4e586f6110aa08c4d2df47f8aa16a053748f8268
[linux-2.6.git] / arch / ia64 / kvm / kvm-ia64.c
1 /*
2  * kvm_ia64.c: Basic KVM suppport On Itanium series processors
3  *
4  *
5  *      Copyright (C) 2007, Intel Corporation.
6  *      Xiantao Zhang  (xiantao.zhang@intel.com)
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms and conditions of the GNU General Public License,
10  * version 2, as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope it will be useful, but WITHOUT
13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
15  * more details.
16  *
17  * You should have received a copy of the GNU General Public License along with
18  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
19  * Place - Suite 330, Boston, MA 02111-1307 USA.
20  *
21  */
22
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
27 #include <linux/fs.h>
28 #include <linux/smp.h>
29 #include <linux/kvm_host.h>
30 #include <linux/kvm.h>
31 #include <linux/bitops.h>
32 #include <linux/hrtimer.h>
33 #include <linux/uaccess.h>
34 #include <linux/iommu.h>
35 #include <linux/intel-iommu.h>
36
37 #include <asm/pgtable.h>
38 #include <asm/gcc_intrin.h>
39 #include <asm/pal.h>
40 #include <asm/cacheflush.h>
41 #include <asm/div64.h>
42 #include <asm/tlb.h>
43 #include <asm/elf.h>
44
45 #include "misc.h"
46 #include "vti.h"
47 #include "iodev.h"
48 #include "ioapic.h"
49 #include "lapic.h"
50 #include "irq.h"
51
52 static unsigned long kvm_vmm_base;
53 static unsigned long kvm_vsa_base;
54 static unsigned long kvm_vm_buffer;
55 static unsigned long kvm_vm_buffer_size;
56 unsigned long kvm_vmm_gp;
57
58 static long vp_env_info;
59
60 static struct kvm_vmm_info *kvm_vmm_info;
61
62 static DEFINE_PER_CPU(struct kvm_vcpu *, last_vcpu);
63
64 struct kvm_stats_debugfs_item debugfs_entries[] = {
65         { NULL }
66 };
67
68 static void kvm_flush_icache(unsigned long start, unsigned long len)
69 {
70         int l;
71
72         for (l = 0; l < (len + 32); l += 32)
73                 ia64_fc(start + l);
74
75         ia64_sync_i();
76         ia64_srlz_i();
77 }
78
79 static void kvm_flush_tlb_all(void)
80 {
81         unsigned long i, j, count0, count1, stride0, stride1, addr;
82         long flags;
83
84         addr    = local_cpu_data->ptce_base;
85         count0  = local_cpu_data->ptce_count[0];
86         count1  = local_cpu_data->ptce_count[1];
87         stride0 = local_cpu_data->ptce_stride[0];
88         stride1 = local_cpu_data->ptce_stride[1];
89
90         local_irq_save(flags);
91         for (i = 0; i < count0; ++i) {
92                 for (j = 0; j < count1; ++j) {
93                         ia64_ptce(addr);
94                         addr += stride1;
95                 }
96                 addr += stride0;
97         }
98         local_irq_restore(flags);
99         ia64_srlz_i();                  /* srlz.i implies srlz.d */
100 }
101
102 long ia64_pal_vp_create(u64 *vpd, u64 *host_iva, u64 *opt_handler)
103 {
104         struct ia64_pal_retval iprv;
105
106         PAL_CALL_STK(iprv, PAL_VP_CREATE, (u64)vpd, (u64)host_iva,
107                         (u64)opt_handler);
108
109         return iprv.status;
110 }
111
112 static  DEFINE_SPINLOCK(vp_lock);
113
114 void kvm_arch_hardware_enable(void *garbage)
115 {
116         long  status;
117         long  tmp_base;
118         unsigned long pte;
119         unsigned long saved_psr;
120         int slot;
121
122         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
123                                 PAGE_KERNEL));
124         local_irq_save(saved_psr);
125         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
126         local_irq_restore(saved_psr);
127         if (slot < 0)
128                 return;
129
130         spin_lock(&vp_lock);
131         status = ia64_pal_vp_init_env(kvm_vsa_base ?
132                                 VP_INIT_ENV : VP_INIT_ENV_INITALIZE,
133                         __pa(kvm_vm_buffer), KVM_VM_BUFFER_BASE, &tmp_base);
134         if (status != 0) {
135                 printk(KERN_WARNING"kvm: Failed to Enable VT Support!!!!\n");
136                 return ;
137         }
138
139         if (!kvm_vsa_base) {
140                 kvm_vsa_base = tmp_base;
141                 printk(KERN_INFO"kvm: kvm_vsa_base:0x%lx\n", kvm_vsa_base);
142         }
143         spin_unlock(&vp_lock);
144         ia64_ptr_entry(0x3, slot);
145 }
146
147 void kvm_arch_hardware_disable(void *garbage)
148 {
149
150         long status;
151         int slot;
152         unsigned long pte;
153         unsigned long saved_psr;
154         unsigned long host_iva = ia64_getreg(_IA64_REG_CR_IVA);
155
156         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
157                                 PAGE_KERNEL));
158
159         local_irq_save(saved_psr);
160         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
161         local_irq_restore(saved_psr);
162         if (slot < 0)
163                 return;
164
165         status = ia64_pal_vp_exit_env(host_iva);
166         if (status)
167                 printk(KERN_DEBUG"kvm: Failed to disable VT support! :%ld\n",
168                                 status);
169         ia64_ptr_entry(0x3, slot);
170 }
171
172 void kvm_arch_check_processor_compat(void *rtn)
173 {
174         *(int *)rtn = 0;
175 }
176
177 int kvm_dev_ioctl_check_extension(long ext)
178 {
179
180         int r;
181
182         switch (ext) {
183         case KVM_CAP_IRQCHIP:
184         case KVM_CAP_MP_STATE:
185
186                 r = 1;
187                 break;
188         case KVM_CAP_COALESCED_MMIO:
189                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
190                 break;
191         case KVM_CAP_IOMMU:
192                 r = iommu_found();
193                 break;
194         default:
195                 r = 0;
196         }
197         return r;
198
199 }
200
201 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
202                                         gpa_t addr, int len, int is_write)
203 {
204         struct kvm_io_device *dev;
205
206         dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len, is_write);
207
208         return dev;
209 }
210
211 static int handle_vm_error(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
212 {
213         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
214         kvm_run->hw.hardware_exit_reason = 1;
215         return 0;
216 }
217
218 static int handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
219 {
220         struct kvm_mmio_req *p;
221         struct kvm_io_device *mmio_dev;
222
223         p = kvm_get_vcpu_ioreq(vcpu);
224
225         if ((p->addr & PAGE_MASK) == IOAPIC_DEFAULT_BASE_ADDRESS)
226                 goto mmio;
227         vcpu->mmio_needed = 1;
228         vcpu->mmio_phys_addr = kvm_run->mmio.phys_addr = p->addr;
229         vcpu->mmio_size = kvm_run->mmio.len = p->size;
230         vcpu->mmio_is_write = kvm_run->mmio.is_write = !p->dir;
231
232         if (vcpu->mmio_is_write)
233                 memcpy(vcpu->mmio_data, &p->data, p->size);
234         memcpy(kvm_run->mmio.data, &p->data, p->size);
235         kvm_run->exit_reason = KVM_EXIT_MMIO;
236         return 0;
237 mmio:
238         mmio_dev = vcpu_find_mmio_dev(vcpu, p->addr, p->size, !p->dir);
239         if (mmio_dev) {
240                 if (!p->dir)
241                         kvm_iodevice_write(mmio_dev, p->addr, p->size,
242                                                 &p->data);
243                 else
244                         kvm_iodevice_read(mmio_dev, p->addr, p->size,
245                                                 &p->data);
246
247         } else
248                 printk(KERN_ERR"kvm: No iodevice found! addr:%lx\n", p->addr);
249         p->state = STATE_IORESP_READY;
250
251         return 1;
252 }
253
254 static int handle_pal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
255 {
256         struct exit_ctl_data *p;
257
258         p = kvm_get_exit_data(vcpu);
259
260         if (p->exit_reason == EXIT_REASON_PAL_CALL)
261                 return kvm_pal_emul(vcpu, kvm_run);
262         else {
263                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
264                 kvm_run->hw.hardware_exit_reason = 2;
265                 return 0;
266         }
267 }
268
269 static int handle_sal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
270 {
271         struct exit_ctl_data *p;
272
273         p = kvm_get_exit_data(vcpu);
274
275         if (p->exit_reason == EXIT_REASON_SAL_CALL) {
276                 kvm_sal_emul(vcpu);
277                 return 1;
278         } else {
279                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
280                 kvm_run->hw.hardware_exit_reason = 3;
281                 return 0;
282         }
283
284 }
285
286 /*
287  *  offset: address offset to IPI space.
