KVM: Consolidate userspace memory capability reporting into common code
[linux-2.6] / 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/intel-iommu.h>
35
36 #include <asm/pgtable.h>
37 #include <asm/gcc_intrin.h>
38 #include <asm/pal.h>
39 #include <asm/cacheflush.h>
40 #include <asm/div64.h>
41 #include <asm/tlb.h>
42 #include <asm/elf.h>
43
44 #include "misc.h"
45 #include "vti.h"
46 #include "iodev.h"
47 #include "ioapic.h"
48 #include "lapic.h"
49 #include "irq.h"
50
51 static unsigned long kvm_vmm_base;
52 static unsigned long kvm_vsa_base;
53 static unsigned long kvm_vm_buffer;
54 static unsigned long kvm_vm_buffer_size;
55 unsigned long kvm_vmm_gp;
56
57 static long vp_env_info;
58
59 static struct kvm_vmm_info *kvm_vmm_info;
60
61 static DEFINE_PER_CPU(struct kvm_vcpu *, last_vcpu);
62
63 struct kvm_stats_debugfs_item debugfs_entries[] = {
64         { NULL }
65 };
66
67 static void kvm_flush_icache(unsigned long start, unsigned long len)
68 {
69         int l;
70
71         for (l = 0; l < (len + 32); l += 32)
72                 ia64_fc(start + l);
73
74         ia64_sync_i();
75         ia64_srlz_i();
76 }
77
78 static void kvm_flush_tlb_all(void)
79 {
80         unsigned long i, j, count0, count1, stride0, stride1, addr;
81         long flags;
82
83         addr    = local_cpu_data->ptce_base;
84         count0  = local_cpu_data->ptce_count[0];
85         count1  = local_cpu_data->ptce_count[1];
86         stride0 = local_cpu_data->ptce_stride[0];
87         stride1 = local_cpu_data->ptce_stride[1];
88
89         local_irq_save(flags);
90         for (i = 0; i < count0; ++i) {
91                 for (j = 0; j < count1; ++j) {
92                         ia64_ptce(addr);
93                         addr += stride1;
94                 }
95                 addr += stride0;
96         }
97         local_irq_restore(flags);
98         ia64_srlz_i();                  /* srlz.i implies srlz.d */
99 }
100
101 long ia64_pal_vp_create(u64 *vpd, u64 *host_iva, u64 *opt_handler)
102 {
103         struct ia64_pal_retval iprv;
104
105         PAL_CALL_STK(iprv, PAL_VP_CREATE, (u64)vpd, (u64)host_iva,
106                         (u64)opt_handler);
107
108         return iprv.status;
109 }
110
111 static  DEFINE_SPINLOCK(vp_lock);
112
113 void kvm_arch_hardware_enable(void *garbage)
114 {
115         long  status;
116         long  tmp_base;
117         unsigned long pte;
118         unsigned long saved_psr;
119         int slot;
120
121         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
122                                 PAGE_KERNEL));
123         local_irq_save(saved_psr);
124         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
125         local_irq_restore(saved_psr);
126         if (slot < 0)
127                 return;
128
129         spin_lock(&vp_lock);
130         status = ia64_pal_vp_init_env(kvm_vsa_base ?
131                                 VP_INIT_ENV : VP_INIT_ENV_INITALIZE,
132                         __pa(kvm_vm_buffer), KVM_VM_BUFFER_BASE, &tmp_base);
133         if (status != 0) {
134                 printk(KERN_WARNING"kvm: Failed to Enable VT Support!!!!\n");
135                 return ;
136         }
137
138         if (!kvm_vsa_base) {
139                 kvm_vsa_base = tmp_base;
140                 printk(KERN_INFO"kvm: kvm_vsa_base:0x%lx\n", kvm_vsa_base);
141         }
142         spin_unlock(&vp_lock);
143         ia64_ptr_entry(0x3, slot);
144 }
145
146 void kvm_arch_hardware_disable(void *garbage)
147 {
148
149         long status;
150         int slot;
151         unsigned long pte;
152         unsigned long saved_psr;
153         unsigned long host_iva = ia64_getreg(_IA64_REG_CR_IVA);
154
155         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
156                                 PAGE_KERNEL));
157
158         local_irq_save(saved_psr);
159         slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
160         local_irq_restore(saved_psr);
161         if (slot < 0)
162                 return;
163
164         status = ia64_pal_vp_exit_env(host_iva);
165         if (status)
166                 printk(KERN_DEBUG"kvm: Failed to disable VT support! :%ld\n",
167                                 status);
168         ia64_ptr_entry(0x3, slot);
169 }
170
171 void kvm_arch_check_processor_compat(void *rtn)
172 {
173         *(int *)rtn = 0;
174 }
175
176 int kvm_dev_ioctl_check_extension(long ext)
177 {
178
179         int r;
180
181         switch (ext) {
182         case KVM_CAP_IRQCHIP:
183         case KVM_CAP_MP_STATE:
184
185                 r = 1;
186                 break;
187         case KVM_CAP_COALESCED_MMIO:
188                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
189                 break;
190         case KVM_CAP_IOMMU:
191                 r = intel_iommu_found();
192                 break;
193         default:
194                 r = 0;
195         }
196         return r;
197
198 }
199
200 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
201                                         gpa_t addr, int len, int is_write)
202 {
203         struct kvm_io_device *dev;
204
205         dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len, is_write);
206
207         return dev;
208 }
209
210 static int handle_vm_error(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
211 {
212         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
213         kvm_run->hw.hardware_exit_reason = 1;
214         return 0;
215 }
216
217 static int handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
218 {
219         struct kvm_mmio_req *p;
220         struct kvm_io_device *mmio_dev;
221
222         p = kvm_get_vcpu_ioreq(vcpu);
223
224         if ((p->addr & PAGE_MASK) == IOAPIC_DEFAULT_BASE_ADDRESS)
225                 goto mmio;
226         vcpu->mmio_needed = 1;
227         vcpu->mmio_phys_addr = kvm_run->mmio.phys_addr = p->addr;
228         vcpu->mmio_size = kvm_run->mmio.len = p->size;
229         vcpu->mmio_is_write = kvm_run->mmio.is_write = !p->dir;
230
231         if (vcpu->mmio_is_write)
232                 memcpy(vcpu->mmio_data, &p->data, p->size);
233         memcpy(kvm_run->mmio.data, &p->data, p->size);
234         kvm_run->exit_reason = KVM_EXIT_MMIO;
235         return 0;
236 mmio:
237         mmio_dev = vcpu_find_mmio_dev(vcpu, p->addr, p->size, !p->dir);
238         if (mmio_dev) {
239                 if (!p->dir)
240                         kvm_iodevice_write(mmio_dev, p->addr, p->size,
241                                                 &p->data);
242                 else
243                         kvm_iodevice_read(mmio_dev, p->addr, p->size,
244                                                 &p->data);
245
246         } else
247                 printk(KERN_ERR"kvm: No iodevice found! addr:%lx\n", p->addr);
248         p->state = STATE_IORESP_READY;
249
250         return 1;
251 }
252
253 static int handle_pal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
254 {
255         struct exit_ctl_data *p;
256
257         p = kvm_get_exit_data(vcpu);
258
259         if (p->exit_reason == EXIT_REASON_PAL_CALL)
260                 return kvm_pal_emul(vcpu, kvm_run);
261         else {
262                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
263                 kvm_run->hw.hardware_exit_reason = 2;
264                 return 0;
265         }
266 }
267
268 static int handle_sal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
269 {
270         struct exit_ctl_data *p;
271
272         p = kvm_get_exit_data(vcpu);
273
274         if (p->exit_reason == EXIT_REASON_SAL_CALL) {
275                 kvm_sal_emul(vcpu);
276                 return 1;
277         } else {
278                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
279                 kvm_run->hw.hardware_exit_reason = 3;
280                 return 0;
281         }
282
283 }
284
285 /*
286  *  offset: address offset to IPI space.
