kvm: Replace vcpu->swait with rcuwait
[linux-2.6-microblaze.git] / arch / mips / kvm / mips.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: MIPS specific KVM APIs
7  *
8  * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
9  * Authors: Sanjay Lal <sanjayl@kymasys.com>
10  */
11
12 #include <linux/bitops.h>
13 #include <linux/errno.h>
14 #include <linux/err.h>
15 #include <linux/kdebug.h>
16 #include <linux/module.h>
17 #include <linux/uaccess.h>
18 #include <linux/vmalloc.h>
19 #include <linux/sched/signal.h>
20 #include <linux/fs.h>
21 #include <linux/memblock.h>
22
23 #include <asm/fpu.h>
24 #include <asm/page.h>
25 #include <asm/cacheflush.h>
26 #include <asm/mmu_context.h>
27 #include <asm/pgalloc.h>
28 #include <asm/pgtable.h>
29
30 #include <linux/kvm_host.h>
31
32 #include "interrupt.h"
33 #include "commpage.h"
34
35 #define CREATE_TRACE_POINTS
36 #include "trace.h"
37
38 #ifndef VECTORSPACING
39 #define VECTORSPACING 0x100     /* for EI/VI mode */
40 #endif
41
42 struct kvm_stats_debugfs_item debugfs_entries[] = {
43         VCPU_STAT("wait", wait_exits),
44         VCPU_STAT("cache", cache_exits),
45         VCPU_STAT("signal", signal_exits),
46         VCPU_STAT("interrupt", int_exits),
47         VCPU_STAT("cop_unusable", cop_unusable_exits),
48         VCPU_STAT("tlbmod", tlbmod_exits),
49         VCPU_STAT("tlbmiss_ld", tlbmiss_ld_exits),
50         VCPU_STAT("tlbmiss_st", tlbmiss_st_exits),
51         VCPU_STAT("addrerr_st", addrerr_st_exits),
52         VCPU_STAT("addrerr_ld", addrerr_ld_exits),
53         VCPU_STAT("syscall", syscall_exits),
54         VCPU_STAT("resvd_inst", resvd_inst_exits),
55         VCPU_STAT("break_inst", break_inst_exits),
56         VCPU_STAT("trap_inst", trap_inst_exits),
57         VCPU_STAT("msa_fpe", msa_fpe_exits),
58         VCPU_STAT("fpe", fpe_exits),
59         VCPU_STAT("msa_disabled", msa_disabled_exits),
60         VCPU_STAT("flush_dcache", flush_dcache_exits),
61 #ifdef CONFIG_KVM_MIPS_VZ
62         VCPU_STAT("vz_gpsi", vz_gpsi_exits),
63         VCPU_STAT("vz_gsfc", vz_gsfc_exits),
64         VCPU_STAT("vz_hc", vz_hc_exits),
65         VCPU_STAT("vz_grr", vz_grr_exits),
66         VCPU_STAT("vz_gva", vz_gva_exits),
67         VCPU_STAT("vz_ghfc", vz_ghfc_exits),
68         VCPU_STAT("vz_gpa", vz_gpa_exits),
69         VCPU_STAT("vz_resvd", vz_resvd_exits),
70 #endif
71         VCPU_STAT("halt_successful_poll", halt_successful_poll),
72         VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
73         VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
74         VCPU_STAT("halt_wakeup", halt_wakeup),
75         {NULL}
76 };
77
78 bool kvm_trace_guest_mode_change;
79
80 int kvm_guest_mode_change_trace_reg(void)
81 {
82         kvm_trace_guest_mode_change = 1;
83         return 0;
84 }
85
86 void kvm_guest_mode_change_trace_unreg(void)
87 {
88         kvm_trace_guest_mode_change = 0;
89 }
90
91 /*
92  * XXXKYMA: We are simulatoring a processor that has the WII bit set in
93  * Config7, so we are "runnable" if interrupts are pending
94  */
95 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
96 {
97         return !!(vcpu->arch.pending_exceptions);
98 }
99
100 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
101 {
102         return false;
103 }
104
105 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
106 {
107         return 1;
108 }
109
110 int kvm_arch_hardware_enable(void)
111 {
112         return kvm_mips_callbacks->hardware_enable();
113 }
114
115 void kvm_arch_hardware_disable(void)
116 {
117         kvm_mips_callbacks->hardware_disable();
118 }
119
120 int kvm_arch_hardware_setup(void *opaque)
121 {
122         return 0;
123 }
124
125 int kvm_arch_check_processor_compat(void *opaque)
126 {
127         return 0;
128 }
129
130 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
131 {
132         switch (type) {
133 #ifdef CONFIG_KVM_MIPS_VZ
134         case KVM_VM_MIPS_VZ:
135 #else
136         case KVM_VM_MIPS_TE:
137 #endif
138                 break;
139         default:
140                 /* Unsupported KVM type */
141                 return -EINVAL;
142         };
143
144         /* Allocate page table to map GPA -> RPA */
145         kvm->arch.gpa_mm.pgd = kvm_pgd_alloc();
146         if (!kvm->arch.gpa_mm.pgd)
147                 return -ENOMEM;
148
149         return 0;
150 }
151
152 void kvm_mips_free_vcpus(struct kvm *kvm)
153 {
154         unsigned int i;
155         struct kvm_vcpu *vcpu;
156
157         kvm_for_each_vcpu(i, vcpu, kvm) {
158                 kvm_vcpu_destroy(vcpu);
159         }
160
161         mutex_lock(&kvm->lock);
162
163         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
164                 kvm->vcpus[i] = NULL;
165
166         atomic_set(&kvm->online_vcpus, 0);
167
168         mutex_unlock(&kvm->lock);
169 }
170
171 static void kvm_mips_free_gpa_pt(struct kvm *kvm)
172 {
173         /* It should always be safe to remove after flushing the whole range */
174         WARN_ON(!kvm_mips_flush_gpa_pt(kvm, 0, ~0));
175         pgd_free(NULL, kvm->arch.gpa_mm.pgd);
176 }
177
178 void kvm_arch_destroy_vm(struct kvm *kvm)
179 {
180         kvm_mips_free_vcpus(kvm);
181         kvm_mips_free_gpa_pt(kvm);
182 }
183
184 long kvm_arch_dev_ioctl(struct file *filp, unsigned int ioctl,
185                         unsigned long arg)
186 {
187         return -ENOIOCTLCMD;
188 }
189
190 void kvm_arch_flush_shadow_all(struct kvm *kvm)
191 {
192         /* Flush whole GPA */
193         kvm_mips_flush_gpa_pt(kvm, 0, ~0);
194
195         /* Let implementation do the rest */
196         kvm_mips_callbacks->flush_shadow_all(kvm);
197 }
198
199 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
200                                    struct kvm_memory_slot *slot)
201 {
202         /*
203          * The slot has been made invalid (ready for moving or deletion), so we
204          * need to ensure that it can no longer be accessed by any guest VCPUs.