288  *  value:  deliver value.
289  */
290 static void vcpu_deliver_ipi(struct kvm_vcpu *vcpu, uint64_t dm,
291                                 uint64_t vector)
292 {
293         switch (dm) {
294         case SAPIC_FIXED:
295                 kvm_apic_set_irq(vcpu, vector, 0);
296                 break;
297         case SAPIC_NMI:
298                 kvm_apic_set_irq(vcpu, 2, 0);
299                 break;
300         case SAPIC_EXTINT:
301                 kvm_apic_set_irq(vcpu, 0, 0);
302                 break;
303         case SAPIC_INIT:
304         case SAPIC_PMI:
305         default:
306                 printk(KERN_ERR"kvm: Unimplemented Deliver reserved IPI!\n");
307                 break;
308         }
309 }
310
311 static struct kvm_vcpu *lid_to_vcpu(struct kvm *kvm, unsigned long id,
312                         unsigned long eid)
313 {
314         union ia64_lid lid;
315         int i;
316
317         for (i = 0; i < KVM_MAX_VCPUS; i++) {
318                 if (kvm->vcpus[i]) {
319                         lid.val = VCPU_LID(kvm->vcpus[i]);
320                         if (lid.id == id && lid.eid == eid)
321                                 return kvm->vcpus[i];
322                 }
323         }
324
325         return NULL;
326 }
327
328 static int handle_ipi(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
329 {
330         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
331         struct kvm_vcpu *target_vcpu;
332         struct kvm_pt_regs *regs;
333         union ia64_ipi_a addr = p->u.ipi_data.addr;
334         union ia64_ipi_d data = p->u.ipi_data.data;
335
336         target_vcpu = lid_to_vcpu(vcpu->kvm, addr.id, addr.eid);
337         if (!target_vcpu)
338                 return handle_vm_error(vcpu, kvm_run);
339
340         if (!target_vcpu->arch.launched) {
341                 regs = vcpu_regs(target_vcpu);
342
343                 regs->cr_iip = vcpu->kvm->arch.rdv_sal_data.boot_ip;
344                 regs->r1 = vcpu->kvm->arch.rdv_sal_data.boot_gp;
345
346                 target_vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
347                 if (waitqueue_active(&target_vcpu->wq))
348                         wake_up_interruptible(&target_vcpu->wq);
349         } else {
350                 vcpu_deliver_ipi(target_vcpu, data.dm, data.vector);
351                 if (target_vcpu != vcpu)
352                         kvm_vcpu_kick(target_vcpu);
353         }
354
355         return 1;
356 }
357
358 struct call_data {
359         struct kvm_ptc_g ptc_g_data;
360         struct kvm_vcpu *vcpu;
361 };
362
363 static void vcpu_global_purge(void *info)
364 {
365         struct call_data *p = (struct call_data *)info;
366         struct kvm_vcpu *vcpu = p->vcpu;
367
368         if (test_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
369                 return;
370
371         set_bit(KVM_REQ_PTC_G, &vcpu->requests);
372         if (vcpu->arch.ptc_g_count < MAX_PTC_G_NUM) {
373                 vcpu->arch.ptc_g_data[vcpu->arch.ptc_g_count++] =
374                                                         p->ptc_g_data;
375         } else {
376                 clear_bit(KVM_REQ_PTC_G, &vcpu->requests);
377                 vcpu->arch.ptc_g_count = 0;
378                 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
379         }
380 }
381
382 static int handle_global_purge(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
383 {
384         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
385         struct kvm *kvm = vcpu->kvm;
386         struct call_data call_data;
387         int i;
388
389         call_data.ptc_g_data = p->u.ptc_g_data;
390
391         for (i = 0; i < KVM_MAX_VCPUS; i++) {
392                 if (!kvm->vcpus[i] || kvm->vcpus[i]->arch.mp_state ==
393                                                 KVM_MP_STATE_UNINITIALIZED ||
394                                         vcpu == kvm->vcpus[i])
395                         continue;
396
397                 if (waitqueue_active(&kvm->vcpus[i]->wq))
398                         wake_up_interruptible(&kvm->vcpus[i]->wq);
399
400                 if (kvm->vcpus[i]->cpu != -1) {
401                         call_data.vcpu = kvm->vcpus[i];
402                         smp_call_function_single(kvm->vcpus[i]->cpu,
403                                         vcpu_global_purge, &call_data, 1);
404                 } else
405                         printk(KERN_WARNING"kvm: Uninit vcpu received ipi!\n");
406
407         }
408         return 1;
409 }
410
411 static int handle_switch_rr6(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
412 {
413         return 1;
414 }
415
416 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
417 {
418
419         ktime_t kt;
420         long itc_diff;
421         unsigned long vcpu_now_itc;
422         unsigned long expires;
423         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
424         unsigned long cyc_per_usec = local_cpu_data->cyc_per_usec;
425         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
426
427         if (irqchip_in_kernel(vcpu->kvm)) {
428
429                 vcpu_now_itc = ia64_getreg(_IA64_REG_AR_ITC) + vcpu->arch.itc_offset;
430
431                 if (time_after(vcpu_now_itc, vpd->itm)) {
432                         vcpu->arch.timer_check = 1;
433                         return 1;
434                 }
435                 itc_diff = vpd->itm - vcpu_now_itc;
436                 if (itc_diff < 0)
437                         itc_diff = -itc_diff;
438
439                 expires = div64_u64(itc_diff, cyc_per_usec);
440                 kt = ktime_set(0, 1000 * expires);
441
442                 vcpu->arch.ht_active = 1;
443                 hrtimer_start(p_ht, kt, HRTIMER_MODE_ABS);
444
445                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
446                 kvm_vcpu_block(vcpu);
447                 hrtimer_cancel(p_ht);
448                 vcpu->arch.ht_active = 0;
449
450                 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
451                         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
452                                 vcpu->arch.mp_state =
453                                         KVM_MP_STATE_RUNNABLE;
454
455                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
