 }
915
916 /**
917  * kvm_mips_config3_wrmask() - Find mask of writable bits in guest Config3
918  * @vcpu:       Virtual CPU.
919  *
920  * Finds the mask of bits which are writable in the guest's Config3 CP0
921  * register, by userland (currently read-only to the guest).
922  */
923 unsigned int kvm_mips_config3_wrmask(struct kvm_vcpu *vcpu)
924 {
925         /* Config4 is optional */
926         unsigned int mask = MIPS_CONF_M;
927
928         /* Permit MSA to be present if MSA is supported */
929         if (kvm_mips_guest_can_have_msa(&vcpu->arch))
930                 mask |= MIPS_CONF3_MSA;
931
932         return mask;
933 }
934
935 /**
936  * kvm_mips_config4_wrmask() - Find mask of writable bits in guest Config4
937  * @vcpu:       Virtual CPU.
938  *
939  * Finds the mask of bits which are writable in the guest's Config4 CP0
940  * register, by userland (currently read-only to the guest).
941  */
942 unsigned int kvm_mips_config4_wrmask(struct kvm_vcpu *vcpu)
943 {
944         /* Config5 is optional */
945         return MIPS_CONF_M;
946 }
947
948 /**
949  * kvm_mips_config5_wrmask() - Find mask of writable bits in guest Config5
950  * @vcpu:       Virtual CPU.
951  *
952  * Finds the mask of bits which are writable in the guest's Config5 CP0
953  * register, by the guest itself.
954  */
955 unsigned int kvm_mips_config5_wrmask(struct kvm_vcpu *vcpu)
956 {
957         unsigned int mask = 0;
958
959         /* Permit MSAEn changes if MSA supported and enabled */
960         if (kvm_mips_guest_has_msa(&vcpu->arch))
961                 mask |= MIPS_CONF5_MSAEN;
962
963         /*
964          * Permit guest FPU mode changes if FPU is enabled and the relevant
965          * feature exists according to FIR register.
966          */
967         if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
968                 if (cpu_has_fre)
969                         mask |= MIPS_CONF5_FRE;
970                 /* We don't support UFR or UFE */
971         }
972
973         return mask;
974 }
975
976 enum emulation_result kvm_mips_emulate_CP0(uint32_t inst, uint32_t *opc,
977                                            uint32_t cause, struct kvm_run *run,
978                                            struct kvm_vcpu *vcpu)
979 {
980         struct mips_coproc *cop0 = vcpu->arch.cop0;
981         enum emulation_result er = EMULATE_DONE;
982         int32_t rt, rd, copz, sel, co_bit, op;
983         uint32_t pc = vcpu->arch.pc;
984         unsigned long curr_pc;
985
986         /*
987          * Update PC and hold onto current PC in case there is
988          * an error and we want to rollback the PC
989          */
990         curr_pc = vcpu->arch.pc;
991         er = update_pc(vcpu, cause);
992         if (er == EMULATE_FAIL)
993                 return er;
994
995         copz = (inst >> 21) & 0x1f;
996         rt = (inst >> 16) & 0x1f;
997         rd = (inst >> 11) & 0x1f;
998         sel = inst & 0x7;
999         co_bit = (inst >> 25) & 1;
1000
1001         if (co_bit) {
1002                 op = (inst) & 0xff;
1003
1004                 switch (op) {
1005                 case tlbr_op:   /*  Read indexed TLB entry  */
1006                         er = kvm_mips_emul_tlbr(vcpu);
1007                         break;
1008                 case tlbwi_op:  /*  Write indexed  */
1009                         er = kvm_mips_emul_tlbwi(vcpu);
1010                         break;
1011                 case tlbwr_op:  /*  Write random  */
1012                         er = kvm_mips_emul_tlbwr(vcpu);
1013                         break;
1014                 case tlbp_op:   /* TLB Probe */
1015                         er = kvm_mips_emul_tlbp(vcpu);
1016                         break;
1017                 case rfe_op:
1018                         kvm_err("!!!COP0_RFE!!!\n");
1019                         break;
1020                 case eret_op:
1021                         er = kvm_mips_emul_eret(vcpu);
1022                         goto dont_update_pc;
1023                         break;
1024                 case wait_op:
1025                         er = kvm_mips_emul_wait(vcpu);
1026                         break;
1027                 }
1028         } else {
1029                 switch (copz) {
1030                 case mfc_op:
1031 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
1032                         cop0->stat[rd][sel]++;
1033 #endif
1034                         /* Get reg */
1035                         if ((rd == MIPS_CP0_COUNT) && (sel == 0)) {
1036                                 vcpu->arch.gprs[rt] = kvm_mips_read_count(vcpu);
1037                         } else if ((rd == MIPS_CP0_ERRCTL) && (sel == 0)) {
1038                                 vcpu->arch.gprs[rt] = 0x0;
1039 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1040                                 kvm_mips_trans_mfc0(inst, opc, vcpu);
1041 #endif
1042                         } else {
1043                                 vcpu->arch.gprs[rt] = cop0->reg[rd][sel];
1044
1045 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1046                                 kvm_mips_trans_mfc0(inst, opc, vcpu);
1047 #endif
1048                         }
1049
1050                         kvm_debug
1051                             ("[%#x] MFCz[%d][%d], vcpu->arch.gprs[%d]: %#lx\n",
1052                              pc, rd, sel, rt, vcpu->arch.gprs[rt]);
1053
1054                         break;
1055
1056                 case dmfc_op:
1057                         vcpu->arch.gprs[rt] = cop0->reg[rd][sel];
1058                         break;
1059
1060                 case mtc_op:
1061 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
1062                         cop0->stat[rd][sel]++;
1063 #endif
1064                         if ((rd == MIPS_CP0_TLB_INDEX)
1065                             && (vcpu->arch.gprs[rt] >=
1066                                 KVM_MIPS_GUEST_TLB_SIZE)) {
1067                                 kvm_err("Invalid TLB Index: %ld",
1068                                         vcpu->arch.gprs[rt]);
1069                                 er = EMULATE_FAIL;
1070                                 break;
1071                         }
1072 #define C0_EBASE_CORE_MASK 0xff
1073                         if ((rd == MIPS_CP0_PRID) && (sel == 1)) {
1074                                 /* Preserve CORE number */
1075                                 kvm_change_c0_guest_ebase(cop0,
1076                                                           ~(C0_EBASE_CORE_MASK),
1077                                                           vcpu->arch.gprs[rt]);
1078                                 kvm_err("MTCz, cop0->reg[EBASE]: %#lx\n",
1079                                         kvm_read_c0_guest_ebase(cop0));
1080                         } else if (rd == MIPS_CP0_TLB_HI && sel == 0) {
1081                                 uint32_t nasid =
1082                                         vcpu->arch.gprs[rt] & ASID_MASK;
1083                                 if ((KSEGX(vcpu->arch.gprs[rt]) != CKSEG0) &&
1084                                     ((kvm_read_c0_guest_entryhi(cop0) &
1085                                       ASID_MASK) != nasid)) {
1086                                         kvm_debug("MTCz, change ASID from %#lx to %#lx\n",
1087                                                 kvm_read_c0_guest_entryhi(cop0)
1088                                                 & ASID_MASK,
1089                                                 vcpu->arch.gprs[rt]
1090                                                 & ASID_MASK);
1091
1092                                         /* Blow away the shadow host TLBs */
1093                                         kvm_mips_flush_host_tlb(1);
1094                                 }
1095                                 kvm_write_c0_guest_entryhi(cop0,
1096                                                            vcpu->arch.gprs[rt]);
1097                         }
1098                         /* Are we writing to COUNT */
1099                         else if ((rd == MIPS_CP0_COUNT) && (sel == 0)) {
1100                                 kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
1101                                 goto done;
1102                         } else if ((rd == MIPS_CP0_COMPARE) && (sel == 0)) {
1103                                 kvm_debug("[%#x] MTCz, COMPARE %#lx <- %#lx\n",
1104                                           pc, kvm_read_c0_guest_compare(cop0),
1105                                           vcpu->arch.gprs[rt]);
1106
1107                                 /* If we are writing to COMPARE */
1108                                 /* Clear pending timer interrupt, if any */
1109                                 kvm_mips_write_compare(vcpu,
1110                                                        vcpu->arch.gprs[rt],
1111                                                        true);
1112                         } else if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1113                                 unsigned int old_val, val, change;
1114
1115                                 old_val = kvm_read_c0_guest_status(cop0);
1116                                 val = vcpu->arch.gprs[rt];
1117                                 change = val ^ old_val;
1118
1119                                 /* Make sure that the NMI bit is never set */
1120                                 val &= ~ST0_NMI;
1121
1122                                 /*
1123                                  * Don't allow CU1 or FR to be set unless FPU
1124                                  * capability enabled and exists in guest
1125                                  * configuration.