205          */
206
207         spin_lock(&kvm->mmu_lock);
208         /* Flush slot from GPA */
209         kvm_mips_flush_gpa_pt(kvm, slot->base_gfn,
210                               slot->base_gfn + slot->npages - 1);
211         /* Let implementation do the rest */
212         kvm_mips_callbacks->flush_shadow_memslot(kvm, slot);
213         spin_unlock(&kvm->mmu_lock);
214 }
215
216 int kvm_arch_prepare_memory_region(struct kvm *kvm,
217                                    struct kvm_memory_slot *memslot,
218                                    const struct kvm_userspace_memory_region *mem,
219                                    enum kvm_mr_change change)
220 {
221         return 0;
222 }
223
224 void kvm_arch_commit_memory_region(struct kvm *kvm,
225                                    const struct kvm_userspace_memory_region *mem,
226                                    struct kvm_memory_slot *old,
227                                    const struct kvm_memory_slot *new,
228                                    enum kvm_mr_change change)
229 {
230         int needs_flush;
231
232         kvm_debug("%s: kvm: %p slot: %d, GPA: %llx, size: %llx, QVA: %llx\n",
233                   __func__, kvm, mem->slot, mem->guest_phys_addr,
234                   mem->memory_size, mem->userspace_addr);
235
236         /*
237          * If dirty page logging is enabled, write protect all pages in the slot
238          * ready for dirty logging.
239          *
240          * There is no need to do this in any of the following cases:
241          * CREATE:      No dirty mappings will already exist.
242          * MOVE/DELETE: The old mappings will already have been cleaned up by
243          *              kvm_arch_flush_shadow_memslot()
244          */
245         if (change == KVM_MR_FLAGS_ONLY &&
246             (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
247              new->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
248                 spin_lock(&kvm->mmu_lock);
249                 /* Write protect GPA page table entries */
250                 needs_flush = kvm_mips_mkclean_gpa_pt(kvm, new->base_gfn,
251                                         new->base_gfn + new->npages - 1);
252                 /* Let implementation do the rest */
253                 if (needs_flush)
254                         kvm_mips_callbacks->flush_shadow_memslot(kvm, new);
255                 spin_unlock(&kvm->mmu_lock);
256         }
257 }
258
259 static inline void dump_handler(const char *symbol, void *start, void *end)
260 {
261         u32 *p;
262
263         pr_debug("LEAF(%s)\n", symbol);
264
265         pr_debug("\t.set push\n");
266         pr_debug("\t.set noreorder\n");
267
268         for (p = start; p < (u32 *)end; ++p)
269                 pr_debug("\t.word\t0x%08x\t\t# %p\n", *p, p);
270
271         pr_debug("\t.set\tpop\n");
272
273         pr_debug("\tEND(%s)\n", symbol);
274 }
275
276 /* low level hrtimer wake routine */
277 static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
278 {
279         struct kvm_vcpu *vcpu;
280
281         vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
282
283         kvm_mips_callbacks->queue_timer_int(vcpu);
284
285         vcpu->arch.wait = 0;
286         rcuwait_wake_up(&vcpu->wait);
287
288         return kvm_mips_count_timeout(vcpu);
289 }
290
291 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
292 {
293         return 0;
294 }
295
296 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
297 {
298         int err, size;
299         void *gebase, *p, *handler, *refill_start, *refill_end;
300         int i;
301
302         kvm_debug("kvm @ %p: create cpu %d at %p\n",
303                   vcpu->kvm, vcpu->vcpu_id, vcpu);
304
305         err = kvm_mips_callbacks->vcpu_init(vcpu);
306         if (err)
307                 return err;
308
309         hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
310                      HRTIMER_MODE_REL);
311         vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
312
313         /*
314          * Allocate space for host mode exception handlers that handle
315          * guest mode exits
316          */
317         if (cpu_has_veic || cpu_has_vint)
318                 size = 0x200 + VECTORSPACING * 64;
319         else
320                 size = 0x4000;
321
322         gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
323
324         if (!gebase) {
325                 err = -ENOMEM;
326                 goto out_uninit_vcpu;
327         }
328         kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
329                   ALIGN(size, PAGE_SIZE), gebase);
330
331         /*
332          * Check new ebase actually fits in CP0_EBase. The lack of a write gate
333          * limits us to the low 512MB of physical address space. If the memory
334          * we allocate is out of range, just give up now.
335          */
336         if (!cpu_has_ebase_wg && virt_to_phys(gebase) >= 0x20000000) {
337                 kvm_err("CP0_EBase.WG required for guest exception base %pK\n",
338                         gebase);
339                 err = -ENOMEM;
340                 goto out_free_gebase;
341         }
342
343         /* Save new ebase */
344         vcpu->arch.guest_ebase = gebase;
345
346         /* Build guest exception vectors dynamically in unmapped memory */
347         handler = gebase + 0x2000;
348
349         /* TLB refill (or XTLB refill on 64-bit VZ where KX=1) */
350         refill_start = gebase;
351         if (IS_ENABLED(CONFIG_KVM_MIPS_VZ) && IS_ENABLED(CONFIG_64BIT))
352                 refill_start += 0x080;
353         refill_end = kvm_mips_build_tlb_refill_exception(refill_start, handler);
354
355         /* General Exception Entry point */
356         kvm_mips_build_exception(gebase + 0x180, handler);
357
358         /* For vectored interrupts poke the exception code @ all offsets 0-7 */
359         for (i = 0; i < 8; i++) {
360                 kvm_debug("L1 Vectored handler @ %p\n",
361                           gebase + 0x200 + (i * VECTORSPACING));
362                 kvm_mips_build_exception(gebase + 0x200 + i * VECTORSPACING,
363                                          handler);
364         }
365
366         /* General exit handler */
367         p = handler;
368         p = kvm_mips_build_exit(p);
369
370         /* Guest entry routine */
371         vcpu->arch.vcpu_run = p;
372         p = kvm_mips_build_vcpu_run(p);
373
374         /* Dump the generated code */
375         pr_debug("#include <asm/asm.h>\n");
376         pr_debug("#include <asm/regdef.h>\n");
377         pr_debug("\n");
378         dump_handler("kvm_vcpu_run", vcpu->arch.vcpu_run, p);
379         dump_handler("kvm_tlb_refill", refill_start, refill_end);
380         dump_handler("kvm_gen_exc", gebase + 0x180, gebase + 0x200);
381         dump_handler("kvm_exit", gebase + 0x2000, vcpu->arch.vcpu_run);
382
383         /* Invalidate the icache for these ranges */
384         flush_icache_range((unsigned long)gebase,
385                            (unsigned long)gebase + ALIGN(size, PAGE_SIZE));
386
387         /*
388          * Allocate comm page for guest kernel, a TLB will be reserved for
389          * mapping GVA @ 0xFFFF8000 to this page
390          */
391         vcpu->arch.kseg0_commpage = kzalloc(PAGE_SIZE << 1, GFP_KERNEL);
392
393         if (!vcpu->arch.kseg0_commpage) {
394                 err = -ENOMEM;
395                 goto out_free_gebase;
396         }
397
398         kvm_debug("Allocated COMM page @ %p\n", vcpu->arch.kseg0_commpage);
399         kvm_mips_commpage_init(vcpu);
400
401         /* Init */
402         vcpu->arch.last_sched_cpu = -1;
403         vcpu->arch.last_exec_cpu = -1;
404
405         /* Initial guest state */
406         err = kvm_mips_callbacks->vcpu_setup(vcpu);
407         if (err)
408                 goto out_free_commpage;
409
410         return 0;
411
412 out_free_commpage:
413         kfree(vcpu->arch.kseg0_commpage);
414 out_free_gebase:
415         kfree(gebase);
416 out_uninit_vcpu:
417         kvm_mips_callbacks->vcpu_uninit(vcpu);
418         return err;
419 }
420
421 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
422 {
423         hrtimer_cancel(&vcpu->arch.comparecount_timer);
424
425         kvm_mips_dump_stats(vcpu);
426
427         kvm_mmu_free_memory_caches(vcpu);
428         kfree(vcpu->arch.guest_ebase);
429         kfree(vcpu->arch.kseg0_commpage);
430
431         kvm_mips_callbacks->vcpu_uninit(vcpu);
432 }
433
434 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
435                                         struct kvm_guest_debug *dbg)
436 {
437         return -ENOIOCTLCMD;
438 }
439
440 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
441 {
442         struct kvm_run *run = vcpu->run;
443         int r = -EINTR;
444
445         vcpu_load(vcpu);
446
447         kvm_sigset_activate(vcpu);
448
449         if (vcpu->mmio_needed) {
450                 if (!vcpu->mmio_is_write)
451                         kvm_mips_complete_mmio_load(vcpu, run);
452                 vcpu->mmio_needed = 0;
453         }
454
455         if (run->immediate_exit)
456                 goto out;
457
458         lose_fpu(1);
459
460         local_irq_disable();
461         guest_enter_irqoff();
462         trace_kvm_enter(vcpu);
463
464         /*
465          * Make sure the read of VCPU requests in vcpu_run() callback is not
466          * reordered ahead of the write to vcpu->mode, or we could miss a TLB
467          * flush request while the requester sees the VCPU as outside of guest
468          * mode and not needing an IPI.
469          */
470         smp_store_mb(vcpu->mode, IN_GUEST_MODE);
471
472         r = kvm_mips_callbacks->vcpu_run(run, vcpu);
473
474         trace_kvm_out(vcpu);
475         guest_exit_irqoff();
476         local_irq_enable();
477
478 out:
479         kvm_sigset_deactivate(vcpu);
480
481         vcpu_put(vcpu);
482         return r;
483 }
484
485 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
486                              struct kvm_mips_interrupt *irq)
487 {
488         int intr = (int)irq->irq;
489         struct kvm_vcpu *dvcpu = NULL;
490
491         if (intr == 3 || intr == -3 || intr == 4 || intr == -4)
492                 kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
493                           (int)intr);
494
495         if (irq->cpu == -1)
496                 dvcpu = vcpu;
497         else
498                 dvcpu = vcpu->kvm->vcpus[irq->cpu];
499
500         if (intr == 2 || intr == 3 || intr == 4) {
501                 kvm_mips_callbacks->queue_io_int(dvcpu, irq);
502
503         } else if (intr == -2 || intr == -3 || intr == -4) {
504                 kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
505         } else {
506                 kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
507                         irq->cpu, irq->irq);
508                 return -EINVAL;
509         }
510
511         dvcpu->arch.wait = 0;
512
513         rcuwait_wake_up(&dvcpu->wait);
514
515         return 0;
516 }
517
518 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
519                                     struct kvm_mp_state *mp_state)
520 {
521         return -ENOIOCTLCMD;
522 }
523
524 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
525                                     struct kvm_mp_state *mp_state)
526 {
527         return -ENOIOCTLCMD;
528 }
529
530 static u64 kvm_mips_get_one_regs[] = {
531         KVM_REG_MIPS_R0,
532         KVM_REG_MIPS_R1,
533         KVM_REG_MIPS_R2,
534         KVM_REG_MIPS_R3,
535         KVM_REG_MIPS_R4,
536         KVM_REG_MIPS_R5,
537         KVM_REG_MIPS_R6,
538         KVM_REG_MIPS_R7,
539         KVM_REG_MIPS_R8,
540         KVM_REG_MIPS_R9,
541         KVM_REG_MIPS_R10,
542         KVM_REG_MIPS_R11,
543         KVM_REG_MIPS_R12,
544         KVM_REG_MIPS_R13,
545         KVM_REG_MIPS_R14,
546         KVM_REG_MIPS_R15,
547         KVM_REG_MIPS_R16,
548         KVM_REG_MIPS_R17,
549         KVM_REG_MIPS_R18,
550         KVM_REG_MIPS_R19,
551         KVM_REG_MIPS_R20,
552         KVM_REG_MIPS_R21,
553         KVM_REG_MIPS_R22,
554         KVM_REG_MIPS_R23,
555         KVM_REG_MIPS_R24,
556         KVM_REG_MIPS_R25,
557         KVM_REG_MIPS_R26,
558         KVM_REG_MIPS_R27,
559         KVM_REG_MIPS_R28,
560         KVM_REG_MIPS_R29,
561         KVM_REG_MIPS_R30,
562         KVM_REG_MIPS_R31,
563
564 #ifndef CONFIG_CPU_MIPSR6
565         KVM_REG_MIPS_HI,
566         KVM_REG_MIPS_LO,
567 #endif
568         KVM_REG_MIPS_PC,
569 };
570
571 static u64 kvm_mips_get_one_regs_fpu[] = {
572         KVM_REG_MIPS_FCR_IR,
573         KVM_REG_MIPS_FCR_CSR,
574 };
575
576 static u64 kvm_mips_get_one_regs_msa[] = {
577         KVM_REG_MIPS_MSA_IR,
578         KVM_REG_MIPS_MSA_CSR,
579 };
580
581 static unsigned long kvm_mips_num_regs(struct kvm_vcpu *vcpu)
582 {
583         unsigned long ret;
584
585         ret = ARRAY_SIZE(kvm_mips_get_one_regs);
586         if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
587                 ret += ARRAY_SIZE(kvm_mips_get_one_regs_fpu) + 48;
588                 /* odd doubles */
589                 if (boot_cpu_data.fpu_id & MIPS_FPIR_F64)
590                         ret += 16;
591         }
592         if (kvm_mips_guest_can_have_msa(&vcpu->arch))
593                 ret += ARRAY_SIZE(kvm_mips_get_one_regs_msa) + 32;
594         ret += kvm_mips_callbacks->num_regs(vcpu);
595
596         return ret;
597 }
598
599 static int kvm_mips_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
600 {
601         u64 index;
602         unsigned int i;
603
604         if (copy_to_user(indices, kvm_mips_get_one_regs,
605                          sizeof(kvm_mips_get_one_regs)))
606                 return -EFAULT;
607         indices += ARRAY_SIZE(kvm_mips_get_one_regs);
608
609         if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
610                 if (copy_to_user(indices, kvm_mips_get_one_regs_fpu,
611                                  sizeof(kvm_mips_get_one_regs_fpu)))
612                         return -EFAULT;
613                 indices += ARRAY_SIZE(kvm_mips_get_one_regs_fpu);
614
615                 for (i = 0; i < 32; ++i) {
616                         index = KVM_REG_MIPS_FPR_32(i);
617                         if (copy_to_user(indices, &index, sizeof(index)))
618                                 return -EFAULT;
619                         ++indices;
620
621                         /* skip odd doubles if no F64 */
622                         if (i & 1 && !(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
623                                 continue;
624
625                         index = KVM_REG_MIPS_FPR_64(i);
626                         if (copy_to_user(indices, &index, sizeof(index)))
627                                 return -EFAULT;
628                         ++indices;
629                 }
630         }
631
632         if (kvm_mips_guest_can_have_msa(&vcpu->arch)) {
633                 if (copy_to_user(indices, kvm_mips_get_one_regs_msa,
634                                  sizeof(kvm_mips_get_one_regs_msa)))
635                         return -EFAULT;
636                 indices += ARRAY_SIZE(kvm_mips_get_one_regs_msa);
637
638                 for (i = 0; i < 32; ++i) {
639                         index = KVM_REG_MIPS_VEC_128(i);
640                         if (copy_to_user(indices, &index, sizeof(index)))
641                                 return -EFAULT;
642                         ++indices;
643                 }
644         }
645
646         return kvm_mips_callbacks->copy_reg_indices(vcpu, indices);
647 }
648
649 static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
650                             const struct kvm_one_reg *reg)
651 {
652         struct mips_coproc *cop0 = vcpu->arch.cop0;