287  *  value:  deliver value.
288  */
289 static void vcpu_deliver_ipi(struct kvm_vcpu *vcpu, uint64_t dm,
290                                 uint64_t vector)
291 {
292         switch (dm) {
293         case SAPIC_FIXED:
294                 kvm_apic_set_irq(vcpu, vector, 0);
295                 break;
296         case SAPIC_NMI:
297                 kvm_apic_set_irq(vcpu, 2, 0);
298                 break;
299         case SAPIC_EXTINT:
300                 kvm_apic_set_irq(vcpu, 0, 0);
301                 break;
302         case SAPIC_INIT:
303         case SAPIC_PMI:
304         default:
305                 printk(KERN_ERR"kvm: Unimplemented Deliver reserved IPI!\n");
306                 break;
307         }
308 }
309
310 static struct kvm_vcpu *lid_to_vcpu(struct kvm *kvm, unsigned long id,
311                         unsigned long eid)
312 {
313         union ia64_lid lid;
314         int i;
315
316         for (i = 0; i < KVM_MAX_VCPUS; i++) {
317                 if (kvm->vcpus[i]) {
318                         lid.val = VCPU_LID(kvm->vcpus[i]);
319                         if (lid.id == id && lid.eid == eid)
320                                 return kvm->vcpus[i];
321                 }
322         }
323
324         return NULL;
325 }
326
327 static int handle_ipi(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
328 {
329         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
330         struct kvm_vcpu *target_vcpu;
331         struct kvm_pt_regs *regs;
332         union ia64_ipi_a addr = p->u.ipi_data.addr;
333         union ia64_ipi_d data = p->u.ipi_data.data;
334
335         target_vcpu = lid_to_vcpu(vcpu->kvm, addr.id, addr.eid);
336         if (!target_vcpu)
337                 return handle_vm_error(vcpu, kvm_run);
338
339         if (!target_vcpu->arch.launched) {
340                 regs = vcpu_regs(target_vcpu);
341
342                 regs->cr_iip = vcpu->kvm->arch.rdv_sal_data.boot_ip;
343                 regs->r1 = vcpu->kvm->arch.rdv_sal_data.boot_gp;
344
345                 target_vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
346                 if (waitqueue_active(&target_vcpu->wq))
347                         wake_up_interruptible(&target_vcpu->wq);
348         } else {
349                 vcpu_deliver_ipi(target_vcpu, data.dm, data.vector);
350                 if (target_vcpu != vcpu)
351                         kvm_vcpu_kick(target_vcpu);
352         }
353
354         return 1;
355 }
356
357 struct call_data {
358         struct kvm_ptc_g ptc_g_data;
359         struct kvm_vcpu *vcpu;
360 };
361
362 static void vcpu_global_purge(void *info)
363 {
364         struct call_data *p = (struct call_data *)info;
365         struct kvm_vcpu *vcpu = p->vcpu;
366
367         if (test_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
368                 return;
369
370         set_bit(KVM_REQ_PTC_G, &vcpu->requests);
371         if (vcpu->arch.ptc_g_count < MAX_PTC_G_NUM) {
372                 vcpu->arch.ptc_g_data[vcpu->arch.ptc_g_count++] =
373                                                         p->ptc_g_data;
374         } else {
375                 clear_bit(KVM_REQ_PTC_G, &vcpu->requests);
376                 vcpu->arch.ptc_g_count = 0;
377                 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
378         }
379 }
380
381 static int handle_global_purge(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
382 {
383         struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
384         struct kvm *kvm = vcpu->kvm;
385         struct call_data call_data;
386         int i;
387
388         call_data.ptc_g_data = p->u.ptc_g_data;
389
390         for (i = 0; i < KVM_MAX_VCPUS; i++) {
391                 if (!kvm->vcpus[i] || kvm->vcpus[i]->arch.mp_state ==
392                                                 KVM_MP_STATE_UNINITIALIZED ||
393                                         vcpu == kvm->vcpus[i])
394                         continue;
395
396                 if (waitqueue_active(&kvm->vcpus[i]->wq))
397                         wake_up_interruptible(&kvm->vcpus[i]->wq);
398
399                 if (kvm->vcpus[i]->cpu != -1) {
400                         call_data.vcpu = kvm->vcpus[i];
401                         smp_call_function_single(kvm->vcpus[i]->cpu,
402                                         vcpu_global_purge, &call_data, 1);
403                 } else
404                         printk(KERN_WARNING"kvm: Uninit vcpu received ipi!\n");
405
406         }
407         return 1;
408 }
409
410 static int handle_switch_rr6(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
411 {
412         return 1;
413 }
414
415 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
416 {
417
418         ktime_t kt;
419         long itc_diff;
420         unsigned long vcpu_now_itc;
421         unsigned long expires;
422         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
423         unsigned long cyc_per_usec = local_cpu_data->cyc_per_usec;
424         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
425
426         if (irqchip_in_kernel(vcpu->kvm)) {
427
428                 vcpu_now_itc = ia64_getreg(_IA64_REG_AR_ITC) + vcpu->arch.itc_offset;
429
430                 if (time_after(vcpu_now_itc, vpd->itm)) {
431                         vcpu->arch.timer_check = 1;
432                         return 1;
433                 }
434                 itc_diff = vpd->itm - vcpu_now_itc;
435                 if (itc_diff < 0)
436                         itc_diff = -itc_diff;
437
438                 expires = div64_u64(itc_diff, cyc_per_usec);
439                 kt = ktime_set(0, 1000 * expires);
440
441                 vcpu->arch.ht_active = 1;
442                 hrtimer_start(p_ht, kt, HRTIMER_MODE_ABS);
443
444                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
445                 kvm_vcpu_block(vcpu);
446                 hrtimer_cancel(p_ht);
447                 vcpu->arch.ht_active = 0;
448
449                 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
450                         if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
451                                 vcpu->arch.mp_state =
452                                         KVM_MP_STATE_RUNNABLE;
453