456                         return -EINTR;
457                 return 1;
458         } else {
459                 printk(KERN_ERR"kvm: Unsupported userspace halt!");
460                 return 0;
461         }
462 }
463
464 static int handle_vm_shutdown(struct kvm_vcpu *vcpu,
465                 struct kvm_run *kvm_run)
466 {
467         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
468         return 0;
469 }
470
471 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
472                 struct kvm_run *kvm_run)
473 {
474         return 1;
475 }
476
477 static int handle_vcpu_debug(struct kvm_vcpu *vcpu,
478                                 struct kvm_run *kvm_run)
479 {
480         printk("VMM: %s", vcpu->arch.log_buf);
481         return 1;
482 }
483
484 static int (*kvm_vti_exit_handlers[])(struct kvm_vcpu *vcpu,
485                 struct kvm_run *kvm_run) = {
486         [EXIT_REASON_VM_PANIC]              = handle_vm_error,
487         [EXIT_REASON_MMIO_INSTRUCTION]      = handle_mmio,
488         [EXIT_REASON_PAL_CALL]              = handle_pal_call,
489         [EXIT_REASON_SAL_CALL]              = handle_sal_call,
490         [EXIT_REASON_SWITCH_RR6]            = handle_switch_rr6,
491         [EXIT_REASON_VM_DESTROY]            = handle_vm_shutdown,
492         [EXIT_REASON_EXTERNAL_INTERRUPT]    = handle_external_interrupt,
493         [EXIT_REASON_IPI]                   = handle_ipi,
494         [EXIT_REASON_PTC_G]                 = handle_global_purge,
495         [EXIT_REASON_DEBUG]                 = handle_vcpu_debug,
496
497 };
498
499 static const int kvm_vti_max_exit_handlers =
500                 sizeof(kvm_vti_exit_handlers)/sizeof(*kvm_vti_exit_handlers);
501
502 static uint32_t kvm_get_exit_reason(struct kvm_vcpu *vcpu)
503 {
504         struct exit_ctl_data *p_exit_data;
505
506         p_exit_data = kvm_get_exit_data(vcpu);
507         return p_exit_data->exit_reason;
508 }
509
510 /*
511  * The guest has exited.  See if we can fix it or if we need userspace
512  * assistance.
513  */
514 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
515 {
516         u32 exit_reason = kvm_get_exit_reason(vcpu);
517         vcpu->arch.last_exit = exit_reason;
518
519         if (exit_reason < kvm_vti_max_exit_handlers
520                         && kvm_vti_exit_handlers[exit_reason])
521                 return kvm_vti_exit_handlers[exit_reason](vcpu, kvm_run);
522         else {
523                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
524                 kvm_run->hw.hardware_exit_reason = exit_reason;
525         }
526         return 0;
527 }
528
529 static inline void vti_set_rr6(unsigned long rr6)
530 {
531         ia64_set_rr(RR6, rr6);
532         ia64_srlz_i();
533 }
534
535 static int kvm_insert_vmm_mapping(struct kvm_vcpu *vcpu)
536 {
537         unsigned long pte;
538         struct kvm *kvm = vcpu->kvm;
539         int r;
540
541         /*Insert a pair of tr to map vmm*/
542         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
543         r = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
544         if (r < 0)
545                 goto out;
546         vcpu->arch.vmm_tr_slot = r;
547         /*Insert a pairt of tr to map data of vm*/
548         pte = pte_val(mk_pte_phys(__pa(kvm->arch.vm_base), PAGE_KERNEL));
549         r = ia64_itr_entry(0x3, KVM_VM_DATA_BASE,
550                                         pte, KVM_VM_DATA_SHIFT);
551         if (r < 0)
552                 goto out;
553         vcpu->arch.vm_tr_slot = r;
554         r = 0;
555 out:
556         return r;
557
558 }
559
560 static void kvm_purge_vmm_mapping(struct kvm_vcpu *vcpu)
561 {
562
563         ia64_ptr_entry(0x3, vcpu->arch.vmm_tr_slot);
564         ia64_ptr_entry(0x3, vcpu->arch.vm_tr_slot);
565
566 }
567
568 static int kvm_vcpu_pre_transition(struct kvm_vcpu *vcpu)
569 {
570         int cpu = smp_processor_id();
571
572         if (vcpu->arch.last_run_cpu != cpu ||
573                         per_cpu(last_vcpu, cpu) != vcpu) {
574                 per_cpu(last_vcpu, cpu) = vcpu;
575                 vcpu->arch.last_run_cpu = cpu;
576                 kvm_flush_tlb_all();
577         }
578
579         vcpu->arch.host_rr6 = ia64_get_rr(RR6);
580         vti_set_rr6(vcpu->arch.vmm_rr);
581         return kvm_insert_vmm_mapping(vcpu);
582 }
583 static void kvm_vcpu_post_transition(struct kvm_vcpu *vcpu)
584 {
585         kvm_purge_vmm_mapping(vcpu);
586         vti_set_rr6(vcpu->arch.host_rr6);
587 }
588
589 static int  vti_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
590 {
591         union context *host_ctx, *guest_ctx;
592         int r;
593
594         /*Get host and guest context with guest address space.*/
595         host_ctx = kvm_get_host_context(vcpu);
596         guest_ctx = kvm_get_guest_context(vcpu);
597
598         r = kvm_vcpu_pre_transition(vcpu);
599         if (r < 0)
600                 goto out;
601         kvm_vmm_info->tramp_entry(host_ctx, guest_ctx);
602         kvm_vcpu_post_transition(vcpu);
603         r = 0;
604 out:
605         return r;
606 }
607
608 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
609 {
610         int r;
611
612 again:
613         preempt_disable();
614         local_irq_disable();
615
616         if (signal_pending(current)) {
617                 local_irq_enable();
618                 preempt_enable();
619                 r = -EINTR;
620                 kvm_run->exit_reason = KVM_EXIT_INTR;
621                 goto out;
622         }
623
624         vcpu->guest_mode = 1;
625         kvm_guest_enter();
626         down_read(&vcpu->kvm->slots_lock);
627         r = vti_vcpu_run(vcpu, kvm_run);
628         if (r < 0) {
629                 local_irq_enable();
630                 preempt_enable();
631                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
632                 goto out;
633         }
634
635         vcpu->arch.launched = 1;
636         vcpu->guest_mode = 0;
637         local_irq_enable();
638
639         /*
640          * We must have an instruction between local_irq_enable() and
641          * kvm_guest_exit(), so the timer interrupt isn't delayed by
642          * the interrupt shadow.  The stat.exits increment will do nicely.