653         struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
654         int ret;
655         s64 v;
656         s64 vs[2];
657         unsigned int idx;
658
659         switch (reg->id) {
660         /* General purpose registers */
661         case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
662                 v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
663                 break;
664 #ifndef CONFIG_CPU_MIPSR6
665         case KVM_REG_MIPS_HI:
666                 v = (long)vcpu->arch.hi;
667                 break;
668         case KVM_REG_MIPS_LO:
669                 v = (long)vcpu->arch.lo;
670                 break;
671 #endif
672         case KVM_REG_MIPS_PC:
673                 v = (long)vcpu->arch.pc;
674                 break;
675
676         /* Floating point registers */
677         case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
678                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
679                         return -EINVAL;
680                 idx = reg->id - KVM_REG_MIPS_FPR_32(0);
681                 /* Odd singles in top of even double when FR=0 */
682                 if (kvm_read_c0_guest_status(cop0) & ST0_FR)
683                         v = get_fpr32(&fpu->fpr[idx], 0);
684                 else
685                         v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
686                 break;
687         case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
688                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
689                         return -EINVAL;
690                 idx = reg->id - KVM_REG_MIPS_FPR_64(0);
691                 /* Can't access odd doubles in FR=0 mode */
692                 if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
693                         return -EINVAL;
694                 v = get_fpr64(&fpu->fpr[idx], 0);
695                 break;
696         case KVM_REG_MIPS_FCR_IR:
697                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
698                         return -EINVAL;
699                 v = boot_cpu_data.fpu_id;
700                 break;
701         case KVM_REG_MIPS_FCR_CSR:
702                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
703                         return -EINVAL;
704                 v = fpu->fcr31;
705                 break;
706
707         /* MIPS SIMD Architecture (MSA) registers */
708         case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
709                 if (!kvm_mips_guest_has_msa(&vcpu->arch))
710                         return -EINVAL;
711                 /* Can't access MSA registers in FR=0 mode */
712                 if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
713                         return -EINVAL;
714                 idx = reg->id - KVM_REG_MIPS_VEC_128(0);
715 #ifdef CONFIG_CPU_LITTLE_ENDIAN
716                 /* least significant byte first */
717                 vs[0] = get_fpr64(&fpu->fpr[idx], 0);
718                 vs[1] = get_fpr64(&fpu->fpr[idx], 1);
719 #else
720                 /* most significant byte first */
721                 vs[0] = get_fpr64(&fpu->fpr[idx], 1);
722                 vs[1] = get_fpr64(&fpu->fpr[idx], 0);
723 #endif
724                 break;
725         case KVM_REG_MIPS_MSA_IR:
726                 if (!kvm_mips_guest_has_msa(&vcpu->arch))
727                         return -EINVAL;
728                 v = boot_cpu_data.msa_id;
729                 break;
730         case KVM_REG_MIPS_MSA_CSR:
731                 if (!kvm_mips_guest_has_msa(&vcpu->arch))
732                         return -EINVAL;
733                 v = fpu->msacsr;
734                 break;
735
736         /* registers to be handled specially */
737         default:
738                 ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
739                 if (ret)
740                         return ret;
741                 break;
742         }
743         if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
744                 u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
745
746                 return put_user(v, uaddr64);
747         } else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
748                 u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
749                 u32 v32 = (u32)v;
750
751                 return put_user(v32, uaddr32);
752         } else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
753                 void __user *uaddr = (void __user *)(long)reg->addr;
754
755                 return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
756         } else {
757                 return -EINVAL;
758         }
759 }
760
761 static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
762                             const struct kvm_one_reg *reg)
763 {
764         struct mips_coproc *cop0 = vcpu->arch.cop0;
765         struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
766         s64 v;
767         s64 vs[2];
768         unsigned int idx;
769
770         if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
771                 u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
772
773                 if (get_user(v, uaddr64) != 0)
774                         return -EFAULT;
775         } else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
776                 u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
777                 s32 v32;
778
779                 if (get_user(v32, uaddr32) != 0)
780                         return -EFAULT;
781                 v = (s64)v32;
782         } else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
783                 void __user *uaddr = (void __user *)(long)reg->addr;
784
785                 return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
786         } else {
787                 return -EINVAL;
788         }
789
790         switch (reg->id) {
791         /* General purpose registers */
792         case KVM_REG_MIPS_R0:
793                 /* Silently ignore requests to set $0 */
794                 break;
795         case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
796                 vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
797                 break;
798 #ifndef CONFIG_CPU_MIPSR6
799         case KVM_REG_MIPS_HI:
800                 vcpu->arch.hi = v;
801                 break;
802         case KVM_REG_MIPS_LO:
803                 vcpu->arch.lo = v;
804                 break;
805 #endif
806         case KVM_REG_MIPS_PC:
807                 vcpu->arch.pc = v;
808                 break;
809
810         /* Floating point registers */
811         case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
812                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
813                         return -EINVAL;
814                 idx = reg->id - KVM_REG_MIPS_FPR_32(0);
815                 /* Odd singles in top of even double when FR=0 */
816                 if (kvm_read_c0_guest_status(cop0) & ST0_FR)
817                         set_fpr32(&fpu->fpr[idx], 0, v);
818                 else
819                         set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
820                 break;
821         case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
822                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
823                         return -EINVAL;
824                 idx = reg->id - KVM_REG_MIPS_FPR_64(0);
825                 /* Can't access odd doubles in FR=0 mode */
826                 if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
827                         return -EINVAL;
828                 set_fpr64(&fpu->fpr[idx], 0, v);
829                 break;
830         case KVM_REG_MIPS_FCR_IR:
831                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
832                         return -EINVAL;
833                 /* Read-only */
834                 break;