1126                                  */
1127                                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1128                                         val &= ~(ST0_CU1 | ST0_FR);
1129
1130                                 /*
1131                                  * Also don't allow FR to be set if host doesn't
1132                                  * support it.
1133                                  */
1134                                 if (!(current_cpu_data.fpu_id & MIPS_FPIR_F64))
1135                                         val &= ~ST0_FR;
1136
1137
1138                                 /* Handle changes in FPU mode */
1139                                 preempt_disable();
1140
1141                                 /*
1142                                  * FPU and Vector register state is made
1143                                  * UNPREDICTABLE by a change of FR, so don't
1144                                  * even bother saving it.
1145                                  */
1146                                 if (change & ST0_FR)
1147                                         kvm_drop_fpu(vcpu);
1148
1149                                 /*
1150                                  * If MSA state is already live, it is undefined
1151                                  * how it interacts with FR=0 FPU state, and we
1152                                  * don't want to hit reserved instruction
1153                                  * exceptions trying to save the MSA state later
1154                                  * when CU=1 && FR=1, so play it safe and save
1155                                  * it first.
1156                                  */
1157                                 if (change & ST0_CU1 && !(val & ST0_FR) &&
1158                                     vcpu->arch.fpu_inuse & KVM_MIPS_FPU_MSA)
1159                                         kvm_lose_fpu(vcpu);
1160
1161                                 /*
1162                                  * Propagate CU1 (FPU enable) changes
1163                                  * immediately if the FPU context is already
1164                                  * loaded. When disabling we leave the context
1165                                  * loaded so it can be quickly enabled again in
1166                                  * the near future.
1167                                  */
1168                                 if (change & ST0_CU1 &&
1169                                     vcpu->arch.fpu_inuse & KVM_MIPS_FPU_FPU)
1170                                         change_c0_status(ST0_CU1, val);
1171
1172                                 preempt_enable();
1173
1174                                 kvm_write_c0_guest_status(cop0, val);
1175
1176 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1177                                 /*
1178                                  * If FPU present, we need CU1/FR bits to take
1179                                  * effect fairly soon.
1180                                  */
1181                                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1182                                         kvm_mips_trans_mtc0(inst, opc, vcpu);
1183 #endif
1184                         } else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1185                                 unsigned int old_val, val, change, wrmask;
1186
1187                                 old_val = kvm_read_c0_guest_config5(cop0);
1188                                 val = vcpu->arch.gprs[rt];
1189
1190                                 /* Only a few bits are writable in Config5 */
1191                                 wrmask = kvm_mips_config5_wrmask(vcpu);
1192                                 change = (val ^ old_val) & wrmask;
1193                                 val = old_val ^ change;
1194
1195
1196                                 /* Handle changes in FPU/MSA modes */
1197                                 preempt_disable();
1198
1199                                 /*
1200                                  * Propagate FRE changes immediately if the FPU
1201                                  * context is already loaded.
1202                                  */
1203                                 if (change & MIPS_CONF5_FRE &&
1204                                     vcpu->arch.fpu_inuse & KVM_MIPS_FPU_FPU)
1205                                         change_c0_config5(MIPS_CONF5_FRE, val);
1206
1207                                 /*
1208                                  * Propagate MSAEn changes immediately if the
1209                                  * MSA context is already loaded. When disabling
1210                                  * we leave the context loaded so it can be
1211                                  * quickly enabled again in the near future.
1212                                  */
1213                                 if (change & MIPS_CONF5_MSAEN &&
1214                                     vcpu->arch.fpu_inuse & KVM_MIPS_FPU_MSA)
1215                                         change_c0_config5(MIPS_CONF5_MSAEN,
1216                                                           val);
1217
1218                                 preempt_enable();
1219
1220                                 kvm_write_c0_guest_config5(cop0, val);
1221                         } else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1222                                 uint32_t old_cause, new_cause;
1223
1224                                 old_cause = kvm_read_c0_guest_cause(cop0);
1225                                 new_cause = vcpu->arch.gprs[rt];
1226                                 /* Update R/W bits */
1227                                 kvm_change_c0_guest_cause(cop0, 0x08800300,
1228                                                           new_cause);
1229                                 /* DC bit enabling/disabling timer? */
1230                                 if ((old_cause ^ new_cause) & CAUSEF_DC) {
1231                                         if (new_cause & CAUSEF_DC)
1232                                                 kvm_mips_count_disable_cause(vcpu);
1233                                         else
1234                                                 kvm_mips_count_enable_cause(vcpu);
1235                                 }
1236                         } else {
1237                                 cop0->reg[rd][sel] = vcpu->arch.gprs[rt];
1238 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1239                                 kvm_mips_trans_mtc0(inst, opc, vcpu);
1240 #endif
1241                         }
1242
1243                         kvm_debug("[%#x] MTCz, cop0->reg[%d][%d]: %#lx\n", pc,
1244                                   rd, sel, cop0->reg[rd][sel]);
1245                         break;
1246
1247                 case dmtc_op:
1248                         kvm_err("!!!!!!![%#lx]dmtc_op: rt: %d, rd: %d, sel: %d!!!!!!\n",
1249                                 vcpu->arch.pc, rt, rd, sel);
1250                         er = EMULATE_FAIL;
1251                         break;
1252
1253                 case mfmcz_op:
1254 #ifdef KVM_MIPS_DEBUG_COP0_COUNTERS
1255                         cop0->stat[MIPS_CP0_STATUS][0]++;
1256 #endif
1257                         if (rt != 0) {
1258                                 vcpu->arch.gprs[rt] =
1259                                     kvm_read_c0_guest_status(cop0);
1260                         }
1261                         /* EI */
1262                         if (inst & 0x20) {
1263                                 kvm_debug("[%#lx] mfmcz_op: EI\n",
1264                                           vcpu->arch.pc);
1265                                 kvm_set_c0_guest_status(cop0, ST0_IE);
1266                         } else {
1267                                 kvm_debug("[%#lx] mfmcz_op: DI\n",
1268                                           vcpu->arch.pc);
1269                                 kvm_clear_c0_guest_status(cop0, ST0_IE);
1270                         }
1271
1272                         break;
1273
1274                 case wrpgpr_op:
1275                         {
1276                                 uint32_t css =
1277                                     cop0->reg[MIPS_CP0_STATUS][2] & 0xf;
1278                                 uint32_t pss =
1279                                     (cop0->reg[MIPS_CP0_STATUS][2] >> 6) & 0xf;
1280                                 /*
1281                                  * We don't support any shadow register sets, so
1282                                  * SRSCtl[PSS] == SRSCtl[CSS] = 0
1283                                  */
1284                                 if (css || pss) {
1285                                         er = EMULATE_FAIL;
1286                                         break;
1287                                 }
1288                                 kvm_debug("WRPGPR[%d][%d] = %#lx\n", pss, rd,
1289                                           vcpu->arch.gprs[rt]);
1290                                 vcpu->arch.gprs[rd] = vcpu->arch.gprs[rt];
1291                         }
1292                         break;
1293                 default:
1294                         kvm_err("[%#lx]MachEmulateCP0: unsupported COP0, copz: 0x%x\n",
1295                                 vcpu->arch.pc, copz);
1296                         er = EMULATE_FAIL;
1297                         break;
1298                 }
1299         }
1300
1301 done:
1302         /* Rollback PC only if emulation was unsuccessful */
1303         if (er == EMULATE_FAIL)
1304                 vcpu->arch.pc = curr_pc;
1305
1306 dont_update_pc:
1307         /*
1308          * This is for special instructions whose emulation
1309          * updates the PC, so do not overwrite the PC under
1310          * any circumstances
1311          */
1312
1313         return er;
1314 }
1315
1316 enum emulation_result kvm_mips_emulate_store(uint32_t inst, uint32_t cause,
1317                                              struct kvm_run *run,
1318                                              struct kvm_vcpu *vcpu)
1319 {
1320         enum emulation_result er = EMULATE_DO_MMIO;
1321         int32_t op, base, rt, offset;
1322         uint32_t bytes;
1323         void *data = run->mmio.data;
1324         unsigned long curr_pc;
1325
1326         /*
1327          * Update PC and hold onto current PC in case there is
1328          * an error and we want to rollback the PC
1329          */
1330         curr_pc = vcpu->arch.pc;
1331         er = update_pc(vcpu, cause);
1332         if (er == EMULATE_FAIL)
1333                 return er;
1334
1335         rt = (inst >> 16) & 0x1f;
1336         base = (inst >> 21) & 0x1f;
1337         offset = inst & 0xffff;
1338         op = (inst >> 26) & 0x3f;
1339
1340         switch (op) {
1341         case sb_op:
1342                 bytes = 1;
1343                 if (bytes > sizeof(run->mmio.data)) {
1344                         kvm_err("%s: bad MMIO length: %d\n", __func__,
1345                                run->mmio.len);
1346                 }
1347                 run->mmio.phys_addr =
1348                     kvm_mips_callbacks->gva_to_gpa(vcpu->arch.