454                 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
455                         return -EINTR;
456                 return 1;
457         } else {
458                 printk(KERN_ERR"kvm: Unsupported userspace halt!");
459                 return 0;
460         }
461 }
462
463 static int handle_vm_shutdown(struct kvm_vcpu *vcpu,
464                 struct kvm_run *kvm_run)
465 {
466         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
467         return 0;
468 }
469
470 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
471                 struct kvm_run *kvm_run)
472 {
473         return 1;
474 }
475
476 static int handle_vcpu_debug(struct kvm_vcpu *vcpu,
477                                 struct kvm_run *kvm_run)
478 {
479         printk("VMM: %s", vcpu->arch.log_buf);
480         return 1;
481 }
482
483 static int (*kvm_vti_exit_handlers[])(struct kvm_vcpu *vcpu,
484                 struct kvm_run *kvm_run) = {
485         [EXIT_REASON_VM_PANIC]              = handle_vm_error,
486         [EXIT_REASON_MMIO_INSTRUCTION]      = handle_mmio,
487         [EXIT_REASON_PAL_CALL]              = handle_pal_call,
488         [EXIT_REASON_SAL_CALL]              = handle_sal_call,
489         [EXIT_REASON_SWITCH_RR6]            = handle_switch_rr6,
490         [EXIT_REASON_VM_DESTROY]            = handle_vm_shutdown,
491         [EXIT_REASON_EXTERNAL_INTERRUPT]    = handle_external_interrupt,
492         [EXIT_REASON_IPI]                   = handle_ipi,
493         [EXIT_REASON_PTC_G]                 = handle_global_purge,
494         [EXIT_REASON_DEBUG]                 = handle_vcpu_debug,
495
496 };
497
498 static const int kvm_vti_max_exit_handlers =
499                 sizeof(kvm_vti_exit_handlers)/sizeof(*kvm_vti_exit_handlers);
500
501 static uint32_t kvm_get_exit_reason(struct kvm_vcpu *vcpu)
502 {
503         struct exit_ctl_data *p_exit_data;
504
505         p_exit_data = kvm_get_exit_data(vcpu);
506         return p_exit_data->exit_reason;
507 }
508
509 /*
510  * The guest has exited.  See if we can fix it or if we need userspace
511  * assistance.
512  */
513 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
514 {
515         u32 exit_reason = kvm_get_exit_reason(vcpu);
516         vcpu->arch.last_exit = exit_reason;
517
518         if (exit_reason < kvm_vti_max_exit_handlers
519                         && kvm_vti_exit_handlers[exit_reason])
520                 return kvm_vti_exit_handlers[exit_reason](vcpu, kvm_run);
521         else {
522                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
523                 kvm_run->hw.hardware_exit_reason = exit_reason;
524         }
525         return 0;
526 }
527
528 static inline void vti_set_rr6(unsigned long rr6)
529 {
530         ia64_set_rr(RR6, rr6);
531         ia64_srlz_i();
532 }
533
534 static int kvm_insert_vmm_mapping(struct kvm_vcpu *vcpu)
535 {
536         unsigned long pte;
537         struct kvm *kvm = vcpu->kvm;
538         int r;
539
540         /*Insert a pair of tr to map vmm*/
541         pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
542         r = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
543         if (r < 0)
544                 goto out;
545         vcpu->arch.vmm_tr_slot = r;
546         /*Insert a pairt of tr to map data of vm*/
547         pte = pte_val(mk_pte_phys(__pa(kvm->arch.vm_base), PAGE_KERNEL));
548         r = ia64_itr_entry(0x3, KVM_VM_DATA_BASE,
549                                         pte, KVM_VM_DATA_SHIFT);
550         if (r < 0)
551                 goto out;
552         vcpu->arch.vm_tr_slot = r;
553         r = 0;
554 out:
555         return r;
556
557 }
558
559 static void kvm_purge_vmm_mapping(struct kvm_vcpu *vcpu)
560 {
561
562         ia64_ptr_entry(0x3, vcpu->arch.vmm_tr_slot);
563         ia64_ptr_entry(0x3, vcpu->arch.vm_tr_slot);
564
565 }
566
567 static int kvm_vcpu_pre_transition(struct kvm_vcpu *vcpu)
568 {
569         int cpu = smp_processor_id();
570
571         if (vcpu->arch.last_run_cpu != cpu ||
572                         per_cpu(last_vcpu, cpu) != vcpu) {
573                 per_cpu(last_vcpu, cpu) = vcpu;
574                 vcpu->arch.last_run_cpu = cpu;
575                 kvm_flush_tlb_all();
576         }
577
578         vcpu->arch.host_rr6 = ia64_get_rr(RR6);
579         vti_set_rr6(vcpu->arch.vmm_rr);
580         return kvm_insert_vmm_mapping(vcpu);
581 }
582 static void kvm_vcpu_post_transition(struct kvm_vcpu *vcpu)
583 {
584         kvm_purge_vmm_mapping(vcpu);
585         vti_set_rr6(vcpu->arch.host_rr6);
586 }
587
588 static int  vti_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
589 {
590         union context *host_ctx, *guest_ctx;
591         int r;
592
593         /*Get host and guest context with guest address space.*/
594         host_ctx = kvm_get_host_context(vcpu);
595         guest_ctx = kvm_get_guest_context(vcpu);
596
597         r = kvm_vcpu_pre_transition(vcpu);
598         if (r < 0)
599                 goto out;
600         kvm_vmm_info->tramp_entry(host_ctx, guest_ctx);
601         kvm_vcpu_post_transition(vcpu);
602         r = 0;
603 out:
604         return r;
605 }
606
607 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
608 {
609         int r;
610
611 again:
612         preempt_disable();
613         local_irq_disable();
614
615         if (signal_pending(current)) {
616                 local_irq_enable();
617                 preempt_enable();
618                 r = -EINTR;
619                 kvm_run->exit_reason = KVM_EXIT_INTR;
620                 goto out;
621         }
622
623         vcpu->guest_mode = 1;
624         kvm_guest_enter();
625         down_read(&vcpu->kvm->slots_lock);
626         r = vti_vcpu_run(vcpu, kvm_run);
627         if (r < 0) {
628                 local_irq_enable();
629                 preempt_enable();
630                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
631                 goto out;
632         }
633
634         vcpu->arch.launched = 1;
635         vcpu->guest_mode = 0;
636         local_irq_enable();
637
638         /*
639          * We must have an instruction between local_irq_enable() and
640          * kvm_guest_exit(), so the timer interrupt isn't delayed by
641          * the interrupt shadow.  The stat.exits increment will do nicely.