835         case KVM_REG_MIPS_FCR_CSR:
836                 if (!kvm_mips_guest_has_fpu(&vcpu->arch))
837                         return -EINVAL;
838                 fpu->fcr31 = v;
839                 break;
840
841         /* MIPS SIMD Architecture (MSA) registers */
842         case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
843                 if (!kvm_mips_guest_has_msa(&vcpu->arch))
844                         return -EINVAL;
845                 idx = reg->id - KVM_REG_MIPS_VEC_128(0);
846 #ifdef CONFIG_CPU_LITTLE_ENDIAN
847                 /* least significant byte first */
848                 set_fpr64(&fpu->fpr[idx], 0, vs[0]);
849                 set_fpr64(&fpu->fpr[idx], 1, vs[1]);
850 #else
851                 /* most significant byte first */
852                 set_fpr64(&fpu->fpr[idx], 1, vs[0]);
853                 set_fpr64(&fpu->fpr[idx], 0, vs[1]);
854 #endif
855                 break;
856         case KVM_REG_MIPS_MSA_IR:
857                 if (!kvm_mips_guest_has_msa(&vcpu->arch))
858                         return -EINVAL;
859                 /* Read-only */
860                 break;
861         case KVM_REG_MIPS_MSA_CSR:
862                 if (!kvm_mips_guest_has_msa(&vcpu->arch))
863                         return -EINVAL;
864                 fpu->msacsr = v;
865                 break;
866
867         /* registers to be handled specially */
868         default:
869                 return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
870         }
871         return 0;
872 }
873
874 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
875                                      struct kvm_enable_cap *cap)
876 {
877         int r = 0;
878
879         if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
880                 return -EINVAL;
881         if (cap->flags)
882                 return -EINVAL;
883         if (cap->args[0])
884                 return -EINVAL;
885
886         switch (cap->cap) {
887         case KVM_CAP_MIPS_FPU:
888                 vcpu->arch.fpu_enabled = true;
889                 break;
890         case KVM_CAP_MIPS_MSA:
891                 vcpu->arch.msa_enabled = true;
892                 break;
893         default:
894                 r = -EINVAL;
895                 break;
896         }
897
898         return r;
899 }
900
901 long kvm_arch_vcpu_async_ioctl(struct file *filp, unsigned int ioctl,
902                                unsigned long arg)
903 {
904         struct kvm_vcpu *vcpu = filp->private_data;
905         void __user *argp = (void __user *)arg;
906
907         if (ioctl == KVM_INTERRUPT) {
908                 struct kvm_mips_interrupt irq;
909
910                 if (copy_from_user(&irq, argp, sizeof(irq)))
911                         return -EFAULT;
912                 kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
913                           irq.irq);
914
915                 return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
916         }
917
918         return -ENOIOCTLCMD;
919 }
920
921 long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
922                          unsigned long arg)
923 {
924         struct kvm_vcpu *vcpu = filp->private_data;
925         void __user *argp = (void __user *)arg;
926         long r;
927
928         vcpu_load(vcpu);
929
930         switch (ioctl) {
931         case KVM_SET_ONE_REG:
932         case KVM_GET_ONE_REG: {
933                 struct kvm_one_reg reg;
934
935                 r = -EFAULT;
936                 if (copy_from_user(&reg, argp, sizeof(reg)))
937                         break;
938                 if (ioctl == KVM_SET_ONE_REG)
939                         r = kvm_mips_set_reg(vcpu, &reg);
940                 else
941                         r = kvm_mips_get_reg(vcpu, &reg);
942                 break;
943         }
944         case KVM_GET_REG_LIST: {
945                 struct kvm_reg_list __user *user_list = argp;
946                 struct kvm_reg_list reg_list;
947                 unsigned n;
948
949                 r = -EFAULT;
950                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
951                         break;
952                 n = reg_list.n;
953                 reg_list.n = kvm_mips_num_regs(vcpu);
954                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
955                         break;
956                 r = -E2BIG;
957                 if (n < reg_list.n)
958                         break;
959                 r = kvm_mips_copy_reg_indices(vcpu, user_list->reg);
960                 break;
961         }
962         case KVM_ENABLE_CAP: {
963                 struct kvm_enable_cap cap;
964
965                 r = -EFAULT;
966                 if (copy_from_user(&cap, argp, sizeof(cap)))
967                         break;
968                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
969                 break;
970         }
971         default:
972                 r = -ENOIOCTLCMD;
973         }
974
975         vcpu_put(vcpu);
976         return r;
977 }
978
979 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
980 {
981
982 }
983
984 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
985                                         struct kvm_memory_slot *memslot)
986 {
987         /* Let implementation handle TLB/GVA invalidation */
988         kvm_mips_callbacks->flush_shadow_memslot(kvm, memslot);
989 }
990
991 long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
992 {
993         long r;
994
995         switch (ioctl) {
996         default:
997                 r = -ENOIOCTLCMD;
998         }
999
1000         return r;
1001 }
1002
1003 int kvm_arch_init(void *opaque)
1004 {
1005         if (kvm_mips_callbacks) {
1006                 kvm_err("kvm: module already exists\n");
1007                 return -EEXIST;
1008         }
1009
1010         return kvm_mips_emulation_init(&kvm_mips_callbacks);
1011 }
1012
1013 void kvm_arch_exit(void)
1014 {
1015         kvm_mips_callbacks = NULL;
1016 }
1017
1018 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1019                                   struct kvm_sregs *sregs)
1020 {
1021         return -ENOIOCTLCMD;
1022 }
1023
1024 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1025                                   struct kvm_sregs *sregs)
1026 {
1027         return -ENOIOCTLCMD;
1028 }
1029
1030 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1031 {
1032 }
1033
1034 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1035 {
1036         return -ENOIOCTLCMD;
1037 }
1038
1039 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1040 {
1041         return -ENOIOCTLCMD;
1042 }
1043
1044 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1045 {
1046         return VM_FAULT_SIGBUS;
1047 }
1048
1049 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1050 {
1051         int r;
1052
1053         switch (ext) {
1054         case KVM_CAP_ONE_REG:
1055         case KVM_CAP_ENABLE_CAP:
1056         case KVM_CAP_READONLY_MEM:
1057         case KVM_CAP_SYNC_MMU:
1058         case KVM_CAP_IMMEDIATE_EXIT:
1059                 r = 1;
1060                 break;
1061         case KVM_CAP_NR_VCPUS:
1062                 r = num_online_cpus();
1063                 break;
1064         case KVM_CAP_MAX_VCPUS:
1065                 r = KVM_MAX_VCPUS;
1066                 break;
1067         case KVM_CAP_MAX_VCPU_ID:
1068                 r = KVM_MAX_VCPU_ID;
1069                 break;
1070         case KVM_CAP_MIPS_FPU:
1071                 /* We don't handle systems with inconsistent cpu_has_fpu */
1072                 r = !!raw_cpu_has_fpu;
1073                 break;
1074         case KVM_CAP_MIPS_MSA:
1075                 /*
1076                  * We don't support MSA vector partitioning yet:
1077                  * 1) It would require explicit support which can't be tested
1078                  *    yet due to lack of support in current hardware.