643          * But we need to prevent reordering, hence this barrier():
644          */
645         barrier();
646         kvm_guest_exit();
647         up_read(&vcpu->kvm->slots_lock);
648         preempt_enable();
649
650         r = kvm_handle_exit(kvm_run, vcpu);
651
652         if (r > 0) {
653                 if (!need_resched())
654                         goto again;
655         }
656
657 out:
658         if (r > 0) {
659                 kvm_resched(vcpu);
660                 goto again;
661         }
662
663         return r;
664 }
665
666 static void kvm_set_mmio_data(struct kvm_vcpu *vcpu)
667 {
668         struct kvm_mmio_req *p = kvm_get_vcpu_ioreq(vcpu);
669
670         if (!vcpu->mmio_is_write)
671                 memcpy(&p->data, vcpu->mmio_data, 8);
672         p->state = STATE_IORESP_READY;
673 }
674
675 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
676 {
677         int r;
678         sigset_t sigsaved;
679
680         vcpu_load(vcpu);
681
682         if (vcpu->sigset_active)
683                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
684
685         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
686                 kvm_vcpu_block(vcpu);
687                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
688                 r = -EAGAIN;
689                 goto out;
690         }
691
692         if (vcpu->mmio_needed) {
693                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
694                 kvm_set_mmio_data(vcpu);
695                 vcpu->mmio_read_completed = 1;
696                 vcpu->mmio_needed = 0;
697         }
698         r = __vcpu_run(vcpu, kvm_run);
699 out:
700         if (vcpu->sigset_active)
701                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
702
703         vcpu_put(vcpu);
704         return r;
705 }
706
707 static struct kvm *kvm_alloc_kvm(void)
708 {
709
710         struct kvm *kvm;
711         uint64_t  vm_base;
712
713         BUG_ON(sizeof(struct kvm) > KVM_VM_STRUCT_SIZE);
714
715         vm_base = __get_free_pages(GFP_KERNEL, get_order(KVM_VM_DATA_SIZE));
716
717         if (!vm_base)
718                 return ERR_PTR(-ENOMEM);
719
720         memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
721         kvm = (struct kvm *)(vm_base +
722                         offsetof(struct kvm_vm_data, kvm_vm_struct));
723         kvm->arch.vm_base = vm_base;
724         printk(KERN_DEBUG"kvm: vm's data area:0x%lx\n", vm_base);
725
726         return kvm;
727 }
728
729 struct kvm_io_range {
730         unsigned long start;
731         unsigned long size;
732         unsigned long type;
733 };
734
735 static const struct kvm_io_range io_ranges[] = {
736         {VGA_IO_START, VGA_IO_SIZE, GPFN_FRAME_BUFFER},
737         {MMIO_START, MMIO_SIZE, GPFN_LOW_MMIO},
738         {LEGACY_IO_START, LEGACY_IO_SIZE, GPFN_LEGACY_IO},
739         {IO_SAPIC_START, IO_SAPIC_SIZE, GPFN_IOSAPIC},
740         {PIB_START, PIB_SIZE, GPFN_PIB},
741 };
742
743 static void kvm_build_io_pmt(struct kvm *kvm)
744 {
745         unsigned long i, j;
746
747         /* Mark I/O ranges */
748         for (i = 0; i < (sizeof(io_ranges) / sizeof(struct kvm_io_range));
749                                                         i++) {
750                 for (j = io_ranges[i].start;
751                                 j < io_ranges[i].start + io_ranges[i].size;
752                                 j += PAGE_SIZE)
753                         kvm_set_pmt_entry(kvm, j >> PAGE_SHIFT,
754                                         io_ranges[i].type, 0);
755         }
756
757 }
758
759 /*Use unused rids to virtualize guest rid.*/
760 #define GUEST_PHYSICAL_RR0      0x1739
761 #define GUEST_PHYSICAL_RR4      0x2739
762 #define VMM_INIT_RR             0x1660
763
764 static void kvm_init_vm(struct kvm *kvm)
765 {
766         BUG_ON(!kvm);
767
768         kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
769         kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
770         kvm->arch.vmm_init_rr = VMM_INIT_RR;
771
772         /*
773          *Fill P2M entries for MMIO/IO ranges
774          */
775         kvm_build_io_pmt(kvm);
776
777         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
778
779         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
780         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
781 }
782
783 struct  kvm *kvm_arch_create_vm(void)
784 {
785         struct kvm *kvm = kvm_alloc_kvm();
786
787         if (IS_ERR(kvm))
788                 return ERR_PTR(-ENOMEM);
789         kvm_init_vm(kvm);
790
791         return kvm;
792
793 }
794
795 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
796                                         struct kvm_irqchip *chip)
797 {
798         int r;
799
800         r = 0;
801         switch (chip->chip_id) {
802         case KVM_IRQCHIP_IOAPIC:
803                 memcpy(&chip->chip.ioapic, ioapic_irqchip(kvm),
804                                 sizeof(struct kvm_ioapic_state));
805                 break;
806         default:
807                 r = -EINVAL;
808                 break;
809         }
810         return r;
811 }
812
813 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
814 {
815         int r;
816
817         r = 0;
818         switch (chip->chip_id) {
819         case KVM_IRQCHIP_IOAPIC:
820                 memcpy(ioapic_irqchip(kvm),
821                                 &chip->chip.ioapic,
822                                 sizeof(struct kvm_ioapic_state));
823                 break;
824         default:
825                 r = -EINVAL;
826                 break;
827         }
828         return r;
829 }
830
831 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
832
833 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
834 {
835         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
836         int i;
837
838         vcpu_load(vcpu);
839
840         for (i = 0; i < 16; i++) {
841                 vpd->vgr[i] = regs->vpd.vgr[i];
842                 vpd->vbgr[i] = regs->vpd.vbgr[i];
843         }
844         for (i = 0; i < 128; i++)
845                 vpd->vcr[i] = regs->vpd.vcr[i];
846         vpd->vhpi = regs->vpd.vhpi;
847         vpd->vnat = regs->vpd.vnat;
848         vpd->vbnat = regs->vpd.vbnat;
849         vpd->vpsr = regs->vpd.vpsr;
850
851         vpd->vpr = regs->vpd.vpr;
852
853         memcpy(&vcpu->arch.guest, &regs->saved_guest, sizeof(union context));
854
855         RESTORE_REGS(mp_state);
856         RESTORE_REGS(vmm_rr);
857         memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
858         memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
859         RESTORE_REGS(itr_regions);
860         RESTORE_REGS(dtr_regions);
861         RESTORE_REGS(tc_regions);
862         RESTORE_REGS(irq_check);
863         RESTORE_REGS(itc_check);
864         RESTORE_REGS(timer_check);
865         RESTORE_REGS(timer_pending);
866         RESTORE_REGS(last_itc);
867         for (i = 0; i < 8; i++) {
868                 vcpu->arch.vrr[i] = regs->vrr[i];
869                 vcpu->arch.ibr[i] = regs->ibr[i];
870                 vcpu->arch.dbr[i] = regs->dbr[i];
871         }
872         for (i = 0; i < 4; i++)
873                 vcpu->arch.insvc[i] = regs->insvc[i];
874         RESTORE_REGS(xtp);
875         RESTORE_REGS(metaphysical_rr0);
876         RESTORE_REGS(metaphysical_rr4);
877         RESTORE_REGS(metaphysical_saved_rr0);
878         RESTORE_REGS(metaphysical_saved_rr4);
879         RESTORE_REGS(fp_psr);
880         RESTORE_REGS(saved_gp);
881
882         vcpu->arch.irq_new_pending = 1;
883         vcpu->arch.itc_offset = regs->saved_itc - ia64_getreg(_IA64_REG_AR_ITC);
884         set_bit(KVM_REQ_RESUME, &vcpu->requests);
885
886         vcpu_put(vcpu);
887
888         return 0;
889 }
890
891 long kvm_arch_vm_ioctl(struct file *filp,