1349                                                    host_cp0_badvaddr);
1350                 if (run->mmio.phys_addr == KVM_INVALID_ADDR) {
1351                         er = EMULATE_FAIL;
1352                         break;
1353                 }
1354                 run->mmio.len = bytes;
1355                 run->mmio.is_write = 1;
1356                 vcpu->mmio_needed = 1;
1357                 vcpu->mmio_is_write = 1;
1358                 *(u8 *) data = vcpu->arch.gprs[rt];
1359                 kvm_debug("OP_SB: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1360                           vcpu->arch.host_cp0_badvaddr, vcpu->arch.gprs[rt],
1361                           *(uint8_t *) data);
1362
1363                 break;
1364
1365         case sw_op:
1366                 bytes = 4;
1367                 if (bytes > sizeof(run->mmio.data)) {
1368                         kvm_err("%s: bad MMIO length: %d\n", __func__,
1369                                run->mmio.len);
1370                 }
1371                 run->mmio.phys_addr =
1372                     kvm_mips_callbacks->gva_to_gpa(vcpu->arch.
1373                                                    host_cp0_badvaddr);
1374                 if (run->mmio.phys_addr == KVM_INVALID_ADDR) {
1375                         er = EMULATE_FAIL;
1376                         break;
1377                 }
1378
1379                 run->mmio.len = bytes;
1380                 run->mmio.is_write = 1;
1381                 vcpu->mmio_needed = 1;
1382                 vcpu->mmio_is_write = 1;
1383                 *(uint32_t *) data = vcpu->arch.gprs[rt];
1384
1385                 kvm_debug("[%#lx] OP_SW: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1386                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1387                           vcpu->arch.gprs[rt], *(uint32_t *) data);
1388                 break;
1389
1390         case sh_op:
1391                 bytes = 2;
1392                 if (bytes > sizeof(run->mmio.data)) {
1393                         kvm_err("%s: bad MMIO length: %d\n", __func__,
1394                                run->mmio.len);
1395                 }
1396                 run->mmio.phys_addr =
1397                     kvm_mips_callbacks->gva_to_gpa(vcpu->arch.
1398                                                    host_cp0_badvaddr);
1399                 if (run->mmio.phys_addr == KVM_INVALID_ADDR) {
1400                         er = EMULATE_FAIL;
1401                         break;
1402                 }
1403
1404                 run->mmio.len = bytes;
1405                 run->mmio.is_write = 1;
1406                 vcpu->mmio_needed = 1;
1407                 vcpu->mmio_is_write = 1;
1408                 *(uint16_t *) data = vcpu->arch.gprs[rt];
1409
1410                 kvm_debug("[%#lx] OP_SH: eaddr: %#lx, gpr: %#lx, data: %#x\n",
1411                           vcpu->arch.pc, vcpu->arch.host_cp0_badvaddr,
1412                           vcpu->arch.gprs[rt], *(uint32_t *) data);
1413                 break;
1414
1415         default:
1416                 kvm_err("Store not yet supported");
1417                 er = EMULATE_FAIL;
1418                 break;
1419         }
1420
1421         /* Rollback PC if emulation was unsuccessful */
1422         if (er == EMULATE_FAIL)
1423                 vcpu->arch.pc = curr_pc;
1424
1425         return er;
1426 }
1427
1428 enum emulation_result kvm_mips_emulate_load(uint32_t inst, uint32_t cause,
1429                                             struct kvm_run *run,
1430                                             struct kvm_vcpu *vcpu)
1431 {
1432         enum emulation_result er = EMULATE_DO_MMIO;
1433         int32_t op, base, rt, offset;
1434         uint32_t bytes;
1435
1436         rt = (inst >> 16) & 0x1f;
1437         base = (inst >> 21) & 0x1f;
1438         offset = inst & 0xffff;
1439         op = (inst >> 26) & 0x3f;
1440
1441         vcpu->arch.pending_load_cause = cause;
1442         vcpu->arch.io_gpr = rt;
1443
1444         switch (op) {
1445         case lw_op:
1446                 bytes = 4;
1447                 if (bytes > sizeof(run->mmio.data)) {
1448                         kvm_err("%s: bad MMIO length: %d\n", __func__,
1449                                run->mmio.len);
1450                         er = EMULATE_FAIL;
1451                         break;
1452                 }
1453                 run->mmio.phys_addr =
1454                     kvm_mips_callbacks->gva_to_gpa(vcpu->arch.
1455                                                    host_cp0_badvaddr);
1456                 if (run->mmio.phys_addr == KVM_INVALID_ADDR) {
1457                         er = EMULATE_FAIL;
1458                         break;
1459                 }
1460
1461                 run->mmio.len = bytes;
1462                 run->mmio.is_write = 0;
1463                 vcpu->mmio_needed = 1;
1464                 vcpu->mmio_is_write = 0;
1465                 break;
1466
1467         case lh_op:
1468         case lhu_op:
1469                 bytes = 2;
1470                 if (bytes > sizeof(run->mmio.data)) {
1471                         kvm_err("%s: bad MMIO length: %d\n", __func__,
1472                                run->mmio.len);
1473                         er = EMULATE_FAIL;
1474                         break;
1475                 }
1476                 run->mmio.phys_addr =
1477                     kvm_mips_callbacks->gva_to_gpa(vcpu->arch.
1478                                                    host_cp0_badvaddr);
1479                 if (run->mmio.phys_addr == KVM_INVALID_ADDR) {
1480                         er = EMULATE_FAIL;
1481                         break;
1482                 }
1483
1484                 run->mmio.len = bytes;
1485                 run->mmio.is_write = 0;
1486                 vcpu->mmio_needed = 1;
1487                 vcpu->mmio_is_write = 0;
1488
1489                 if (op == lh_op)
1490                         vcpu->mmio_needed = 2;
1491                 else
1492                         vcpu->mmio_needed = 1;
1493
1494                 break;
1495
1496         case lbu_op:
1497         case lb_op:
1498                 bytes = 1;
1499                 if (bytes > sizeof(run->mmio.data)) {
1500                         kvm_err("%s: bad MMIO length: %d\n", __func__,
1501                                run->mmio.len);
1502                         er = EMULATE_FAIL;
1503                         break;
1504                 }
1505                 run->mmio.phys_addr =
1506                     kvm_mips_callbacks->gva_to_gpa(vcpu->arch.