642          * But we need to prevent reordering, hence this barrier():
643          */
644         barrier();
645         kvm_guest_exit();
646         up_read(&vcpu->kvm->slots_lock);
647         preempt_enable();
648
649         r = kvm_handle_exit(kvm_run, vcpu);
650
651         if (r > 0) {
652                 if (!need_resched())
653                         goto again;
654         }
655
656 out:
657         if (r > 0) {
658                 kvm_resched(vcpu);
659                 goto again;
660         }
661
662         return r;
663 }
664
665 static void kvm_set_mmio_data(struct kvm_vcpu *vcpu)
666 {
667         struct kvm_mmio_req *p = kvm_get_vcpu_ioreq(vcpu);
668
669         if (!vcpu->mmio_is_write)
670                 memcpy(&p->data, vcpu->mmio_data, 8);
671         p->state = STATE_IORESP_READY;
672 }
673
674 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
675 {
676         int r;
677         sigset_t sigsaved;
678
679         vcpu_load(vcpu);
680
681         if (vcpu->sigset_active)
682                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
683
684         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
685                 kvm_vcpu_block(vcpu);
686                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
687                 r = -EAGAIN;
688                 goto out;
689         }
690
691         if (vcpu->mmio_needed) {
692                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
693                 kvm_set_mmio_data(vcpu);
694                 vcpu->mmio_read_completed = 1;
695                 vcpu->mmio_needed = 0;
696         }
697         r = __vcpu_run(vcpu, kvm_run);
698 out:
699         if (vcpu->sigset_active)
700                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
701
702         vcpu_put(vcpu);
703         return r;
704 }
705
706 static struct kvm *kvm_alloc_kvm(void)
707 {
708
709         struct kvm *kvm;
710         uint64_t  vm_base;
711
712         BUG_ON(sizeof(struct kvm) > KVM_VM_STRUCT_SIZE);
713
714         vm_base = __get_free_pages(GFP_KERNEL, get_order(KVM_VM_DATA_SIZE));
715
716         if (!vm_base)
717                 return ERR_PTR(-ENOMEM);
718
719         memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
720         kvm = (struct kvm *)(vm_base +
721                         offsetof(struct kvm_vm_data, kvm_vm_struct));
722         kvm->arch.vm_base = vm_base;
723         printk(KERN_DEBUG"kvm: vm's data area:0x%lx\n", vm_base);
724
725         return kvm;
726 }
727
728 struct kvm_io_range {
729         unsigned long start;
730         unsigned long size;
731         unsigned long type;
732 };
733
734 static const struct kvm_io_range io_ranges[] = {
735         {VGA_IO_START, VGA_IO_SIZE, GPFN_FRAME_BUFFER},
736         {MMIO_START, MMIO_SIZE, GPFN_LOW_MMIO},
737         {LEGACY_IO_START, LEGACY_IO_SIZE, GPFN_LEGACY_IO},
738         {IO_SAPIC_START, IO_SAPIC_SIZE, GPFN_IOSAPIC},
739         {PIB_START, PIB_SIZE, GPFN_PIB},
740 };
741
742 static void kvm_build_io_pmt(struct kvm *kvm)
743 {
744         unsigned long i, j;
745
746         /* Mark I/O ranges */
747         for (i = 0; i < (sizeof(io_ranges) / sizeof(struct kvm_io_range));
748                                                         i++) {
749                 for (j = io_ranges[i].start;
750                                 j < io_ranges[i].start + io_ranges[i].size;
751                                 j += PAGE_SIZE)
752                         kvm_set_pmt_entry(kvm, j >> PAGE_SHIFT,
753                                         io_ranges[i].type, 0);
754         }
755
756 }
757
758 /*Use unused rids to virtualize guest rid.*/
759 #define GUEST_PHYSICAL_RR0      0x1739
760 #define GUEST_PHYSICAL_RR4      0x2739
761 #define VMM_INIT_RR             0x1660
762
763 static void kvm_init_vm(struct kvm *kvm)
764 {
765         BUG_ON(!kvm);
766
767         kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
768         kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
769         kvm->arch.vmm_init_rr = VMM_INIT_RR;
770
771         /*
772          *Fill P2M entries for MMIO/IO ranges
773          */
774         kvm_build_io_pmt(kvm);
775
776         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
777
778         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
779         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
780 }
781
782 struct  kvm *kvm_arch_create_vm(void)
783 {
784         struct kvm *kvm = kvm_alloc_kvm();
785
786         if (IS_ERR(kvm))
787                 return ERR_PTR(-ENOMEM);
788         kvm_init_vm(kvm);
789
790         return kvm;
791
792 }
793
794 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
795                                         struct kvm_irqchip *chip)
796 {
797         int r;
798
799         r = 0;
800         switch (chip->chip_id) {
801         case KVM_IRQCHIP_IOAPIC:
802                 memcpy(&chip->chip.ioapic, ioapic_irqchip(kvm),
803                                 sizeof(struct kvm_ioapic_state));
804                 break;
805         default:
806                 r = -EINVAL;
807                 break;
808         }
809         return r;
810 }
811
812 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
813 {
814         int r;
815
816         r = 0;
817         switch (chip->chip_id) {
818         case KVM_IRQCHIP_IOAPIC:
819                 memcpy(ioapic_irqchip(kvm),
820                                 &chip->chip.ioapic,
821                                 sizeof(struct kvm_ioapic_state));
822                 break;
823         default:
824                 r = -EINVAL;
825                 break;
826         }
827         return r;
828 }
829
830 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
831
832 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
833 {
834         int i;
835         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
836         int r;
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         r = -EFAULT;
854         r = copy_from_user(&vcpu->arch.guest, regs->saved_guest,
855                                                 sizeof(union context));
856         if (r)
857                 goto out;
858         r = copy_from_user(vcpu + 1, regs->saved_stack +
859                         sizeof(struct kvm_vcpu),
860                         KVM_STK_OFFSET - sizeof(struct kvm_vcpu));
861         if (r)
862                 goto out;
863         vcpu->arch.exit_data =
864                 ((struct kvm_vcpu *)(regs->saved_stack))->arch.exit_data;
865
866         RESTORE_REGS(mp_state);
867         RESTORE_REGS(vmm_rr);
868         memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
869         memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
870         RESTORE_REGS(itr_regions);
871         RESTORE_REGS(dtr_regions);
872         RESTORE_REGS(tc_regions);
873         RESTORE_REGS(irq_check);
874         RESTORE_REGS(itc_check);
875         RESTORE_REGS(timer_check);
876         RESTORE_REGS(timer_pending);
877         RESTORE_REGS(last_itc);
878         for (i = 0; i < 8; i++) {
879                 vcpu->arch.vrr[i] = regs->vrr[i];
880                 vcpu->arch.ibr[i] = regs->ibr[i];
881                 vcpu->arch.dbr[i] = regs->dbr[i];
882         }
883         for (i = 0; i < 4; i++)
884                 vcpu->arch.insvc[i] = regs->insvc[i];
885         RESTORE_REGS(xtp);
886         RESTORE_REGS(metaphysical_rr0);
887         RESTORE_REGS(metaphysical_rr4);
888         RESTORE_REGS(metaphysical_saved_rr0);
889         RESTORE_REGS(metaphysical_saved_rr4);
890         RESTORE_REGS(fp_psr);