892                 unsigned int ioctl, unsigned long arg)
893 {
894         struct kvm *kvm = filp->private_data;
895         void __user *argp = (void __user *)arg;
896         int r = -EINVAL;
897
898         switch (ioctl) {
899         case KVM_SET_MEMORY_REGION: {
900                 struct kvm_memory_region kvm_mem;
901                 struct kvm_userspace_memory_region kvm_userspace_mem;
902
903                 r = -EFAULT;
904                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
905                         goto out;
906                 kvm_userspace_mem.slot = kvm_mem.slot;
907                 kvm_userspace_mem.flags = kvm_mem.flags;
908                 kvm_userspace_mem.guest_phys_addr =
909                                         kvm_mem.guest_phys_addr;
910                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
911                 r = kvm_vm_ioctl_set_memory_region(kvm,
912                                         &kvm_userspace_mem, 0);
913                 if (r)
914                         goto out;
915                 break;
916                 }
917         case KVM_CREATE_IRQCHIP:
918                 r = -EFAULT;
919                 r = kvm_ioapic_init(kvm);
920                 if (r)
921                         goto out;
922                 break;
923         case KVM_IRQ_LINE: {
924                 struct kvm_irq_level irq_event;
925
926                 r = -EFAULT;
927                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
928                         goto out;
929                 if (irqchip_in_kernel(kvm)) {
930                         mutex_lock(&kvm->lock);
931                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
932                                     irq_event.irq, irq_event.level);
933                         mutex_unlock(&kvm->lock);
934                         r = 0;
935                 }
936                 break;
937                 }
938         case KVM_GET_IRQCHIP: {
939                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
940                 struct kvm_irqchip chip;
941
942                 r = -EFAULT;
943                 if (copy_from_user(&chip, argp, sizeof chip))
944                                 goto out;
945                 r = -ENXIO;
946                 if (!irqchip_in_kernel(kvm))
947                         goto out;
948                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
949                 if (r)
950                         goto out;
951                 r = -EFAULT;
952                 if (copy_to_user(argp, &chip, sizeof chip))
953                                 goto out;
954                 r = 0;
955                 break;
956                 }
957         case KVM_SET_IRQCHIP: {
958                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
959                 struct kvm_irqchip chip;
960
961                 r = -EFAULT;
962                 if (copy_from_user(&chip, argp, sizeof chip))
963                                 goto out;
964                 r = -ENXIO;
965                 if (!irqchip_in_kernel(kvm))
966                         goto out;
967                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
968                 if (r)
969                         goto out;
970                 r = 0;
971                 break;
972                 }
973         default:
974                 ;
975         }
976 out:
977         return r;
978 }
979
980 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
981                 struct kvm_sregs *sregs)
982 {
983         return -EINVAL;
984 }
985
986 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
987                 struct kvm_sregs *sregs)
988 {
989         return -EINVAL;
990
991 }
992 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
993                 struct kvm_translation *tr)
994 {
995
996         return -EINVAL;
997 }
998
999 static int kvm_alloc_vmm_area(void)
1000 {
1001         if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1002                 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1003                                 get_order(KVM_VMM_SIZE));
1004                 if (!kvm_vmm_base)
1005                         return -ENOMEM;
1006
1007                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1008                 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1009
1010                 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1011                                 kvm_vmm_base, kvm_vm_buffer);
1012         }
1013
1014         return 0;
1015 }
1016
1017 static void kvm_free_vmm_area(void)
1018 {
1019         if (kvm_vmm_base) {
1020                 /*Zero this area before free to avoid bits leak!!*/
1021                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1022                 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1023                 kvm_vmm_base  = 0;
1024                 kvm_vm_buffer = 0;
1025                 kvm_vsa_base = 0;
1026         }
1027 }
1028
1029 static void vti_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1030 {
1031 }
1032
1033 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1034 {
1035         int i;
1036         union cpuid3_t cpuid3;
1037         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1038
1039         if (IS_ERR(vpd))
1040                 return PTR_ERR(vpd);
1041
1042         /* CPUID init */
1043         for (i = 0; i < 5; i++)
1044                 vpd->vcpuid[i] = ia64_get_cpuid(i);
1045
1046         /* Limit the CPUID number to 5 */
1047         cpuid3.value = vpd->vcpuid[3];
1048         cpuid3.number = 4;      /* 5 - 1 */
1049         vpd->vcpuid[3] = cpuid3.value;
1050
1051         /*Set vac and vdc fields*/
1052         vpd->vac.a_from_int_cr = 1;
1053         vpd->vac.a_to_int_cr = 1;
1054         vpd->vac.a_from_psr = 1;
1055         vpd->vac.a_from_cpuid = 1;
1056         vpd->vac.a_cover = 1;
1057         vpd->vac.a_bsw = 1;
1058         vpd->vac.a_int = 1;
1059         vpd->vdc.d_vmsw = 1;
1060
1061         /*Set virtual buffer*/
1062         vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1063
1064         return 0;
1065 }
1066
1067 static int vti_create_vp(struct kvm_vcpu *vcpu)
1068 {
1069         long ret;
1070         struct vpd *vpd = vcpu->arch.vpd;
1071         unsigned long  vmm_ivt;
1072
1073         vmm_ivt = kvm_vmm_info->vmm_ivt;
1074
1075         printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1076
1077         ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1078
1079         if (ret) {
1080                 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1081                 return -EINVAL;
1082         }
1083         return 0;
1084 }
1085
1086 static void init_ptce_info(struct kvm_vcpu *vcpu)
1087 {
1088         ia64_ptce_info_t ptce = {0};
1089
1090         ia64_get_ptce(&ptce);
1091         vcpu->arch.ptce_base = ptce.base;
1092         vcpu->arch.ptce_count[0] = ptce.count[0];
1093         vcpu->arch.ptce_count[1] = ptce.count[1];
1094         vcpu->arch.ptce_stride[0] = ptce.stride[0];
1095         vcpu->arch.ptce_stride[1] = ptce.stride[1];
1096 }
1097
1098 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1099 {
1100         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1101
1102         if (hrtimer_cancel(p_ht))
1103                 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1104 }
1105
1106 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1107 {
1108         struct kvm_vcpu *vcpu;
1109         wait_queue_head_t *q;
1110
1111         vcpu  = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1112         q = &vcpu->wq;
1113
1114         if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1115                 goto out;
1116
1117         if (waitqueue_active(q))