1079                  * 2) It extends the state that would need to be saved/restored
1080                  *    by e.g. QEMU for migration.
1081                  *
1082                  * When vector partitioning hardware becomes available, support
1083                  * could be added by requiring a flag when enabling
1084                  * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1085                  * to save/restore the appropriate extra state.
1086                  */
1087                 r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1088                 break;
1089         default:
1090                 r = kvm_mips_callbacks->check_extension(kvm, ext);
1091                 break;
1092         }
1093         return r;
1094 }
1095
1096 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1097 {
1098         return kvm_mips_pending_timer(vcpu) ||
1099                 kvm_read_c0_guest_cause(vcpu->arch.cop0) & C_TI;
1100 }
1101
1102 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1103 {
1104         int i;
1105         struct mips_coproc *cop0;
1106
1107         if (!vcpu)
1108                 return -1;
1109
1110         kvm_debug("VCPU Register Dump:\n");
1111         kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1112         kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1113
1114         for (i = 0; i < 32; i += 4) {
1115                 kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1116                        vcpu->arch.gprs[i],
1117                        vcpu->arch.gprs[i + 1],
1118                        vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1119         }
1120         kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1121         kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1122
1123         cop0 = vcpu->arch.cop0;
1124         kvm_debug("\tStatus: 0x%08x, Cause: 0x%08x\n",
1125                   kvm_read_c0_guest_status(cop0),
1126                   kvm_read_c0_guest_cause(cop0));
1127
1128         kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1129
1130         return 0;
1131 }
1132
1133 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1134 {
1135         int i;
1136
1137         vcpu_load(vcpu);
1138
1139         for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1140                 vcpu->arch.gprs[i] = regs->gpr[i];
1141         vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1142         vcpu->arch.hi = regs->hi;
1143         vcpu->arch.lo = regs->lo;
1144         vcpu->arch.pc = regs->pc;
1145
1146         vcpu_put(vcpu);
1147         return 0;
1148 }
1149
1150 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1151 {
1152         int i;
1153
1154         vcpu_load(vcpu);
1155
1156         for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1157                 regs->gpr[i] = vcpu->arch.gprs[i];
1158
1159         regs->hi = vcpu->arch.hi;
1160         regs->lo = vcpu->arch.lo;
1161         regs->pc = vcpu->arch.pc;
1162
1163         vcpu_put(vcpu);
1164         return 0;
1165 }
1166
1167 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1168                                   struct kvm_translation *tr)
1169 {
1170         return 0;
1171 }
1172
1173 static void kvm_mips_set_c0_status(void)
1174 {
1175         u32 status = read_c0_status();
1176
1177         if (cpu_has_dsp)
1178                 status |= (ST0_MX);
1179
1180         write_c0_status(status);
1181         ehb();
1182 }
1183
1184 /*
1185  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1186  */
1187 int kvm_mips_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1188 {
1189         u32 cause = vcpu->arch.host_cp0_cause;
1190         u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1191         u32 __user *opc = (u32 __user *) vcpu->arch.pc;
1192         unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1193         enum emulation_result er = EMULATE_DONE;
1194         u32 inst;
1195         int ret = RESUME_GUEST;
1196
1197         vcpu->mode = OUTSIDE_GUEST_MODE;
1198
1199         /* re-enable HTW before enabling interrupts */
1200         if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ))
1201                 htw_start();
1202
1203         /* Set a default exit reason */
1204         run->exit_reason = KVM_EXIT_UNKNOWN;
1205         run->ready_for_interrupt_injection = 1;
1206
1207         /*
1208          * Set the appropriate status bits based on host CPU features,
1209          * before we hit the scheduler
1210          */
1211         kvm_mips_set_c0_status();
1212
1213         local_irq_enable();
1214
1215         kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1216                         cause, opc, run, vcpu);
1217         trace_kvm_exit(vcpu, exccode);
1218
1219         if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1220                 /*
1221                  * Do a privilege check, if in UM most of these exit conditions
1222                  * end up causing an exception to be delivered to the Guest
1223                  * Kernel
1224                  */
1225                 er = kvm_mips_check_privilege(cause, opc, run, vcpu);
1226                 if (er == EMULATE_PRIV_FAIL) {
1227                         goto skip_emul;
1228                 } else if (er == EMULATE_FAIL) {
1229                         run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1230                         ret = RESUME_HOST;
1231                         goto skip_emul;
1232                 }
1233         }
1234
1235         switch (exccode) {
1236         case EXCCODE_INT:
1237                 kvm_debug("[%d]EXCCODE_INT @ %p\n", vcpu->vcpu_id, opc);
1238
1239                 ++vcpu->stat.int_exits;
1240
1241                 if (need_resched())
1242                         cond_resched();
1243
1244                 ret = RESUME_GUEST;
1245                 break;
1246
1247         case EXCCODE_CPU:
1248                 kvm_debug("EXCCODE_CPU: @ PC: %p\n", opc);
1249
1250                 ++vcpu->stat.cop_unusable_exits;
1251                 ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1252                 /* XXXKYMA: Might need to return to user space */
1253                 if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1254                         ret = RESUME_HOST;
1255                 break;
1256
1257         case EXCCODE_MOD:
1258                 ++vcpu->stat.tlbmod_exits;
1259                 ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1260                 break;
1261
1262         case EXCCODE_TLBS:
1263                 kvm_debug("TLB ST fault:  cause %#x, status %#x, PC: %p, BadVaddr: %#lx\n",
1264                           cause, kvm_read_c0_guest_status(vcpu->arch.cop0), opc,
1265                           badvaddr);
1266
1267                 ++vcpu->stat.tlbmiss_st_exits;
1268                 ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1269                 break;
1270
1271         case EXCCODE_TLBL:
1272                 kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1273                           cause, opc, badvaddr);
1274
1275                 ++vcpu->stat.tlbmiss_ld_exits;
1276                 ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1277                 break;
1278
1279         case EXCCODE_ADES:
1280                 ++vcpu->stat.addrerr_st_exits;