1507                                                    host_cp0_badvaddr);
1508                 if (run->mmio.phys_addr == KVM_INVALID_ADDR) {
1509                         er = EMULATE_FAIL;
1510                         break;
1511                 }
1512
1513                 run->mmio.len = bytes;
1514                 run->mmio.is_write = 0;
1515                 vcpu->mmio_is_write = 0;
1516
1517                 if (op == lb_op)
1518                         vcpu->mmio_needed = 2;
1519                 else
1520                         vcpu->mmio_needed = 1;
1521
1522                 break;
1523
1524         default:
1525                 kvm_err("Load not yet supported");
1526                 er = EMULATE_FAIL;
1527                 break;
1528         }
1529
1530         return er;
1531 }
1532
1533 int kvm_mips_sync_icache(unsigned long va, struct kvm_vcpu *vcpu)
1534 {
1535         unsigned long offset = (va & ~PAGE_MASK);
1536         struct kvm *kvm = vcpu->kvm;
1537         unsigned long pa;
1538         gfn_t gfn;
1539         pfn_t pfn;
1540
1541         gfn = va >> PAGE_SHIFT;
1542
1543         if (gfn >= kvm->arch.guest_pmap_npages) {
1544                 kvm_err("%s: Invalid gfn: %#llx\n", __func__, gfn);
1545                 kvm_mips_dump_host_tlbs();
1546                 kvm_arch_vcpu_dump_regs(vcpu);
1547                 return -1;
1548         }
1549         pfn = kvm->arch.guest_pmap[gfn];
1550         pa = (pfn << PAGE_SHIFT) | offset;
1551
1552         kvm_debug("%s: va: %#lx, unmapped: %#x\n", __func__, va,
1553                   CKSEG0ADDR(pa));
1554
1555         local_flush_icache_range(CKSEG0ADDR(pa), 32);
1556         return 0;
1557 }
1558
1559 #define MIPS_CACHE_OP_INDEX_INV         0x0
1560 #define MIPS_CACHE_OP_INDEX_LD_TAG      0x1
1561 #define MIPS_CACHE_OP_INDEX_ST_TAG      0x2
1562 #define MIPS_CACHE_OP_IMP               0x3
1563 #define MIPS_CACHE_OP_HIT_INV           0x4
1564 #define MIPS_CACHE_OP_FILL_WB_INV       0x5
1565 #define MIPS_CACHE_OP_HIT_HB            0x6
1566 #define MIPS_CACHE_OP_FETCH_LOCK        0x7
1567
1568 #define MIPS_CACHE_ICACHE               0x0
1569 #define MIPS_CACHE_DCACHE               0x1
1570 #define MIPS_CACHE_SEC                  0x3
1571
1572 enum emulation_result kvm_mips_emulate_cache(uint32_t inst, uint32_t *opc,
1573                                              uint32_t cause,
1574                                              struct kvm_run *run,
1575                                              struct kvm_vcpu *vcpu)
1576 {
1577         struct mips_coproc *cop0 = vcpu->arch.cop0;
1578         enum emulation_result er = EMULATE_DONE;
1579         int32_t offset, cache, op_inst, op, base;
1580         struct kvm_vcpu_arch *arch = &vcpu->arch;
1581         unsigned long va;
1582         unsigned long curr_pc;
1583
1584         /*
1585          * Update PC and hold onto current PC in case there is
1586          * an error and we want to rollback the PC
1587          */
1588         curr_pc = vcpu->arch.pc;
1589         er = update_pc(vcpu, cause);
1590         if (er == EMULATE_FAIL)
1591                 return er;
1592
1593         base = (inst >> 21) & 0x1f;
1594         op_inst = (inst >> 16) & 0x1f;
1595         offset = (int16_t)inst;
1596         cache = (inst >> 16) & 0x3;
1597         op = (inst >> 18) & 0x7;
1598
1599         va = arch->gprs[base] + offset;
1600
1601         kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1602                   cache, op, base, arch->gprs[base], offset);
1603
1604         /*
1605          * Treat INDEX_INV as a nop, basically issued by Linux on startup to
1606          * invalidate the caches entirely by stepping through all the
1607          * ways/indexes
1608          */
1609         if (op == MIPS_CACHE_OP_INDEX_INV) {
1610                 kvm_debug("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1611                           vcpu->arch.pc, vcpu->arch.gprs[31], cache, op, base,
1612                           arch->gprs[base], offset);
1613
1614                 if (cache == MIPS_CACHE_DCACHE)
1615                         r4k_blast_dcache();
1616                 else if (cache == MIPS_CACHE_ICACHE)
1617                         r4k_blast_icache();
1618                 else {
1619                         kvm_err("%s: unsupported CACHE INDEX operation\n",
1620                                 __func__);
1621                         return EMULATE_FAIL;
1622                 }
1623
1624 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1625                 kvm_mips_trans_cache_index(inst, opc, vcpu);
1626 #endif
1627                 goto done;
1628         }
1629
1630         preempt_disable();
1631         if (KVM_GUEST_KSEGX(va) == KVM_GUEST_KSEG0) {
1632                 if (kvm_mips_host_tlb_lookup(vcpu, va) < 0)
1633                         kvm_mips_handle_kseg0_tlb_fault(va, vcpu);
1634         } else if ((KVM_GUEST_KSEGX(va) < KVM_GUEST_KSEG0) ||
1635                    KVM_GUEST_KSEGX(va) == KVM_GUEST_KSEG23) {
1636                 int index;
1637
1638                 /* If an entry already exists then skip */
1639                 if (kvm_mips_host_tlb_lookup(vcpu, va) >= 0)
1640                         goto skip_fault;
1641
1642                 /*
1643                  * If address not in the guest TLB, then give the guest a fault,
1644                  * the resulting handler will do the right thing
1645                  */
1646                 index = kvm_mips_guest_tlb_lookup(vcpu, (va & VPN2_MASK) |
1647                                                   (kvm_read_c0_guest_entryhi
1648                                                    (cop0) & ASID_MASK));
1649
1650                 if (index < 0) {
1651                         vcpu->arch.host_cp0_entryhi = (va & VPN2_MASK);
1652                         vcpu->arch.host_cp0_badvaddr = va;
1653                         er = kvm_mips_emulate_tlbmiss_ld(cause, NULL, run,
1654                                                          vcpu);
1655                         preempt_enable();
1656                         goto dont_update_pc;
1657                 } else {
1658                         struct kvm_mips_tlb *tlb = &vcpu->arch.guest_tlb[index];
1659                         /*
1660                          * Check if the entry is valid, if not then setup a TLB
1661                          * invalid exception to the guest
1662                          */
1663                         if (!TLB_IS_VALID(*tlb, va)) {
1664                                 er = kvm_mips_emulate_tlbinv_ld(cause, NULL,
1665                                                                 run, vcpu);
1666                                 preempt_enable();
1667                                 goto dont_update_pc;
1668                         } else {
1669                                 /*
1670                                  * We fault an entry from the guest tlb to the
1671                                  * shadow host TLB
1672                                  */
1673                                 kvm_mips_handle_mapped_seg_tlb_fault(vcpu, tlb,
1674                                                                      NULL,
1675                                                                      NULL);
1676                         }
1677                 }
1678         } else {
1679                 kvm_err("INVALID CACHE INDEX/ADDRESS (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1680                         cache, op, base, arch->gprs[base], offset);
1681                 er = EMULATE_FAIL;
1682                 preempt_enable();
1683                 goto dont_update_pc;
1684
1685         }
1686
1687 skip_fault:
1688         /* XXXKYMA: Only a subset of cache ops are supported, used by Linux */
1689         if (cache == MIPS_CACHE_DCACHE
1690             && (op == MIPS_CACHE_OP_FILL_WB_INV
1691                 || op == MIPS_CACHE_OP_HIT_INV)) {
1692                 flush_dcache_line(va);
1693
1694 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1695                 /*
1696                  * Replace the CACHE instruction, with a SYNCI, not the same,
1697                  * but avoids a trap
1698                  */
1699                 kvm_mips_trans_cache_va(inst, opc, vcpu);
1700 #endif
1701         } else if (op == MIPS_CACHE_OP_HIT_INV && cache == MIPS_CACHE_ICACHE) {
1702                 flush_dcache_line(va);
1703                 flush_icache_line(va);
1704
1705 #ifdef CONFIG_KVM_MIPS_DYN_TRANS
1706                 /* Replace the CACHE instruction, with a SYNCI */
1707                 kvm_mips_trans_cache_va(inst, opc, vcpu);
1708 #endif
1709         } else {
1710                 kvm_err("NO-OP CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1711                         cache, op, base, arch->gprs[base], offset);
1712                 er = EMULATE_FAIL;
1713                 preempt_enable();
1714                 goto dont_update_pc;
1715         }
1716
1717         preempt_enable();
1718
1719 dont_update_pc:
1720         /* Rollback PC */
1721         vcpu->arch.pc = curr_pc;
1722 done:
1723         return er;
1724 }
1725
1726 enum emulation_result kvm_mips_emulate_inst(unsigned long cause, uint32_t *opc,
1727                                             struct kvm_run *run,
1728                                             struct kvm_vcpu *vcpu)
1729 {
1730         enum emulation_result er = EMULATE_DONE;
1731         uint32_t inst;
1732