891         RESTORE_REGS(saved_gp);
892
893         vcpu->arch.irq_new_pending = 1;
894         vcpu->arch.itc_offset = regs->saved_itc - ia64_getreg(_IA64_REG_AR_ITC);
895         set_bit(KVM_REQ_RESUME, &vcpu->requests);
896
897         vcpu_put(vcpu);
898         r = 0;
899 out:
900         return r;
901 }
902
903 long kvm_arch_vm_ioctl(struct file *filp,
904                 unsigned int ioctl, unsigned long arg)
905 {
906         struct kvm *kvm = filp->private_data;
907         void __user *argp = (void __user *)arg;
908         int r = -EINVAL;
909
910         switch (ioctl) {
911         case KVM_SET_MEMORY_REGION: {
912                 struct kvm_memory_region kvm_mem;
913                 struct kvm_userspace_memory_region kvm_userspace_mem;
914
915                 r = -EFAULT;
916                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
917                         goto out;
918                 kvm_userspace_mem.slot = kvm_mem.slot;
919                 kvm_userspace_mem.flags = kvm_mem.flags;
920                 kvm_userspace_mem.guest_phys_addr =
921                                         kvm_mem.guest_phys_addr;
922                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
923                 r = kvm_vm_ioctl_set_memory_region(kvm,
924                                         &kvm_userspace_mem, 0);
925                 if (r)
926                         goto out;
927                 break;
928                 }
929         case KVM_CREATE_IRQCHIP:
930                 r = -EFAULT;
931                 r = kvm_ioapic_init(kvm);
932                 if (r)
933                         goto out;
934                 break;
935         case KVM_IRQ_LINE: {
936                 struct kvm_irq_level irq_event;
937
938                 r = -EFAULT;
939                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
940                         goto out;
941                 if (irqchip_in_kernel(kvm)) {
942                         mutex_lock(&kvm->lock);
943                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
944                                     irq_event.irq, irq_event.level);
945                         mutex_unlock(&kvm->lock);
946                         r = 0;
947                 }
948                 break;
949                 }
950         case KVM_GET_IRQCHIP: {
951                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
952                 struct kvm_irqchip chip;
953
954                 r = -EFAULT;
955                 if (copy_from_user(&chip, argp, sizeof chip))
956                                 goto out;
957                 r = -ENXIO;
958                 if (!irqchip_in_kernel(kvm))
959                         goto out;
960                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
961                 if (r)
962                         goto out;
963                 r = -EFAULT;
964                 if (copy_to_user(argp, &chip, sizeof chip))
965                                 goto out;
966                 r = 0;
967                 break;
968                 }
969         case KVM_SET_IRQCHIP: {
970                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
971                 struct kvm_irqchip chip;
972
973                 r = -EFAULT;
974                 if (copy_from_user(&chip, argp, sizeof chip))
975                                 goto out;
976                 r = -ENXIO;
977                 if (!irqchip_in_kernel(kvm))
978                         goto out;
979                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
980                 if (r)
981                         goto out;
982                 r = 0;
983                 break;
984                 }
985         default:
986                 ;
987         }
988 out:
989         return r;
990 }
991
992 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
993                 struct kvm_sregs *sregs)
994 {
995         return -EINVAL;
996 }
997
998 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
999                 struct kvm_sregs *sregs)
1000 {
1001         return -EINVAL;
1002
1003 }
1004 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1005                 struct kvm_translation *tr)
1006 {
1007
1008         return -EINVAL;
1009 }
1010
1011 static int kvm_alloc_vmm_area(void)
1012 {
1013         if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1014                 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1015                                 get_order(KVM_VMM_SIZE));
1016                 if (!kvm_vmm_base)
1017                         return -ENOMEM;
1018
1019                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1020                 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1021
1022                 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1023                                 kvm_vmm_base, kvm_vm_buffer);
1024         }
1025
1026         return 0;
1027 }
1028
1029 static void kvm_free_vmm_area(void)
1030 {
1031         if (kvm_vmm_base) {
1032                 /*Zero this area before free to avoid bits leak!!*/
1033                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1034                 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1035                 kvm_vmm_base  = 0;
1036                 kvm_vm_buffer = 0;
1037                 kvm_vsa_base = 0;
1038         }
1039 }
1040
1041 static void vti_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1042 {
1043 }
1044
1045 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1046 {
1047         int i;
1048         union cpuid3_t cpuid3;
1049         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1050
1051         if (IS_ERR(vpd))
1052                 return PTR_ERR(vpd);
1053
1054         /* CPUID init */
1055         for (i = 0; i < 5; i++)
1056                 vpd->vcpuid[i] = ia64_get_cpuid(i);
1057
1058         /* Limit the CPUID number to 5 */
1059         cpuid3.value = vpd->vcpuid[3];
1060         cpuid3.number = 4;      /* 5 - 1 */
1061         vpd->vcpuid[3] = cpuid3.value;
1062
1063         /*Set vac and vdc fields*/
1064         vpd->vac.a_from_int_cr = 1;
1065         vpd->vac.a_to_int_cr = 1;
1066         vpd->vac.a_from_psr = 1;
1067         vpd->vac.a_from_cpuid = 1;
1068         vpd->vac.a_cover = 1;
1069         vpd->vac.a_bsw = 1;
1070         vpd->vac.a_int = 1;
1071         vpd->vdc.d_vmsw = 1;
1072
1073         /*Set virtual buffer*/
1074         vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1075
1076         return 0;
1077 }
1078
1079 static int vti_create_vp(struct kvm_vcpu *vcpu)
1080 {
1081         long ret;
1082         struct vpd *vpd = vcpu->arch.vpd;
1083         unsigned long  vmm_ivt;
1084
1085         vmm_ivt = kvm_vmm_info->vmm_ivt;
1086
1087         printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1088
1089         ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1090
1091         if (ret) {
1092                 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1093                 return -EINVAL;
1094         }
1095         return 0;
1096 }
1097
1098 static void init_ptce_info(struct kvm_vcpu *vcpu)
1099 {
1100         ia64_ptce_info_t ptce = {0};
1101
1102         ia64_get_ptce(&ptce);
1103         vcpu->arch.ptce_base = ptce.base;
1104         vcpu->arch.ptce_count[0] = ptce.count[0];
1105         vcpu->arch.ptce_count[1] = ptce.count[1];
1106         vcpu->arch.ptce_stride[0] = ptce.stride[0];
1107         vcpu->arch.ptce_stride[1] = ptce.stride[1];
1108 }
1109