1118                 wake_up_interruptible(q);
1119
1120 out:
1121         vcpu->arch.timer_fired = 1;
1122         vcpu->arch.timer_check = 1;
1123         return HRTIMER_NORESTART;
1124 }
1125
1126 #define PALE_RESET_ENTRY    0x80000000ffffffb0UL
1127
1128 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1129 {
1130         struct kvm_vcpu *v;
1131         int r;
1132         int i;
1133         long itc_offset;
1134         struct kvm *kvm = vcpu->kvm;
1135         struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1136
1137         union context *p_ctx = &vcpu->arch.guest;
1138         struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1139
1140         /*Init vcpu context for first run.*/
1141         if (IS_ERR(vmm_vcpu))
1142                 return PTR_ERR(vmm_vcpu);
1143
1144         if (vcpu->vcpu_id == 0) {
1145                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1146
1147                 /*Set entry address for first run.*/
1148                 regs->cr_iip = PALE_RESET_ENTRY;
1149
1150                 /*Initialize itc offset for vcpus*/
1151                 itc_offset = 0UL - ia64_getreg(_IA64_REG_AR_ITC);
1152                 for (i = 0; i < KVM_MAX_VCPUS; i++) {
1153                         v = (struct kvm_vcpu *)((char *)vcpu +
1154                                         sizeof(struct kvm_vcpu_data) * i);
1155                         v->arch.itc_offset = itc_offset;
1156                         v->arch.last_itc = 0;
1157                 }
1158         } else
1159                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1160
1161         r = -ENOMEM;
1162         vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1163         if (!vcpu->arch.apic)
1164                 goto out;
1165         vcpu->arch.apic->vcpu = vcpu;
1166
1167         p_ctx->gr[1] = 0;
1168         p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + KVM_STK_OFFSET);
1169         p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1170         p_ctx->psr = 0x1008522000UL;
1171         p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1172         p_ctx->caller_unat = 0;
1173         p_ctx->pr = 0x0;
1174         p_ctx->ar[36] = 0x0; /*unat*/
1175         p_ctx->ar[19] = 0x0; /*rnat*/
1176         p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1177                                 ((sizeof(struct kvm_vcpu)+15) & ~15);
1178         p_ctx->ar[64] = 0x0; /*pfs*/
1179         p_ctx->cr[0] = 0x7e04UL;
1180         p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1181         p_ctx->cr[8] = 0x3c;
1182
1183         /*Initilize region register*/
1184         p_ctx->rr[0] = 0x30;
1185         p_ctx->rr[1] = 0x30;
1186         p_ctx->rr[2] = 0x30;
1187         p_ctx->rr[3] = 0x30;
1188         p_ctx->rr[4] = 0x30;
1189         p_ctx->rr[5] = 0x30;
1190         p_ctx->rr[7] = 0x30;
1191
1192         /*Initilize branch register 0*/
1193         p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1194
1195         vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1196         vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1197         vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1198
1199         hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1200         vcpu->arch.hlt_timer.function = hlt_timer_fn;
1201
1202         vcpu->arch.last_run_cpu = -1;
1203         vcpu->arch.vpd = (struct vpd *)VPD_BASE(vcpu->vcpu_id);
1204         vcpu->arch.vsa_base = kvm_vsa_base;
1205         vcpu->arch.__gp = kvm_vmm_gp;
1206         vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1207         vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_BASE(vcpu->vcpu_id);
1208         vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_BASE(vcpu->vcpu_id);
1209         init_ptce_info(vcpu);
1210
1211         r = 0;
1212 out:
1213         return r;
1214 }
1215
1216 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1217 {
1218         unsigned long psr;
1219         int r;
1220
1221         local_irq_save(psr);
1222         r = kvm_insert_vmm_mapping(vcpu);
1223         if (r)
1224                 goto fail;
1225         r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1226         if (r)
1227                 goto fail;
1228
1229         r = vti_init_vpd(vcpu);
1230         if (r) {
1231                 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1232                 goto uninit;
1233         }
1234
1235         r = vti_create_vp(vcpu);
1236         if (r)
1237                 goto uninit;
1238
1239         kvm_purge_vmm_mapping(vcpu);
1240         local_irq_restore(psr);
1241
1242         return 0;
1243 uninit:
1244         kvm_vcpu_uninit(vcpu);
1245 fail:
1246         local_irq_restore(psr);
1247         return r;
1248 }
1249
1250 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1251                 unsigned int id)
1252 {
1253         struct kvm_vcpu *vcpu;
1254         unsigned long vm_base = kvm->arch.vm_base;
1255         int r;
1256         int cpu;
1257
1258         BUG_ON(sizeof(struct kvm_vcpu) > VCPU_STRUCT_SIZE/2);
1259
1260         r = -EINVAL;
1261         if (id >= KVM_MAX_VCPUS) {
1262                 printk(KERN_ERR"kvm: Can't configure vcpus > %ld",
1263                                 KVM_MAX_VCPUS);
1264                 goto fail;
1265         }
1266
1267         r = -ENOMEM;
1268         if (!vm_base) {
1269                 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1270                 goto fail;
1271         }
1272         vcpu = (struct kvm_vcpu *)(vm_base + offsetof(struct kvm_vm_data,
1273                                         vcpu_data[id].vcpu_struct));
1274         vcpu->kvm = kvm;
1275
1276         cpu = get_cpu();
1277         vti_vcpu_load(vcpu, cpu);
1278         r = vti_vcpu_setup(vcpu, id);
1279         put_cpu();
1280
1281         if (r) {
1282                 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1283                 goto fail;
1284         }
1285
1286         return vcpu;
1287 fail:
1288         return ERR_PTR(r);
1289 }
1290
1291 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1292 {
1293         return 0;
1294 }
1295
1296 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1297 {
1298         return -EINVAL;
1299 }
1300
1301 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1302 {
1303         return -EINVAL;
1304 }
1305
1306 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1307                 struct kvm_debug_guest *dbg)
1308 {
1309         return -EINVAL;
1310 }
1311
1312 static void free_kvm(struct kvm *kvm)
1313 {
1314         unsigned long vm_base = kvm->arch.vm_base;
1315
1316         if (vm_base) {
1317                 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1318                 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1319         }
1320
1321 }
1322
1323 static void kvm_release_vm_pages(struct kvm *kvm)
1324 {
1325         struct kvm_memory_slot *memslot;
1326         int i, j;
1327         unsigned long base_gfn;
1328
1329         for (i = 0; i < kvm->nmemslots; i++) {
1330                 memslot = &kvm->memslots[i];
1331                 base_gfn = memslot->base_gfn;
1332
1333                 for (j = 0; j < memslot->npages; j++) {
1334                         if (memslot->rmap[j])
1335                                 put_page((struct page *)memslot->rmap[j]);
1336                 }
1337         }
1338 }
1339
1340 void kvm_arch_destroy_vm(struct kvm *kvm)
1341 {
1342         kvm_iommu_unmap_guest(kvm);
1343 #ifdef  KVM_CAP_DEVICE_ASSIGNMENT
1344         kvm_free_all_assigned_devices(kvm);
1345 #endif
1346         kfree(kvm->arch.vioapic);
1347         kvm_release_vm_pages(kvm);
1348         kvm_free_physmem(kvm);
1349         free_kvm(kvm);
1350 }
1351
1352 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1353 {
1354 }
1355
1356 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1357 {