1733         /* Fetch the instruction. */
1734         if (cause & CAUSEF_BD)
1735                 opc += 1;
1736
1737         inst = kvm_get_inst(opc, vcpu);
1738
1739         switch (((union mips_instruction)inst).r_format.opcode) {
1740         case cop0_op:
1741                 er = kvm_mips_emulate_CP0(inst, opc, cause, run, vcpu);
1742                 break;
1743         case sb_op:
1744         case sh_op:
1745         case sw_op:
1746                 er = kvm_mips_emulate_store(inst, cause, run, vcpu);
1747                 break;
1748         case lb_op:
1749         case lbu_op:
1750         case lhu_op:
1751         case lh_op:
1752         case lw_op:
1753                 er = kvm_mips_emulate_load(inst, cause, run, vcpu);
1754                 break;
1755
1756         case cache_op:
1757                 ++vcpu->stat.cache_exits;
1758                 trace_kvm_exit(vcpu, CACHE_EXITS);
1759                 er = kvm_mips_emulate_cache(inst, opc, cause, run, vcpu);
1760                 break;
1761
1762         default:
1763                 kvm_err("Instruction emulation not supported (%p/%#x)\n", opc,
1764                         inst);
1765                 kvm_arch_vcpu_dump_regs(vcpu);
1766                 er = EMULATE_FAIL;
1767                 break;
1768         }
1769
1770         return er;
1771 }
1772
1773 enum emulation_result kvm_mips_emulate_syscall(unsigned long cause,
1774                                                uint32_t *opc,
1775                                                struct kvm_run *run,
1776                                                struct kvm_vcpu *vcpu)
1777 {
1778         struct mips_coproc *cop0 = vcpu->arch.cop0;
1779         struct kvm_vcpu_arch *arch = &vcpu->arch;
1780         enum emulation_result er = EMULATE_DONE;
1781
1782         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
1783                 /* save old pc */
1784                 kvm_write_c0_guest_epc(cop0, arch->pc);
1785                 kvm_set_c0_guest_status(cop0, ST0_EXL);
1786
1787                 if (cause & CAUSEF_BD)
1788                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
1789                 else
1790                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
1791
1792                 kvm_debug("Delivering SYSCALL @ pc %#lx\n", arch->pc);
1793
1794                 kvm_change_c0_guest_cause(cop0, (0xff),
1795                                           (T_SYSCALL << CAUSEB_EXCCODE));
1796
1797                 /* Set PC to the exception entry point */
1798                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
1799
1800         } else {
1801                 kvm_err("Trying to deliver SYSCALL when EXL is already set\n");
1802                 er = EMULATE_FAIL;
1803         }
1804
1805         return er;
1806 }
1807
1808 enum emulation_result kvm_mips_emulate_tlbmiss_ld(unsigned long cause,
1809                                                   uint32_t *opc,
1810                                                   struct kvm_run *run,
1811                                                   struct kvm_vcpu *vcpu)
1812 {
1813         struct mips_coproc *cop0 = vcpu->arch.cop0;
1814         struct kvm_vcpu_arch *arch = &vcpu->arch;
1815         unsigned long entryhi = (vcpu->arch.  host_cp0_badvaddr & VPN2_MASK) |
1816                                 (kvm_read_c0_guest_entryhi(cop0) & ASID_MASK);
1817
1818         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
1819                 /* save old pc */
1820                 kvm_write_c0_guest_epc(cop0, arch->pc);
1821                 kvm_set_c0_guest_status(cop0, ST0_EXL);
1822
1823                 if (cause & CAUSEF_BD)
1824                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
1825                 else
1826                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
1827
1828                 kvm_debug("[EXL == 0] delivering TLB MISS @ pc %#lx\n",
1829                           arch->pc);
1830
1831                 /* set pc to the exception entry point */
1832                 arch->pc = KVM_GUEST_KSEG0 + 0x0;
1833
1834         } else {
1835                 kvm_debug("[EXL == 1] delivering TLB MISS @ pc %#lx\n",
1836                           arch->pc);
1837
1838                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
1839         }
1840
1841         kvm_change_c0_guest_cause(cop0, (0xff),
1842                                   (T_TLB_LD_MISS << CAUSEB_EXCCODE));
1843
1844         /* setup badvaddr, context and entryhi registers for the guest */
1845         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
1846         /* XXXKYMA: is the context register used by linux??? */
1847         kvm_write_c0_guest_entryhi(cop0, entryhi);
1848         /* Blow away the shadow host TLBs */
1849         kvm_mips_flush_host_tlb(1);
1850
1851         return EMULATE_DONE;
1852 }
1853
1854 enum emulation_result kvm_mips_emulate_tlbinv_ld(unsigned long cause,
1855                                                  uint32_t *opc,
1856                                                  struct kvm_run *run,
1857                                                  struct kvm_vcpu *vcpu)
1858 {
1859         struct mips_coproc *cop0 = vcpu->arch.cop0;
1860         struct kvm_vcpu_arch *arch = &vcpu->arch;
1861         unsigned long entryhi =
1862                 (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
1863                 (kvm_read_c0_guest_entryhi(cop0) & ASID_MASK);
1864
1865         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
1866                 /* save old pc */
1867                 kvm_write_c0_guest_epc(cop0, arch->pc);
1868                 kvm_set_c0_guest_status(cop0, ST0_EXL);
1869
1870                 if (cause & CAUSEF_BD)
1871                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
1872                 else
1873                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
1874
1875                 kvm_debug("[EXL == 0] delivering TLB INV @ pc %#lx\n",
1876                           arch->pc);
1877
1878                 /* set pc to the exception entry point */
1879                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
1880
1881         } else {
1882                 kvm_debug("[EXL == 1] delivering TLB MISS @ pc %#lx\n",
1883                           arch->pc);
1884                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
1885         }
1886
1887         kvm_change_c0_guest_cause(cop0, (0xff),
1888                                   (T_TLB_LD_MISS << CAUSEB_EXCCODE));
1889
1890         /* setup badvaddr, context and entryhi registers for the guest */
1891         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
1892         /* XXXKYMA: is the context register used by linux??? */
1893         kvm_write_c0_guest_entryhi(cop0, entryhi);
1894         /* Blow away the shadow host TLBs */
1895         kvm_mips_flush_host_tlb(1);
1896
1897         return EMULATE_DONE;
1898 }
1899
1900 enum emulation_result kvm_mips_emulate_tlbmiss_st(unsigned long cause,
1901                                                   uint32_t *opc,
1902                                                   struct kvm_run *run,
1903                                                   struct kvm_vcpu *vcpu)
1904 {
1905         struct mips_coproc *cop0 = vcpu->arch.cop0;
1906         struct kvm_vcpu_arch *arch = &vcpu->arch;
1907         unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
1908                                 (kvm_read_c0_guest_entryhi(cop0) & ASID_MASK);
1909
1910         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
1911                 /* save old pc */
1912                 kvm_write_c0_guest_epc(cop0, arch->pc);
1913                 kvm_set_c0_guest_status(cop0, ST0_EXL);
1914
1915                 if (cause & CAUSEF_BD)
1916                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
1917                 else
1918                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
1919
1920                 kvm_debug("[EXL == 0] Delivering TLB MISS @ pc %#lx\n",
1921                           arch->pc);
1922
1923                 /* Set PC to the exception entry point */
1924                 arch->pc = KVM_GUEST_KSEG0 + 0x0;
1925         } else {
1926                 kvm_debug("[EXL == 1] Delivering TLB MISS @ pc %#lx\n",
1927                           arch->pc);
1928                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
1929         }
1930
1931         kvm_change_c0_guest_cause(cop0, (0xff),
1932                                   (T_TLB_ST_MISS << CAUSEB_EXCCODE));
1933
1934         /* setup badvaddr, context and entryhi registers for the guest */
1935         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
1936         /* XXXKYMA: is the context register used by linux??? */
1937         kvm_write_c0_guest_entryhi(cop0, entryhi);
1938         /* Blow away the shadow host TLBs */
1939         kvm_mips_flush_host_tlb(1);
1940
1941         return EMULATE_DONE;
1942 }
1943
1944 enum emulation_result kvm_mips_emulate_tlbinv_st(unsigned long cause,
1945                                                  uint32_t *opc,
1946                                                  struct kvm_run *run,
1947                                                  struct kvm_vcpu *vcpu)
1948 {
1949         struct mips_coproc *cop0 = vcpu->arch.cop0;
1950         struct kvm_vcpu_arch *arch = &vcpu->arch;
1951         unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