1110 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1111 {
1112         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1113
1114         if (hrtimer_cancel(p_ht))
1115                 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1116 }
1117
1118 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1119 {
1120         struct kvm_vcpu *vcpu;
1121         wait_queue_head_t *q;
1122
1123         vcpu  = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1124         q = &vcpu->wq;
1125
1126         if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1127                 goto out;
1128
1129         if (waitqueue_active(q))
1130                 wake_up_interruptible(q);
1131
1132 out:
1133         vcpu->arch.timer_fired = 1;
1134         vcpu->arch.timer_check = 1;
1135         return HRTIMER_NORESTART;
1136 }
1137
1138 #define PALE_RESET_ENTRY    0x80000000ffffffb0UL
1139
1140 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1141 {
1142         struct kvm_vcpu *v;
1143         int r;
1144         int i;
1145         long itc_offset;
1146         struct kvm *kvm = vcpu->kvm;
1147         struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1148
1149         union context *p_ctx = &vcpu->arch.guest;
1150         struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1151
1152         /*Init vcpu context for first run.*/
1153         if (IS_ERR(vmm_vcpu))
1154                 return PTR_ERR(vmm_vcpu);
1155
1156         if (vcpu->vcpu_id == 0) {
1157                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1158
1159                 /*Set entry address for first run.*/
1160                 regs->cr_iip = PALE_RESET_ENTRY;
1161
1162                 /*Initialize itc offset for vcpus*/
1163                 itc_offset = 0UL - ia64_getreg(_IA64_REG_AR_ITC);
1164                 for (i = 0; i < KVM_MAX_VCPUS; i++) {
1165                         v = (struct kvm_vcpu *)((char *)vcpu +
1166                                         sizeof(struct kvm_vcpu_data) * i);
1167                         v->arch.itc_offset = itc_offset;
1168                         v->arch.last_itc = 0;
1169                 }
1170         } else
1171                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1172
1173         r = -ENOMEM;
1174         vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1175         if (!vcpu->arch.apic)
1176                 goto out;
1177         vcpu->arch.apic->vcpu = vcpu;
1178
1179         p_ctx->gr[1] = 0;
1180         p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + KVM_STK_OFFSET);
1181         p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1182         p_ctx->psr = 0x1008522000UL;
1183         p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1184         p_ctx->caller_unat = 0;
1185         p_ctx->pr = 0x0;
1186         p_ctx->ar[36] = 0x0; /*unat*/
1187         p_ctx->ar[19] = 0x0; /*rnat*/
1188         p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1189                                 ((sizeof(struct kvm_vcpu)+15) & ~15);
1190         p_ctx->ar[64] = 0x0; /*pfs*/
1191         p_ctx->cr[0] = 0x7e04UL;
1192         p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1193         p_ctx->cr[8] = 0x3c;
1194
1195         /*Initilize region register*/
1196         p_ctx->rr[0] = 0x30;
1197         p_ctx->rr[1] = 0x30;
1198         p_ctx->rr[2] = 0x30;
1199         p_ctx->rr[3] = 0x30;
1200         p_ctx->rr[4] = 0x30;
1201         p_ctx->rr[5] = 0x30;
1202         p_ctx->rr[7] = 0x30;
1203
1204         /*Initilize branch register 0*/
1205         p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1206
1207         vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1208         vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1209         vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1210
1211         hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1212         vcpu->arch.hlt_timer.function = hlt_timer_fn;
1213
1214         vcpu->arch.last_run_cpu = -1;
1215         vcpu->arch.vpd = (struct vpd *)VPD_BASE(vcpu->vcpu_id);
1216         vcpu->arch.vsa_base = kvm_vsa_base;
1217         vcpu->arch.__gp = kvm_vmm_gp;
1218         vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1219         vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_BASE(vcpu->vcpu_id);
1220         vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_BASE(vcpu->vcpu_id);
1221         init_ptce_info(vcpu);
1222
1223         r = 0;
1224 out:
1225         return r;
1226 }
1227
1228 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1229 {
1230         unsigned long psr;
1231         int r;
1232
1233         local_irq_save(psr);
1234         r = kvm_insert_vmm_mapping(vcpu);
1235         if (r)
1236                 goto fail;
1237         r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1238         if (r)
1239                 goto fail;
1240
1241         r = vti_init_vpd(vcpu);
1242         if (r) {
1243                 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1244                 goto uninit;
1245         }
1246
1247         r = vti_create_vp(vcpu);
1248         if (r)
1249                 goto uninit;
1250
1251         kvm_purge_vmm_mapping(vcpu);
1252         local_irq_restore(psr);
1253
1254         return 0;
1255 uninit:
1256         kvm_vcpu_uninit(vcpu);
1257 fail:
1258         local_irq_restore(psr);
1259         return r;
1260 }
1261
1262 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1263                 unsigned int id)
1264 {
1265         struct kvm_vcpu *vcpu;
1266         unsigned long vm_base = kvm->arch.vm_base;
1267         int r;
1268         int cpu;
1269
1270         BUG_ON(sizeof(struct kvm_vcpu) > VCPU_STRUCT_SIZE/2);
1271
1272         r = -EINVAL;
1273         if (id >= KVM_MAX_VCPUS) {
1274                 printk(KERN_ERR"kvm: Can't configure vcpus > %ld",
1275                                 KVM_MAX_VCPUS);
1276                 goto fail;
1277         }
1278
1279         r = -ENOMEM;
1280         if (!vm_base) {
1281                 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1282                 goto fail;
1283         }
1284         vcpu = (struct kvm_vcpu *)(vm_base + offsetof(struct kvm_vm_data,
1285                                         vcpu_data[id].vcpu_struct));
1286         vcpu->kvm = kvm;
1287
1288         cpu = get_cpu();
1289         vti_vcpu_load(vcpu, cpu);
1290         r = vti_vcpu_setup(vcpu, id);
1291         put_cpu();
1292
1293         if (r) {
1294                 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1295                 goto fail;
1296         }
1297
1298         return vcpu;
1299 fail:
1300         return ERR_PTR(r);
1301 }
1302
1303 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1304 {
1305         return 0;
1306 }
1307
1308 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1309 {
1310         return -EINVAL;
1311 }
1312
1313 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1314 {
1315         return -EINVAL;
1316 }
1317
1318 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1319                 struct kvm_debug_guest *dbg)
1320 {
1321         return -EINVAL;
1322 }
1323
1324 static void free_kvm(struct kvm *kvm)
1325 {
1326         unsigned long vm_base = kvm->arch.vm_base;
1327
1328         if (vm_base) {
1329                 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1330                 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1331         }
1332
1333 }
1334
1335 static void kvm_release_vm_pages(struct kvm *kvm)
1336 {
1337         struct kvm_memory_slot *memslot;