1358         if (cpu != vcpu->cpu) {
1359                 vcpu->cpu = cpu;
1360                 if (vcpu->arch.ht_active)
1361                         kvm_migrate_hlt_timer(vcpu);
1362         }
1363 }
1364
1365 #define SAVE_REGS(_x)   regs->_x = vcpu->arch._x
1366
1367 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1368 {
1369         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1370         int i;
1371
1372         vcpu_load(vcpu);
1373
1374         for (i = 0; i < 16; i++) {
1375                 regs->vpd.vgr[i] = vpd->vgr[i];
1376                 regs->vpd.vbgr[i] = vpd->vbgr[i];
1377         }
1378         for (i = 0; i < 128; i++)
1379                 regs->vpd.vcr[i] = vpd->vcr[i];
1380         regs->vpd.vhpi = vpd->vhpi;
1381         regs->vpd.vnat = vpd->vnat;
1382         regs->vpd.vbnat = vpd->vbnat;
1383         regs->vpd.vpsr = vpd->vpsr;
1384         regs->vpd.vpr = vpd->vpr;
1385
1386         memcpy(&regs->saved_guest, &vcpu->arch.guest, sizeof(union context));
1387
1388         SAVE_REGS(mp_state);
1389         SAVE_REGS(vmm_rr);
1390         memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1391         memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1392         SAVE_REGS(itr_regions);
1393         SAVE_REGS(dtr_regions);
1394         SAVE_REGS(tc_regions);
1395         SAVE_REGS(irq_check);
1396         SAVE_REGS(itc_check);
1397         SAVE_REGS(timer_check);
1398         SAVE_REGS(timer_pending);
1399         SAVE_REGS(last_itc);
1400         for (i = 0; i < 8; i++) {
1401                 regs->vrr[i] = vcpu->arch.vrr[i];
1402                 regs->ibr[i] = vcpu->arch.ibr[i];
1403                 regs->dbr[i] = vcpu->arch.dbr[i];
1404         }
1405         for (i = 0; i < 4; i++)
1406                 regs->insvc[i] = vcpu->arch.insvc[i];
1407         regs->saved_itc = vcpu->arch.itc_offset + ia64_getreg(_IA64_REG_AR_ITC);
1408         SAVE_REGS(xtp);
1409         SAVE_REGS(metaphysical_rr0);
1410         SAVE_REGS(metaphysical_rr4);
1411         SAVE_REGS(metaphysical_saved_rr0);
1412         SAVE_REGS(metaphysical_saved_rr4);
1413         SAVE_REGS(fp_psr);
1414         SAVE_REGS(saved_gp);
1415
1416         vcpu_put(vcpu);
1417         return 0;
1418 }
1419
1420 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1421 {
1422
1423         hrtimer_cancel(&vcpu->arch.hlt_timer);
1424         kfree(vcpu->arch.apic);
1425 }
1426
1427
1428 long kvm_arch_vcpu_ioctl(struct file *filp,
1429                 unsigned int ioctl, unsigned long arg)
1430 {
1431         return -EINVAL;
1432 }
1433
1434 int kvm_arch_set_memory_region(struct kvm *kvm,
1435                 struct kvm_userspace_memory_region *mem,
1436                 struct kvm_memory_slot old,
1437                 int user_alloc)
1438 {
1439         unsigned long i;
1440         unsigned long pfn;
1441         int npages = mem->memory_size >> PAGE_SHIFT;
1442         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
1443         unsigned long base_gfn = memslot->base_gfn;
1444
1445         if (base_gfn + npages > (KVM_MAX_MEM_SIZE >> PAGE_SHIFT))
1446                 return -ENOMEM;
1447
1448         for (i = 0; i < npages; i++) {
1449                 pfn = gfn_to_pfn(kvm, base_gfn + i);
1450                 if (!kvm_is_mmio_pfn(pfn)) {
1451                         kvm_set_pmt_entry(kvm, base_gfn + i,
1452                                         pfn << PAGE_SHIFT,
1453                                 _PAGE_AR_RWX | _PAGE_MA_WB);
1454                         memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1455                 } else {
1456                         kvm_set_pmt_entry(kvm, base_gfn + i,
1457                                         GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1458                                         _PAGE_MA_UC);
1459                         memslot->rmap[i] = 0;
1460                         }
1461         }
1462
1463         return 0;
1464 }
1465
1466 void kvm_arch_flush_shadow(struct kvm *kvm)
1467 {
1468 }
1469
1470 long kvm_arch_dev_ioctl(struct file *filp,
1471                 unsigned int ioctl, unsigned long arg)
1472 {
1473         return -EINVAL;
1474 }
1475
1476 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1477 {
1478         kvm_vcpu_uninit(vcpu);
1479 }
1480
1481 static int vti_cpu_has_kvm_support(void)
1482 {
1483         long  avail = 1, status = 1, control = 1;
1484         long ret;
1485
1486         ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1487         if (ret)
1488                 goto out;
1489
1490         if (!(avail & PAL_PROC_VM_BIT))
1491                 goto out;
1492
1493         printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1494
1495         ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1496         if (ret)
1497                 goto out;
1498         printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1499
1500         if (!(vp_env_info & VP_OPCODE)) {
1501                 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1502                                 "vm_env_info:0x%lx\n", vp_env_info);
1503         }
1504
1505         return 1;
1506 out:
1507         return 0;
1508 }
1509
1510 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1511                                                 struct module *module)
1512 {
1513         unsigned long module_base;
1514         unsigned long vmm_size;
1515
1516         unsigned long vmm_offset, func_offset, fdesc_offset;
1517         struct fdesc *p_fdesc;
1518
1519         BUG_ON(!module);
1520
1521         if (!kvm_vmm_base) {
1522                 printk("kvm: kvm area hasn't been initilized yet!!\n");
1523                 return -EFAULT;
1524         }
1525
1526         /*Calculate new position of relocated vmm module.*/
1527         module_base = (unsigned long)module->module_core;
1528         vmm_size = module->core_size;
1529         if (unlikely(vmm_size > KVM_VMM_SIZE))
1530                 return -EFAULT;
1531
1532         memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1533         kvm_flush_icache(kvm_vmm_base, vmm_size);
1534
1535         /*Recalculate kvm_vmm_info based on new VMM*/
1536         vmm_offset = vmm_info->vmm_ivt - module_base;
1537         kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1538         printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1539                         kvm_vmm_info->vmm_ivt);
1540
1541         fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1542         kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1543                                                         fdesc_offset);
1544         func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1545         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1546         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1547         p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1548
1549         printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1550                         KVM_VMM_BASE+func_offset);
1551
1552         fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1553         kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1554                         fdesc_offset);
1555         func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1556         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1557         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1558         p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1559
1560         kvm_vmm_gp = p_fdesc->gp;
1561
1562         printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1563                                                 kvm_vmm_info->vmm_entry);
1564         printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1565                                                 KVM_VMM_BASE + func_offset);