1281                 ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1282                 break;
1283
1284         case EXCCODE_ADEL:
1285                 ++vcpu->stat.addrerr_ld_exits;
1286                 ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1287                 break;
1288
1289         case EXCCODE_SYS:
1290                 ++vcpu->stat.syscall_exits;
1291                 ret = kvm_mips_callbacks->handle_syscall(vcpu);
1292                 break;
1293
1294         case EXCCODE_RI:
1295                 ++vcpu->stat.resvd_inst_exits;
1296                 ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1297                 break;
1298
1299         case EXCCODE_BP:
1300                 ++vcpu->stat.break_inst_exits;
1301                 ret = kvm_mips_callbacks->handle_break(vcpu);
1302                 break;
1303
1304         case EXCCODE_TR:
1305                 ++vcpu->stat.trap_inst_exits;
1306                 ret = kvm_mips_callbacks->handle_trap(vcpu);
1307                 break;
1308
1309         case EXCCODE_MSAFPE:
1310                 ++vcpu->stat.msa_fpe_exits;
1311                 ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1312                 break;
1313
1314         case EXCCODE_FPE:
1315                 ++vcpu->stat.fpe_exits;
1316                 ret = kvm_mips_callbacks->handle_fpe(vcpu);
1317                 break;
1318
1319         case EXCCODE_MSADIS:
1320                 ++vcpu->stat.msa_disabled_exits;
1321                 ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1322                 break;
1323
1324         case EXCCODE_GE:
1325                 /* defer exit accounting to handler */
1326                 ret = kvm_mips_callbacks->handle_guest_exit(vcpu);
1327                 break;
1328
1329         default:
1330                 if (cause & CAUSEF_BD)
1331                         opc += 1;
1332                 inst = 0;
1333                 kvm_get_badinstr(opc, vcpu, &inst);
1334                 kvm_err("Exception Code: %d, not yet handled, @ PC: %p, inst: 0x%08x  BadVaddr: %#lx Status: %#x\n",
1335                         exccode, opc, inst, badvaddr,
1336                         kvm_read_c0_guest_status(vcpu->arch.cop0));
1337                 kvm_arch_vcpu_dump_regs(vcpu);
1338                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1339                 ret = RESUME_HOST;
1340                 break;
1341
1342         }
1343
1344 skip_emul:
1345         local_irq_disable();
1346
1347         if (ret == RESUME_GUEST)
1348                 kvm_vz_acquire_htimer(vcpu);
1349
1350         if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1351                 kvm_mips_deliver_interrupts(vcpu, cause);
1352
1353         if (!(ret & RESUME_HOST)) {
1354                 /* Only check for signals if not already exiting to userspace */
1355                 if (signal_pending(current)) {
1356                         run->exit_reason = KVM_EXIT_INTR;
1357                         ret = (-EINTR << 2) | RESUME_HOST;
1358                         ++vcpu->stat.signal_exits;
1359                         trace_kvm_exit(vcpu, KVM_TRACE_EXIT_SIGNAL);
1360                 }
1361         }
1362
1363         if (ret == RESUME_GUEST) {
1364                 trace_kvm_reenter(vcpu);
1365
1366                 /*
1367                  * Make sure the read of VCPU requests in vcpu_reenter()
1368                  * callback is not reordered ahead of the write to vcpu->mode,
1369                  * or we could miss a TLB flush request while the requester sees
1370                  * the VCPU as outside of guest mode and not needing an IPI.
1371                  */
1372                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1373
1374                 kvm_mips_callbacks->vcpu_reenter(run, vcpu);
1375
1376                 /*
1377                  * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1378                  * is live), restore FCR31 / MSACSR.
1379                  *
1380                  * This should be before returning to the guest exception
1381                  * vector, as it may well cause an [MSA] FP exception if there
1382                  * are pending exception bits unmasked. (see
1383                  * kvm_mips_csr_die_notifier() for how that is handled).
1384                  */
1385                 if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1386                     read_c0_status() & ST0_CU1)
1387                         __kvm_restore_fcsr(&vcpu->arch);
1388
1389                 if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1390                     read_c0_config5() & MIPS_CONF5_MSAEN)
1391                         __kvm_restore_msacsr(&vcpu->arch);
1392         }
1393
1394         /* Disable HTW before returning to guest or host */
1395         if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ))
1396                 htw_stop();
1397
1398         return ret;
1399 }
1400
1401 /* Enable FPU for guest and restore context */
1402 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1403 {
1404         struct mips_coproc *cop0 = vcpu->arch.cop0;
1405         unsigned int sr, cfg5;
1406
1407         preempt_disable();
1408
1409         sr = kvm_read_c0_guest_status(cop0);
1410
1411         /*
1412          * If MSA state is already live, it is undefined how it interacts with
1413          * FR=0 FPU state, and we don't want to hit reserved instruction
1414          * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1415          * play it safe and save it first.
1416          *
1417          * In theory we shouldn't ever hit this case since kvm_lose_fpu() should
1418          * get called when guest CU1 is set, however we can't trust the guest
1419          * not to clobber the status register directly via the commpage.
1420          */
1421         if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1422             vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1423                 kvm_lose_fpu(vcpu);
1424
1425         /*
1426          * Enable FPU for guest
1427          * We set FR and FRE according to guest context
1428          */
1429         change_c0_status(ST0_CU1 | ST0_FR, sr);
1430         if (cpu_has_fre) {
1431                 cfg5 = kvm_read_c0_guest_config5(cop0);
1432                 change_c0_config5(MIPS_CONF5_FRE, cfg5);
1433         }
1434         enable_fpu_hazard();
1435
1436         /* If guest FPU state not active, restore it now */
1437         if (!(vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1438                 __kvm_restore_fpu(&vcpu->arch);
1439                 vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1440                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1441         } else {
1442                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_FPU);
1443         }
1444
1445         preempt_enable();
1446 }
1447
1448 #ifdef CONFIG_CPU_HAS_MSA
1449 /* Enable MSA for guest and restore context */
1450 void kvm_own_msa(struct kvm_vcpu *vcpu)
1451 {
1452         struct mips_coproc *cop0 = vcpu->arch.cop0;
1453         unsigned int sr, cfg5;
1454
1455         preempt_disable();
1456
1457         /*
1458          * Enable FPU if enabled in guest, since we're restoring FPU context
1459          * anyway. We set FR and FRE according to guest context.