1338         int i, j;
1339         unsigned long base_gfn;
1340
1341         for (i = 0; i < kvm->nmemslots; i++) {
1342                 memslot = &kvm->memslots[i];
1343                 base_gfn = memslot->base_gfn;
1344
1345                 for (j = 0; j < memslot->npages; j++) {
1346                         if (memslot->rmap[j])
1347                                 put_page((struct page *)memslot->rmap[j]);
1348                 }
1349         }
1350 }
1351
1352 void kvm_arch_destroy_vm(struct kvm *kvm)
1353 {
1354         kvm_iommu_unmap_guest(kvm);
1355 #ifdef  KVM_CAP_DEVICE_ASSIGNMENT
1356         kvm_free_all_assigned_devices(kvm);
1357 #endif
1358         kfree(kvm->arch.vioapic);
1359         kvm_release_vm_pages(kvm);
1360         kvm_free_physmem(kvm);
1361         free_kvm(kvm);
1362 }
1363
1364 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1365 {
1366 }
1367
1368 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1369 {
1370         if (cpu != vcpu->cpu) {
1371                 vcpu->cpu = cpu;
1372                 if (vcpu->arch.ht_active)
1373                         kvm_migrate_hlt_timer(vcpu);
1374         }
1375 }
1376
1377 #define SAVE_REGS(_x)   regs->_x = vcpu->arch._x
1378
1379 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1380 {
1381         int i;
1382         int r;
1383         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1384         vcpu_load(vcpu);
1385
1386         for (i = 0; i < 16; i++) {
1387                 regs->vpd.vgr[i] = vpd->vgr[i];
1388                 regs->vpd.vbgr[i] = vpd->vbgr[i];
1389         }
1390         for (i = 0; i < 128; i++)
1391                 regs->vpd.vcr[i] = vpd->vcr[i];
1392         regs->vpd.vhpi = vpd->vhpi;
1393         regs->vpd.vnat = vpd->vnat;
1394         regs->vpd.vbnat = vpd->vbnat;
1395         regs->vpd.vpsr = vpd->vpsr;
1396         regs->vpd.vpr = vpd->vpr;
1397
1398         r = -EFAULT;
1399         r = copy_to_user(regs->saved_guest, &vcpu->arch.guest,
1400                                         sizeof(union context));
1401         if (r)
1402                 goto out;
1403         r = copy_to_user(regs->saved_stack, (void *)vcpu, KVM_STK_OFFSET);
1404         if (r)
1405                 goto out;
1406         SAVE_REGS(mp_state);
1407         SAVE_REGS(vmm_rr);
1408         memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1409         memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1410         SAVE_REGS(itr_regions);
1411         SAVE_REGS(dtr_regions);
1412         SAVE_REGS(tc_regions);
1413         SAVE_REGS(irq_check);
1414         SAVE_REGS(itc_check);
1415         SAVE_REGS(timer_check);
1416         SAVE_REGS(timer_pending);
1417         SAVE_REGS(last_itc);
1418         for (i = 0; i < 8; i++) {
1419                 regs->vrr[i] = vcpu->arch.vrr[i];
1420                 regs->ibr[i] = vcpu->arch.ibr[i];
1421                 regs->dbr[i] = vcpu->arch.dbr[i];
1422         }
1423         for (i = 0; i < 4; i++)
1424                 regs->insvc[i] = vcpu->arch.insvc[i];
1425         regs->saved_itc = vcpu->arch.itc_offset + ia64_getreg(_IA64_REG_AR_ITC);
1426         SAVE_REGS(xtp);
1427         SAVE_REGS(metaphysical_rr0);
1428         SAVE_REGS(metaphysical_rr4);
1429         SAVE_REGS(metaphysical_saved_rr0);
1430         SAVE_REGS(metaphysical_saved_rr4);
1431         SAVE_REGS(fp_psr);
1432         SAVE_REGS(saved_gp);
1433         vcpu_put(vcpu);
1434         r = 0;
1435 out:
1436         return r;
1437 }
1438
1439 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1440 {
1441
1442         hrtimer_cancel(&vcpu->arch.hlt_timer);
1443         kfree(vcpu->arch.apic);
1444 }
1445
1446
1447 long kvm_arch_vcpu_ioctl(struct file *filp,
1448                 unsigned int ioctl, unsigned long arg)
1449 {
1450         return -EINVAL;
1451 }
1452
1453 int kvm_arch_set_memory_region(struct kvm *kvm,
1454                 struct kvm_userspace_memory_region *mem,
1455                 struct kvm_memory_slot old,
1456                 int user_alloc)
1457 {
1458         unsigned long i;
1459         unsigned long pfn;
1460         int npages = mem->memory_size >> PAGE_SHIFT;
1461         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
1462         unsigned long base_gfn = memslot->base_gfn;
1463
1464         if (base_gfn + npages > (KVM_MAX_MEM_SIZE >> PAGE_SHIFT))
1465                 return -ENOMEM;
1466
1467         for (i = 0; i < npages; i++) {
1468                 pfn = gfn_to_pfn(kvm, base_gfn + i);
1469                 if (!kvm_is_mmio_pfn(pfn)) {
1470                         kvm_set_pmt_entry(kvm, base_gfn + i,
1471                                         pfn << PAGE_SHIFT,
1472                                 _PAGE_AR_RWX | _PAGE_MA_WB);
1473                         memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1474                 } else {
1475                         kvm_set_pmt_entry(kvm, base_gfn + i,
1476                                         GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1477                                         _PAGE_MA_UC);
1478                         memslot->rmap[i] = 0;
1479                         }
1480         }
1481
1482         return 0;
1483 }
1484
1485 void kvm_arch_flush_shadow(struct kvm *kvm)
1486 {
1487 }
1488
1489 long kvm_arch_dev_ioctl(struct file *filp,
1490                 unsigned int ioctl, unsigned long arg)
1491 {
1492         return -EINVAL;
1493 }
1494
1495 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1496 {
1497         kvm_vcpu_uninit(vcpu);
1498 }
1499
1500 static int vti_cpu_has_kvm_support(void)
1501 {
1502         long  avail = 1, status = 1, control = 1;
1503         long ret;
1504
1505         ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1506         if (ret)
1507                 goto out;
1508
1509         if (!(avail & PAL_PROC_VM_BIT))
1510                 goto out;
1511
1512         printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1513
1514         ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1515         if (ret)
1516                 goto out;
1517         printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1518
1519         if (!(vp_env_info & VP_OPCODE)) {
1520                 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1521                                 "vm_env_info:0x%lx\n", vp_env_info);
1522         }
1523
1524         return 1;
1525 out:
1526         return 0;
1527 }
1528
1529 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1530                                                 struct module *module)
1531 {
1532         unsigned long module_base;
1533         unsigned long vmm_size;
1534
1535         unsigned long vmm_offset, func_offset, fdesc_offset;
1536         struct fdesc *p_fdesc;
1537
1538         BUG_ON(!module);
1539
1540         if (!kvm_vmm_base) {
1541                 printk("kvm: kvm area hasn't been initilized yet!!\n");
1542                 return -EFAULT;
1543         }
1544
1545         /*Calculate new position of relocated vmm module.*/
1546         module_base = (unsigned long)module->module_core;
1547         vmm_size = module->core_size;
1548         if (unlikely(vmm_size > KVM_VMM_SIZE))
1549                 return -EFAULT;
1550
1551         memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1552         kvm_flush_icache(kvm_vmm_base, vmm_size);
1553
1554         /*Recalculate kvm_vmm_info based on new VMM*/
1555         vmm_offset = vmm_info->vmm_ivt - module_base;
1556         kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1557         printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1558                         kvm_vmm_info->vmm_ivt);