1952                 (kvm_read_c0_guest_entryhi(cop0) & ASID_MASK);
1953
1954         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
1955                 /* save old pc */
1956                 kvm_write_c0_guest_epc(cop0, arch->pc);
1957                 kvm_set_c0_guest_status(cop0, ST0_EXL);
1958
1959                 if (cause & CAUSEF_BD)
1960                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
1961                 else
1962                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
1963
1964                 kvm_debug("[EXL == 0] Delivering TLB MISS @ pc %#lx\n",
1965                           arch->pc);
1966
1967                 /* Set PC to the exception entry point */
1968                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
1969         } else {
1970                 kvm_debug("[EXL == 1] Delivering TLB MISS @ pc %#lx\n",
1971                           arch->pc);
1972                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
1973         }
1974
1975         kvm_change_c0_guest_cause(cop0, (0xff),
1976                                   (T_TLB_ST_MISS << CAUSEB_EXCCODE));
1977
1978         /* setup badvaddr, context and entryhi registers for the guest */
1979         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
1980         /* XXXKYMA: is the context register used by linux??? */
1981         kvm_write_c0_guest_entryhi(cop0, entryhi);
1982         /* Blow away the shadow host TLBs */
1983         kvm_mips_flush_host_tlb(1);
1984
1985         return EMULATE_DONE;
1986 }
1987
1988 /* TLBMOD: store into address matching TLB with Dirty bit off */
1989 enum emulation_result kvm_mips_handle_tlbmod(unsigned long cause, uint32_t *opc,
1990                                              struct kvm_run *run,
1991                                              struct kvm_vcpu *vcpu)
1992 {
1993         enum emulation_result er = EMULATE_DONE;
1994 #ifdef DEBUG
1995         struct mips_coproc *cop0 = vcpu->arch.cop0;
1996         unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
1997                                 (kvm_read_c0_guest_entryhi(cop0) & ASID_MASK);
1998         int index;
1999
2000         /* If address not in the guest TLB, then we are in trouble */
2001         index = kvm_mips_guest_tlb_lookup(vcpu, entryhi);
2002         if (index < 0) {
2003                 /* XXXKYMA Invalidate and retry */
2004                 kvm_mips_host_tlb_inv(vcpu, vcpu->arch.host_cp0_badvaddr);
2005                 kvm_err("%s: host got TLBMOD for %#lx but entry not present in Guest TLB\n",
2006                      __func__, entryhi);
2007                 kvm_mips_dump_guest_tlbs(vcpu);
2008                 kvm_mips_dump_host_tlbs();
2009                 return EMULATE_FAIL;
2010         }
2011 #endif
2012
2013         er = kvm_mips_emulate_tlbmod(cause, opc, run, vcpu);
2014         return er;
2015 }
2016
2017 enum emulation_result kvm_mips_emulate_tlbmod(unsigned long cause,
2018                                               uint32_t *opc,
2019                                               struct kvm_run *run,
2020                                               struct kvm_vcpu *vcpu)
2021 {
2022         struct mips_coproc *cop0 = vcpu->arch.cop0;
2023         unsigned long entryhi = (vcpu->arch.host_cp0_badvaddr & VPN2_MASK) |
2024                                 (kvm_read_c0_guest_entryhi(cop0) & ASID_MASK);
2025         struct kvm_vcpu_arch *arch = &vcpu->arch;
2026
2027         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2028                 /* save old pc */
2029                 kvm_write_c0_guest_epc(cop0, arch->pc);
2030                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2031
2032                 if (cause & CAUSEF_BD)
2033                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2034                 else
2035                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2036
2037                 kvm_debug("[EXL == 0] Delivering TLB MOD @ pc %#lx\n",
2038                           arch->pc);
2039
2040                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2041         } else {
2042                 kvm_debug("[EXL == 1] Delivering TLB MOD @ pc %#lx\n",
2043                           arch->pc);
2044                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2045         }
2046
2047         kvm_change_c0_guest_cause(cop0, (0xff), (T_TLB_MOD << CAUSEB_EXCCODE));
2048
2049         /* setup badvaddr, context and entryhi registers for the guest */
2050         kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2051         /* XXXKYMA: is the context register used by linux??? */
2052         kvm_write_c0_guest_entryhi(cop0, entryhi);
2053         /* Blow away the shadow host TLBs */
2054         kvm_mips_flush_host_tlb(1);
2055
2056         return EMULATE_DONE;
2057 }
2058
2059 enum emulation_result kvm_mips_emulate_fpu_exc(unsigned long cause,
2060                                                uint32_t *opc,
2061                                                struct kvm_run *run,
2062                                                struct kvm_vcpu *vcpu)
2063 {
2064         struct mips_coproc *cop0 = vcpu->arch.cop0;
2065         struct kvm_vcpu_arch *arch = &vcpu->arch;
2066
2067         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2068                 /* save old pc */
2069                 kvm_write_c0_guest_epc(cop0, arch->pc);
2070                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2071
2072                 if (cause & CAUSEF_BD)
2073                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2074                 else
2075                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2076
2077         }
2078
2079         arch->pc = KVM_GUEST_KSEG0 + 0x180;
2080
2081         kvm_change_c0_guest_cause(cop0, (0xff),
2082                                   (T_COP_UNUSABLE << CAUSEB_EXCCODE));
2083         kvm_change_c0_guest_cause(cop0, (CAUSEF_CE), (0x1 << CAUSEB_CE));
2084
2085         return EMULATE_DONE;
2086 }
2087
2088 enum emulation_result kvm_mips_emulate_ri_exc(unsigned long cause,
2089                                               uint32_t *opc,
2090                                               struct kvm_run *run,
2091                                               struct kvm_vcpu *vcpu)
2092 {
2093         struct mips_coproc *cop0 = vcpu->arch.cop0;
2094         struct kvm_vcpu_arch *arch = &vcpu->arch;
2095         enum emulation_result er = EMULATE_DONE;
2096
2097         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2098                 /* save old pc */
2099                 kvm_write_c0_guest_epc(cop0, arch->pc);
2100                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2101
2102                 if (cause & CAUSEF_BD)
2103                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2104                 else
2105                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2106
2107                 kvm_debug("Delivering RI @ pc %#lx\n", arch->pc);
2108
2109                 kvm_change_c0_guest_cause(cop0, (0xff),
2110                                           (T_RES_INST << CAUSEB_EXCCODE));
2111
2112                 /* Set PC to the exception entry point */
2113                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2114
2115         } else {
2116                 kvm_err("Trying to deliver RI when EXL is already set\n");
2117                 er = EMULATE_FAIL;
2118         }
2119
2120         return er;
2121 }
2122
2123 enum emulation_result kvm_mips_emulate_bp_exc(unsigned long cause,
2124                                               uint32_t *opc,
2125                                               struct kvm_run *run,
2126                                               struct kvm_vcpu *vcpu)
2127 {
2128         struct mips_coproc *cop0 = vcpu->arch.cop0;
2129         struct kvm_vcpu_arch *arch = &vcpu->arch;
2130         enum emulation_result er = EMULATE_DONE;
2131
2132         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2133                 /* save old pc */
2134                 kvm_write_c0_guest_epc(cop0, arch->pc);
2135                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2136
2137                 if (cause & CAUSEF_BD)
2138                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2139                 else
2140                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2141
2142                 kvm_debug("Delivering BP @ pc %#lx\n", arch->pc);
2143
2144                 kvm_change_c0_guest_cause(cop0, (0xff),
2145                                           (T_BREAK << CAUSEB_EXCCODE));
2146
2147                 /* Set PC to the exception entry point */
2148                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2149
2150         } else {
2151                 kvm_err("Trying to deliver BP when EXL is already set\n");
2152                 er = EMULATE_FAIL;
2153         }
2154
2155         return er;
2156 }
2157
2158 enum emulation_result kvm_mips_emulate_trap_exc(unsigned long cause,
2159                                                 uint32_t *opc,
2160                                                 struct kvm_run *run,
2161                                                 struct kvm_vcpu *vcpu)
2162 {
2163         struct mips_coproc *cop0 = vcpu->arch.cop0;
2164         struct kvm_vcpu_arch *arch = &vcpu->arch;