1460          */
1461         if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1462                 sr = kvm_read_c0_guest_status(cop0);
1463
1464                 /*
1465                  * If FR=0 FPU state is already live, it is undefined how it
1466                  * interacts with MSA state, so play it safe and save it first.
1467                  */
1468                 if (!(sr & ST0_FR) &&
1469                     (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU |
1470                                 KVM_MIPS_AUX_MSA)) == KVM_MIPS_AUX_FPU)
1471                         kvm_lose_fpu(vcpu);
1472
1473                 change_c0_status(ST0_CU1 | ST0_FR, sr);
1474                 if (sr & ST0_CU1 && cpu_has_fre) {
1475                         cfg5 = kvm_read_c0_guest_config5(cop0);
1476                         change_c0_config5(MIPS_CONF5_FRE, cfg5);
1477                 }
1478         }
1479
1480         /* Enable MSA for guest */
1481         set_c0_config5(MIPS_CONF5_MSAEN);
1482         enable_fpu_hazard();
1483
1484         switch (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA)) {
1485         case KVM_MIPS_AUX_FPU:
1486                 /*
1487                  * Guest FPU state already loaded, only restore upper MSA state
1488                  */
1489                 __kvm_restore_msa_upper(&vcpu->arch);
1490                 vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1491                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_MSA);
1492                 break;
1493         case 0:
1494                 /* Neither FPU or MSA already active, restore full MSA state */
1495                 __kvm_restore_msa(&vcpu->arch);
1496                 vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1497                 if (kvm_mips_guest_has_fpu(&vcpu->arch))
1498                         vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1499                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE,
1500                               KVM_TRACE_AUX_FPU_MSA);
1501                 break;
1502         default:
1503                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_MSA);
1504                 break;
1505         }
1506
1507         preempt_enable();
1508 }
1509 #endif
1510
1511 /* Drop FPU & MSA without saving it */
1512 void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1513 {
1514         preempt_disable();
1515         if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1516                 disable_msa();
1517                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_MSA);
1518                 vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_MSA;
1519         }
1520         if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1521                 clear_c0_status(ST0_CU1 | ST0_FR);
1522                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_FPU);
1523                 vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1524         }
1525         preempt_enable();
1526 }
1527
1528 /* Save and disable FPU & MSA */
1529 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1530 {
1531         /*
1532          * With T&E, FPU & MSA get disabled in root context (hardware) when it
1533          * is disabled in guest context (software), but the register state in
1534          * the hardware may still be in use.
1535          * This is why we explicitly re-enable the hardware before saving.
1536          */
1537
1538         preempt_disable();
1539         if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1540                 if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1541                         set_c0_config5(MIPS_CONF5_MSAEN);
1542                         enable_fpu_hazard();
1543                 }
1544
1545                 __kvm_save_msa(&vcpu->arch);
1546                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU_MSA);
1547
1548                 /* Disable MSA & FPU */
1549                 disable_msa();
1550                 if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1551                         clear_c0_status(ST0_CU1 | ST0_FR);
1552                         disable_fpu_hazard();
1553                 }
1554                 vcpu->arch.aux_inuse &= ~(KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA);
1555         } else if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1556                 if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1557                         set_c0_status(ST0_CU1);
1558                         enable_fpu_hazard();
1559                 }
1560
1561                 __kvm_save_fpu(&vcpu->arch);
1562                 vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1563                 trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1564
1565                 /* Disable FPU */
1566                 clear_c0_status(ST0_CU1 | ST0_FR);
1567                 disable_fpu_hazard();
1568         }
1569         preempt_enable();
1570 }
1571
1572 /*
1573  * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1574  * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1575  * exception if cause bits are set in the value being written.
1576  */
1577 static int kvm_mips_csr_die_notify(struct notifier_block *self,
1578                                    unsigned long cmd, void *ptr)
1579 {
1580         struct die_args *args = (struct die_args *)ptr;
1581         struct pt_regs *regs = args->regs;
1582         unsigned long pc;
1583
1584         /* Only interested in FPE and MSAFPE */
1585         if (cmd != DIE_FP && cmd != DIE_MSAFP)
1586                 return NOTIFY_DONE;
1587
1588         /* Return immediately if guest context isn't active */
1589         if (!(current->flags & PF_VCPU))
1590                 return NOTIFY_DONE;
1591
1592         /* Should never get here from user mode */
1593         BUG_ON(user_mode(regs));
1594
1595         pc = instruction_pointer(regs);
1596         switch (cmd) {
1597         case DIE_FP:
1598                 /* match 2nd instruction in __kvm_restore_fcsr */
1599                 if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1600                         return NOTIFY_DONE;
1601                 break;
1602         case DIE_MSAFP:
1603                 /* match 2nd/3rd instruction in __kvm_restore_msacsr */
1604                 if (!cpu_has_msa ||
1605                     pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1606                     pc > (unsigned long)&__kvm_restore_msacsr + 8)
1607                         return NOTIFY_DONE;
1608                 break;
1609         }
1610
1611         /* Move PC forward a little and continue executing */
1612         instruction_pointer(regs) += 4;
1613
1614         return NOTIFY_STOP;
1615 }
1616
1617 static struct notifier_block kvm_mips_csr_die_notifier = {
1618         .notifier_call = kvm_mips_csr_die_notify,
1619 };
1620
1621 static int __init kvm_mips_init(void)
1622 {
1623         int ret;
1624
1625         if (cpu_has_mmid) {
1626                 pr_warn("KVM does not yet support MMIDs. KVM Disabled\n");
1627                 return -EOPNOTSUPP;
1628         }
1629
1630         ret = kvm_mips_entry_setup();
1631         if (ret)
1632                 return ret;
1633
1634         ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1635
1636         if (ret)
1637                 return ret;
1638
1639         register_die_notifier(&kvm_mips_csr_die_notifier);
1640
1641         return 0;
1642 }
1643
1644 static void __exit kvm_mips_exit(void)
1645 {
1646         kvm_exit();
1647
1648         unregister_die_notifier(&kvm_mips_csr_die_notifier);
1649 }
1650
1651 module_init(kvm_mips_init);
1652 module_exit(kvm_mips_exit);
1653
1654 EXPORT_TRACEPOINT_SYMBOL(kvm_exit);