1566
1567         return 0;
1568 }
1569
1570 int kvm_arch_init(void *opaque)
1571 {
1572         int r;
1573         struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1574
1575         if (!vti_cpu_has_kvm_support()) {
1576                 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1577                 r = -EOPNOTSUPP;
1578                 goto out;
1579         }
1580
1581         if (kvm_vmm_info) {
1582                 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1583                 r = -EEXIST;
1584                 goto out;
1585         }
1586
1587         r = -ENOMEM;
1588         kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1589         if (!kvm_vmm_info)
1590                 goto out;
1591
1592         if (kvm_alloc_vmm_area())
1593                 goto out_free0;
1594
1595         r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1596         if (r)
1597                 goto out_free1;
1598
1599         return 0;
1600
1601 out_free1:
1602         kvm_free_vmm_area();
1603 out_free0:
1604         kfree(kvm_vmm_info);
1605 out:
1606         return r;
1607 }
1608
1609 void kvm_arch_exit(void)
1610 {
1611         kvm_free_vmm_area();
1612         kfree(kvm_vmm_info);
1613         kvm_vmm_info = NULL;
1614 }
1615
1616 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1617                 struct kvm_dirty_log *log)
1618 {
1619         struct kvm_memory_slot *memslot;
1620         int r, i;
1621         long n, base;
1622         unsigned long *dirty_bitmap = (unsigned long *)(kvm->arch.vm_base +
1623                         offsetof(struct kvm_vm_data, kvm_mem_dirty_log));
1624
1625         r = -EINVAL;
1626         if (log->slot >= KVM_MEMORY_SLOTS)
1627                 goto out;
1628
1629         memslot = &kvm->memslots[log->slot];
1630         r = -ENOENT;
1631         if (!memslot->dirty_bitmap)
1632                 goto out;
1633
1634         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1635         base = memslot->base_gfn / BITS_PER_LONG;
1636
1637         for (i = 0; i < n/sizeof(long); ++i) {
1638                 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1639                 dirty_bitmap[base + i] = 0;
1640         }
1641         r = 0;
1642 out:
1643         return r;
1644 }
1645
1646 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1647                 struct kvm_dirty_log *log)
1648 {
1649         int r;
1650         int n;
1651         struct kvm_memory_slot *memslot;
1652         int is_dirty = 0;
1653
1654         spin_lock(&kvm->arch.dirty_log_lock);
1655
1656         r = kvm_ia64_sync_dirty_log(kvm, log);
1657         if (r)
1658                 goto out;
1659
1660         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1661         if (r)
1662                 goto out;
1663
1664         /* If nothing is dirty, don't bother messing with page tables. */
1665         if (is_dirty) {
1666                 kvm_flush_remote_tlbs(kvm);
1667                 memslot = &kvm->memslots[log->slot];
1668                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1669                 memset(memslot->dirty_bitmap, 0, n);
1670         }
1671         r = 0;
1672 out:
1673         spin_unlock(&kvm->arch.dirty_log_lock);
1674         return r;
1675 }
1676
1677 int kvm_arch_hardware_setup(void)
1678 {
1679         return 0;
1680 }
1681
1682 void kvm_arch_hardware_unsetup(void)
1683 {
1684 }
1685
1686 static void vcpu_kick_intr(void *info)
1687 {
1688 #ifdef DEBUG
1689         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
1690         printk(KERN_DEBUG"vcpu_kick_intr %p \n", vcpu);
1691 #endif
1692 }
1693
1694 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1695 {
1696         int ipi_pcpu = vcpu->cpu;
1697         int cpu = get_cpu();
1698
1699         if (waitqueue_active(&vcpu->wq))
1700                 wake_up_interruptible(&vcpu->wq);
1701
1702         if (vcpu->guest_mode && cpu != ipi_pcpu)
1703                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
1704         put_cpu();
1705 }
1706
1707 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
1708 {
1709
1710         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1711
1712         if (!test_and_set_bit(vec, &vpd->irr[0])) {
1713                 vcpu->arch.irq_new_pending = 1;
1714                 kvm_vcpu_kick(vcpu);
1715                 return 1;
1716         }
1717         return 0;
1718 }
1719
1720 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1721 {
1722         return apic->vcpu->vcpu_id == dest;
1723 }
1724
1725 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1726 {
1727         return 0;
1728 }
1729
1730 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
1731                                        unsigned long bitmap)
1732 {
1733         struct kvm_vcpu *lvcpu = kvm->vcpus[0];
1734         int i;
1735
1736         for (i = 1; i < KVM_MAX_VCPUS; i++) {
1737                 if (!kvm->vcpus[i])
1738                         continue;
1739                 if (lvcpu->arch.xtp > kvm->vcpus[i]->arch.xtp)
1740                         lvcpu = kvm->vcpus[i];
1741         }
1742
1743         return lvcpu;
1744 }
1745
1746 static int find_highest_bits(int *dat)
1747 {
1748         u32  bits, bitnum;
1749         int i;
1750
1751         /* loop for all 256 bits */
1752         for (i = 7; i >= 0 ; i--) {
1753                 bits = dat[i];
1754                 if (bits) {
1755                         bitnum = fls(bits);
1756                         return i * 32 + bitnum - 1;
1757                 }
1758         }
1759
1760         return -1;
1761 }
1762
1763 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1764 {
1765     struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1766
1767     if (vpd->irr[0] & (1UL << NMI_VECTOR))
1768                 return NMI_VECTOR;
1769     if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1770                 return ExtINT_VECTOR;
1771
1772     return find_highest_bits((int *)&vpd->irr[0]);
1773 }
1774
1775 int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
1776 {
1777         if (kvm_highest_pending_irq(vcpu) != -1)
1778                 return 1;
1779         return 0;
1780 }
1781
1782 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1783 {
1784         return vcpu->arch.timer_fired;
1785 }
1786
1787 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1788 {
1789         return gfn;
1790 }
1791
1792 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1793 {
1794         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE;
1795 }
1796
1797 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1798                                     struct kvm_mp_state *mp_state)
1799 {
1800         vcpu_load(vcpu);
1801         mp_state->mp_state = vcpu->arch.mp_state;
1802         vcpu_put(vcpu);
1803         return 0;
1804 }
1805
1806 static int vcpu_reset(struct kvm_vcpu *vcpu)
1807 {
1808         int r;
1809         long psr;
1810         local_irq_save(psr);
1811         r = kvm_insert_vmm_mapping(vcpu);
1812         if (r)
1813                 goto fail;
1814
1815         vcpu->arch.launched = 0;
1816         kvm_arch_vcpu_uninit(vcpu);
1817         r = kvm_arch_vcpu_init(vcpu);
1818         if (r)
1819                 goto fail;
1820
1821         kvm_purge_vmm_mapping(vcpu);
1822         r = 0;
1823 fail:
1824         local_irq_restore(psr);
1825         return r;
1826 }
1827
1828 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1829                                     struct kvm_mp_state *mp_state)
1830 {
1831         int r = 0;
1832
1833         vcpu_load(vcpu);
1834         vcpu->arch.mp_state = mp_state->mp_state;
1835         if (vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)
1836                 r = vcpu_reset(vcpu);
1837         vcpu_put(vcpu);
1838         return r;
1839 }