2165         enum emulation_result er = EMULATE_DONE;
2166
2167         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2168                 /* save old pc */
2169                 kvm_write_c0_guest_epc(cop0, arch->pc);
2170                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2171
2172                 if (cause & CAUSEF_BD)
2173                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2174                 else
2175                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2176
2177                 kvm_debug("Delivering TRAP @ pc %#lx\n", arch->pc);
2178
2179                 kvm_change_c0_guest_cause(cop0, (0xff),
2180                                           (T_TRAP << CAUSEB_EXCCODE));
2181
2182                 /* Set PC to the exception entry point */
2183                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2184
2185         } else {
2186                 kvm_err("Trying to deliver TRAP when EXL is already set\n");
2187                 er = EMULATE_FAIL;
2188         }
2189
2190         return er;
2191 }
2192
2193 enum emulation_result kvm_mips_emulate_msafpe_exc(unsigned long cause,
2194                                                   uint32_t *opc,
2195                                                   struct kvm_run *run,
2196                                                   struct kvm_vcpu *vcpu)
2197 {
2198         struct mips_coproc *cop0 = vcpu->arch.cop0;
2199         struct kvm_vcpu_arch *arch = &vcpu->arch;
2200         enum emulation_result er = EMULATE_DONE;
2201
2202         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2203                 /* save old pc */
2204                 kvm_write_c0_guest_epc(cop0, arch->pc);
2205                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2206
2207                 if (cause & CAUSEF_BD)
2208                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2209                 else
2210                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2211
2212                 kvm_debug("Delivering MSAFPE @ pc %#lx\n", arch->pc);
2213
2214                 kvm_change_c0_guest_cause(cop0, (0xff),
2215                                           (T_MSAFPE << CAUSEB_EXCCODE));
2216
2217                 /* Set PC to the exception entry point */
2218                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2219
2220         } else {
2221                 kvm_err("Trying to deliver MSAFPE when EXL is already set\n");
2222                 er = EMULATE_FAIL;
2223         }
2224
2225         return er;
2226 }
2227
2228 enum emulation_result kvm_mips_emulate_fpe_exc(unsigned long cause,
2229                                                uint32_t *opc,
2230                                                struct kvm_run *run,
2231                                                struct kvm_vcpu *vcpu)
2232 {
2233         struct mips_coproc *cop0 = vcpu->arch.cop0;
2234         struct kvm_vcpu_arch *arch = &vcpu->arch;
2235         enum emulation_result er = EMULATE_DONE;
2236
2237         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2238                 /* save old pc */
2239                 kvm_write_c0_guest_epc(cop0, arch->pc);
2240                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2241
2242                 if (cause & CAUSEF_BD)
2243                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2244                 else
2245                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2246
2247                 kvm_debug("Delivering FPE @ pc %#lx\n", arch->pc);
2248
2249                 kvm_change_c0_guest_cause(cop0, (0xff),
2250                                           (T_FPE << CAUSEB_EXCCODE));
2251
2252                 /* Set PC to the exception entry point */
2253                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2254
2255         } else {
2256                 kvm_err("Trying to deliver FPE when EXL is already set\n");
2257                 er = EMULATE_FAIL;
2258         }
2259
2260         return er;
2261 }
2262
2263 enum emulation_result kvm_mips_emulate_msadis_exc(unsigned long cause,
2264                                                   uint32_t *opc,
2265                                                   struct kvm_run *run,
2266                                                   struct kvm_vcpu *vcpu)
2267 {
2268         struct mips_coproc *cop0 = vcpu->arch.cop0;
2269         struct kvm_vcpu_arch *arch = &vcpu->arch;
2270         enum emulation_result er = EMULATE_DONE;
2271
2272         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2273                 /* save old pc */
2274                 kvm_write_c0_guest_epc(cop0, arch->pc);
2275                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2276
2277                 if (cause & CAUSEF_BD)
2278                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2279                 else
2280                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2281
2282                 kvm_debug("Delivering MSADIS @ pc %#lx\n", arch->pc);
2283
2284                 kvm_change_c0_guest_cause(cop0, (0xff),
2285                                           (T_MSADIS << CAUSEB_EXCCODE));
2286
2287                 /* Set PC to the exception entry point */
2288                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2289
2290         } else {
2291                 kvm_err("Trying to deliver MSADIS when EXL is already set\n");
2292                 er = EMULATE_FAIL;
2293         }
2294
2295         return er;
2296 }
2297
2298 /* ll/sc, rdhwr, sync emulation */
2299
2300 #define OPCODE 0xfc000000
2301 #define BASE   0x03e00000
2302 #define RT     0x001f0000
2303 #define OFFSET 0x0000ffff
2304 #define LL     0xc0000000
2305 #define SC     0xe0000000
2306 #define SPEC0  0x00000000
2307 #define SPEC3  0x7c000000
2308 #define RD     0x0000f800
2309 #define FUNC   0x0000003f
2310 #define SYNC   0x0000000f
2311 #define RDHWR  0x0000003b
2312
2313 enum emulation_result kvm_mips_handle_ri(unsigned long cause, uint32_t *opc,
2314                                          struct kvm_run *run,
2315                                          struct kvm_vcpu *vcpu)
2316 {
2317         struct mips_coproc *cop0 = vcpu->arch.cop0;
2318         struct kvm_vcpu_arch *arch = &vcpu->arch;
2319         enum emulation_result er = EMULATE_DONE;
2320         unsigned long curr_pc;
2321         uint32_t inst;
2322
2323         /*
2324          * Update PC and hold onto current PC in case there is
2325          * an error and we want to rollback the PC
2326          */
2327         curr_pc = vcpu->arch.pc;
2328         er = update_pc(vcpu, cause);
2329         if (er == EMULATE_FAIL)
2330                 return er;
2331
2332         /* Fetch the instruction. */
2333         if (cause & CAUSEF_BD)
2334                 opc += 1;
2335
2336         inst = kvm_get_inst(opc, vcpu);
2337
2338         if (inst == KVM_INVALID_INST) {
2339                 kvm_err("%s: Cannot get inst @ %p\n", __func__, opc);
2340                 return EMULATE_FAIL;
2341         }
2342
2343         if ((inst & OPCODE) == SPEC3 && (inst & FUNC) == RDHWR) {
2344                 int usermode = !KVM_GUEST_KERNEL_MODE(vcpu);
2345                 int rd = (inst & RD) >> 11;
2346                 int rt = (inst & RT) >> 16;
2347                 /* If usermode, check RDHWR rd is allowed by guest HWREna */
2348                 if (usermode && !(kvm_read_c0_guest_hwrena(cop0) & BIT(rd))) {
2349                         kvm_debug("RDHWR %#x disallowed by HWREna @ %p\n",
2350                                   rd, opc);
2351                         goto emulate_ri;
2352                 }
2353                 switch (rd) {
2354                 case 0: /* CPU number */
2355                         arch->gprs[rt] = 0;
2356                         break;
2357                 case 1: /* SYNCI length */
2358                         arch->gprs[rt] = min(current_cpu_data.dcache.linesz,
2359                                              current_cpu_data.icache.linesz);
2360                         break;
2361                 case 2: /* Read count register */
2362                         arch->gprs[rt] = kvm_mips_read_count(vcpu);
2363                         break;
2364                 case 3: /* Count register resolution */
2365                         switch (current_cpu_data.cputype) {
2366                         case CPU_20KC:
2367                         case CPU_25KF:
2368                                 arch->gprs[rt] = 1;
2369                                 break;
2370                         default:
2371                                 arch->gprs[rt] = 2;
2372                         }
2373                         break;
2374                 case 29:
2375                         arch->gprs[rt] = kvm_read_c0_guest_userlocal(cop0);
2376                         break;
2377
2378                 default:
2379                         kvm_debug("RDHWR %#x not supported @ %p\n", rd, opc);
2380                         goto emulate_ri;
2381                 }
2382         } else {
2383                 kvm_debug("Emulate RI not supported @ %p: %#x\n", opc, inst);
2384                 goto emulate_ri;
2385         }
2386
2387         return EMULATE_DONE;
2388
2389 emulate_ri:
2390         /*
2391          * Rollback PC (if in branch delay slot then the PC already points to
2392          * branch target), and pass the RI exception to the guest OS.
2393          */
2394         vcpu->arch.pc = curr_pc;
2395         return kvm_mips_emulate_ri_exc(cause, opc, run, vcpu);
2396 }
2397
2398 enum emulation_result kvm_mips_complete_mmio_load(struct kvm_vcpu *vcpu,
2399                                                   struct kvm_run *run)
2400 {
2401         unsigned long *gpr = &vcpu->arch.gprs[vcpu->arch.io_gpr];
2402         enum emulation_result er = EMULATE_DONE;
2403
2404         if (run->mmio.len > sizeof(*gpr)) {
2405                 kvm_err("Bad MMIO length: %d", run->mmio.len);
2406                 er = EMULATE_FAIL;
2407                 goto done;
2408         }
2409
2410         er = update_pc(vcpu, vcpu->arch.pending_load_cause);
2411         if (er == EMULATE_FAIL)
2412                 return er;
2413
2414         switch (run->mmio.len) {
2415         case 4:
2416                 *gpr = *(int32_t *) run->mmio.data;
2417                 break;
2418
2419         case 2:
2420                 if (vcpu->mmio_needed == 2)
2421                         *gpr = *(int16_t *) run->mmio.data;
2422                 else
2423                         *gpr = *(uint16_t *)run->mmio.data;
2424
2425                 break;
2426         case 1:
2427                 if (vcpu->mmio_needed == 2)
2428                         *gpr = *(int8_t *) run->mmio.data;
2429                 else
2430                         *gpr = *(u8 *) run->mmio.data;
2431                 break;
2432         }
2433
2434         if (vcpu->arch.pending_load_cause & CAUSEF_BD)
2435                 kvm_debug("[%#lx] Completing %d byte BD Load to gpr %d (0x%08lx) type %d\n",
2436                           vcpu->arch.pc, run->mmio.len, vcpu->arch.io_gpr, *gpr,
2437                           vcpu->mmio_needed);
2438
2439 done:
2440         return er;
2441 }
2442
2443 static enum emulation_result kvm_mips_emulate_exc(unsigned long cause,
2444                                                   uint32_t *opc,
2445                                                   struct kvm_run *run,
2446                                                   struct kvm_vcpu *vcpu)
2447 {
2448         uint32_t exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2449         struct mips_coproc *cop0 = vcpu->arch.cop0;
2450         struct kvm_vcpu_arch *arch = &vcpu->arch;
2451         enum emulation_result er = EMULATE_DONE;
2452
2453         if ((kvm_read_c0_guest_status(cop0) & ST0_EXL) == 0) {
2454                 /* save old pc */
2455                 kvm_write_c0_guest_epc(cop0, arch->pc);
2456                 kvm_set_c0_guest_status(cop0, ST0_EXL);
2457
2458                 if (cause & CAUSEF_BD)
2459                         kvm_set_c0_guest_cause(cop0, CAUSEF_BD);
2460                 else
2461                         kvm_clear_c0_guest_cause(cop0, CAUSEF_BD);
2462
2463                 kvm_change_c0_guest_cause(cop0, (0xff),
2464                                           (exccode << CAUSEB_EXCCODE));
2465
2466                 /* Set PC to the exception entry point */
2467                 arch->pc = KVM_GUEST_KSEG0 + 0x180;
2468                 kvm_write_c0_guest_badvaddr(cop0, vcpu->arch.host_cp0_badvaddr);
2469
2470                 kvm_debug("Delivering EXC %d @ pc %#lx, badVaddr: %#lx\n",
2471                           exccode, kvm_read_c0_guest_epc(cop0),
2472                           kvm_read_c0_guest_badvaddr(cop0));
2473         } else {
2474                 kvm_err("Trying to deliver EXC when EXL is already set\n");
2475                 er = EMULATE_FAIL;
2476         }
2477
2478         return er;
2479 }
2480
2481 enum emulation_result kvm_mips_check_privilege(unsigned long cause,
2482                                                uint32_t *opc,
2483                                                struct kvm_run *run,
2484                                                struct kvm_vcpu *vcpu)
2485 {
2486         enum emulation_result er = EMULATE_DONE;
2487         uint32_t exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2488         unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
2489
2490         int usermode = !KVM_GUEST_KERNEL_MODE(vcpu);
2491
2492         if (usermode) {
2493                 switch (exccode) {
2494                 case T_INT:
2495                 case T_SYSCALL:
2496                 case T_BREAK:
2497                 case T_RES_INST:
2498                 case T_TRAP:
2499                 case T_MSAFPE:
2500                 case T_FPE:
2501                 case T_MSADIS:
2502                         break;
2503
2504                 case T_COP_UNUSABLE:
2505                         if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 0)
2506                                 er = EMULATE_PRIV_FAIL;
2507                         break;
2508
2509                 case T_TLB_MOD:
2510                         break;
2511
2512                 case T_TLB_LD_MISS:
2513                         /*
2514                          * We we are accessing Guest kernel space, then send an
2515                          * address error exception to the guest
2516                          */
2517                         if (badvaddr >= (unsigned long) KVM_GUEST_KSEG0) {
2518                                 kvm_debug("%s: LD MISS @ %#lx\n", __func__,
2519                                           badvaddr);
2520                                 cause &= ~0xff;
2521                                 cause |= (T_ADDR_ERR_LD << CAUSEB_EXCCODE);
2522                                 er = EMULATE_PRIV_FAIL;
2523                         }
2524                         break;
2525
2526                 case T_TLB_ST_MISS:
2527                         /*
2528                          * We we are accessing Guest kernel space, then send an
2529                          * address error exception to the guest
2530                          */
2531                         if (badvaddr >= (unsigned long) KVM_GUEST_KSEG0) {
2532                                 kvm_debug("%s: ST MISS @ %#lx\n", __func__,
2533                                           badvaddr);
2534                                 cause &= ~0xff;
2535                                 cause |= (T_ADDR_ERR_ST << CAUSEB_EXCCODE);
2536                                 er = EMULATE_PRIV_FAIL;
2537                         }
2538                         break;
2539
2540                 case T_ADDR_ERR_ST:
2541                         kvm_debug("%s: address error ST @ %#lx\n", __func__,
2542                                   badvaddr);
2543                         if ((badvaddr & PAGE_MASK) == KVM_GUEST_COMMPAGE_ADDR) {
2544                                 cause &= ~0xff;
2545                                 cause |= (T_TLB_ST_MISS << CAUSEB_EXCCODE);
2546                         }
2547                         er = EMULATE_PRIV_FAIL;
2548                         break;
2549                 case T_ADDR_ERR_LD:
2550                         kvm_debug("%s: address error LD @ %#lx\n", __func__,
2551                                   badvaddr);
2552                         if ((badvaddr & PAGE_MASK) == KVM_GUEST_COMMPAGE_ADDR) {
2553                                 cause &= ~0xff;
2554                                 cause |= (T_TLB_LD_MISS << CAUSEB_EXCCODE);
2555                         }
2556                         er = EMULATE_PRIV_FAIL;
2557                         break;
2558                 default:
2559                         er = EMULATE_PRIV_FAIL;
2560                         break;
2561                 }
2562         }
2563
2564         if (er == EMULATE_PRIV_FAIL)
2565                 kvm_mips_emulate_exc(cause, opc, run, vcpu);
2566
2567         return er;
2568 }
2569
2570 /*
2571  * User Address (UA) fault, this could happen if
2572  * (1) TLB entry not present/valid in both Guest and shadow host TLBs, in this
2573  *     case we pass on the fault to the guest kernel and let it handle it.
2574  * (2) TLB entry is present in the Guest TLB but not in the shadow, in this
2575  *     case we inject the TLB from the Guest TLB into the shadow host TLB
2576  */
2577 enum emulation_result kvm_mips_handle_tlbmiss(unsigned long cause,
2578                                               uint32_t *opc,
2579                                               struct kvm_run *run,
2580                                               struct kvm_vcpu *vcpu)
2581 {
2582         enum emulation_result er = EMULATE_DONE;
2583         uint32_t exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
2584         unsigned long va = vcpu->arch.host_cp0_badvaddr;
2585         int index;
2586
2587         kvm_debug("kvm_mips_handle_tlbmiss: badvaddr: %#lx, entryhi: %#lx\n",
2588                   vcpu->arch.host_cp0_badvaddr, vcpu->arch.host_cp0_entryhi);
2589
2590         /*
2591          * KVM would not have got the exception if this entry was valid in the
2592          * shadow host TLB. Check the Guest TLB, if the entry is not there then
2593          * send the guest an exception. The guest exc handler should then inject
2594          * an entry into the guest TLB.
2595          */
2596         index = kvm_mips_guest_tlb_lookup(vcpu,
2597                                           (va & VPN2_MASK) |
2598                                           (kvm_read_c0_guest_entryhi
2599                                            (vcpu->arch.cop0) & ASID_MASK));
2600         if (index < 0) {
2601                 if (exccode == T_TLB_LD_MISS) {
2602                         er = kvm_mips_emulate_tlbmiss_ld(cause, opc, run, vcpu);
2603                 } else if (exccode == T_TLB_ST_MISS) {
2604                         er = kvm_mips_emulate_tlbmiss_st(cause, opc, run, vcpu);
2605                 } else {
2606                         kvm_err("%s: invalid exc code: %d\n", __func__,
2607                                 exccode);
2608                         er = EMULATE_FAIL;
2609                 }
2610         } else {
2611                 struct kvm_mips_tlb *tlb = &vcpu->arch.guest_tlb[index];
2612
2613                 /*
2614                  * Check if the entry is valid, if not then setup a TLB invalid
2615                  * exception to the guest
2616                  */
2617                 if (!TLB_IS_VALID(*tlb, va)) {
2618                         if (exccode == T_TLB_LD_MISS) {
2619                                 er = kvm_mips_emulate_tlbinv_ld(cause, opc, run,
2620                                                                 vcpu);
2621                         } else if (exccode == T_TLB_ST_MISS) {
2622                                 er = kvm_mips_emulate_tlbinv_st(cause, opc, run,
2623                                                                 vcpu);
2624                         } else {
2625                                 kvm_err("%s: invalid exc code: %d\n", __func__,
2626                                         exccode);
2627                                 er = EMULATE_FAIL;
2628                         }
2629                 } else {
2630                         kvm_debug("Injecting hi: %#lx, lo0: %#lx, lo1: %#lx into shadow host TLB\n",
2631                                   tlb->tlb_hi, tlb->tlb_lo0, tlb->tlb_lo1);
2632                         /*
2633                          * OK we have a Guest TLB entry, now inject it into the
2634                          * shadow host TLB
2635                          */
2636                         kvm_mips_handle_mapped_seg_tlb_fault(vcpu, tlb, NULL,
2637                                                              NULL);
2638                 }
2639         }
2640
2641         return er;
2642 }