1559
1560         fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1561         kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1562                                                         fdesc_offset);
1563         func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1564         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1565         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1566         p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1567
1568         printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1569                         KVM_VMM_BASE+func_offset);
1570
1571         fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1572         kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1573                         fdesc_offset);
1574         func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1575         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1576         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1577         p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1578
1579         kvm_vmm_gp = p_fdesc->gp;
1580
1581         printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1582                                                 kvm_vmm_info->vmm_entry);
1583         printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1584                                                 KVM_VMM_BASE + func_offset);
1585
1586         return 0;
1587 }
1588
1589 int kvm_arch_init(void *opaque)
1590 {
1591         int r;
1592         struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1593
1594         if (!vti_cpu_has_kvm_support()) {
1595                 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1596                 r = -EOPNOTSUPP;
1597                 goto out;
1598         }
1599
1600         if (kvm_vmm_info) {
1601                 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1602                 r = -EEXIST;
1603                 goto out;
1604         }
1605
1606         r = -ENOMEM;
1607         kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1608         if (!kvm_vmm_info)
1609                 goto out;
1610
1611         if (kvm_alloc_vmm_area())
1612                 goto out_free0;
1613
1614         r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1615         if (r)
1616                 goto out_free1;
1617
1618         return 0;
1619
1620 out_free1:
1621         kvm_free_vmm_area();
1622 out_free0:
1623         kfree(kvm_vmm_info);
1624 out:
1625         return r;
1626 }
1627
1628 void kvm_arch_exit(void)
1629 {
1630         kvm_free_vmm_area();
1631         kfree(kvm_vmm_info);
1632         kvm_vmm_info = NULL;
1633 }
1634
1635 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1636                 struct kvm_dirty_log *log)
1637 {
1638         struct kvm_memory_slot *memslot;
1639         int r, i;
1640         long n, base;
1641         unsigned long *dirty_bitmap = (unsigned long *)(kvm->arch.vm_base +
1642                         offsetof(struct kvm_vm_data, kvm_mem_dirty_log));
1643
1644         r = -EINVAL;
1645         if (log->slot >= KVM_MEMORY_SLOTS)
1646                 goto out;
1647
1648         memslot = &kvm->memslots[log->slot];
1649         r = -ENOENT;
1650         if (!memslot->dirty_bitmap)
1651                 goto out;
1652
1653         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1654         base = memslot->base_gfn / BITS_PER_LONG;
1655
1656         for (i = 0; i < n/sizeof(long); ++i) {
1657                 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1658                 dirty_bitmap[base + i] = 0;
1659         }
1660         r = 0;
1661 out:
1662         return r;
1663 }
1664
1665 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1666                 struct kvm_dirty_log *log)
1667 {
1668         int r;
1669         int n;
1670         struct kvm_memory_slot *memslot;
1671         int is_dirty = 0;
1672
1673         spin_lock(&kvm->arch.dirty_log_lock);
1674
1675         r = kvm_ia64_sync_dirty_log(kvm, log);
1676         if (r)
1677                 goto out;
1678
1679         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1680         if (r)
1681                 goto out;
1682
1683         /* If nothing is dirty, don't bother messing with page tables. */
1684         if (is_dirty) {
1685                 kvm_flush_remote_tlbs(kvm);
1686                 memslot = &kvm->memslots[log->slot];
1687                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1688                 memset(memslot->dirty_bitmap, 0, n);
1689         }
1690         r = 0;
1691 out:
1692         spin_unlock(&kvm->arch.dirty_log_lock);
1693         return r;
1694 }
1695
1696 int kvm_arch_hardware_setup(void)
1697 {
1698         return 0;
1699 }
1700
1701 void kvm_arch_hardware_unsetup(void)
1702 {
1703 }
1704
1705 static void vcpu_kick_intr(void *info)
1706 {
1707 #ifdef DEBUG
1708         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
1709         printk(KERN_DEBUG"vcpu_kick_intr %p \n", vcpu);
1710 #endif
1711 }
1712
1713 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1714 {
1715         int ipi_pcpu = vcpu->cpu;
1716         int cpu = get_cpu();
1717
1718         if (waitqueue_active(&vcpu->wq))
1719                 wake_up_interruptible(&vcpu->wq);
1720
1721         if (vcpu->guest_mode && cpu != ipi_pcpu)
1722                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
1723         put_cpu();
1724 }
1725
1726 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
1727 {
1728
1729         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1730
1731         if (!test_and_set_bit(vec, &vpd->irr[0])) {
1732                 vcpu->arch.irq_new_pending = 1;
1733                 kvm_vcpu_kick(vcpu);
1734                 return 1;
1735         }
1736         return 0;
1737 }
1738
1739 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1740 {
1741         return apic->vcpu->vcpu_id == dest;
1742 }
1743
1744 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1745 {
1746         return 0;
1747 }
1748
1749 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
1750                                        unsigned long bitmap)
1751 {
1752         struct kvm_vcpu *lvcpu = kvm->vcpus[0];
1753         int i;
1754
1755         for (i = 1; i < KVM_MAX_VCPUS; i++) {
1756                 if (!kvm->vcpus[i])
1757                         continue;
1758                 if (lvcpu->arch.xtp > kvm->vcpus[i]->arch.xtp)
1759                         lvcpu = kvm->vcpus[i];
1760         }
1761
1762         return lvcpu;
1763 }
1764
1765 static int find_highest_bits(int *dat)
1766 {
1767         u32  bits, bitnum;
1768         int i;
1769
1770         /* loop for all 256 bits */
1771         for (i = 7; i >= 0 ; i--) {
1772                 bits = dat[i];
1773                 if (bits) {
1774                         bitnum = fls(bits);
1775                         return i * 32 + bitnum - 1;
1776                 }
1777         }
1778
1779         return -1;
1780 }
1781
1782 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1783 {
1784     struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1785
1786     if (vpd->irr[0] & (1UL << NMI_VECTOR))
1787                 return NMI_VECTOR;
1788     if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1789                 return ExtINT_VECTOR;
1790
1791     return find_highest_bits((int *)&vpd->irr[0]);
1792 }
1793
1794 int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
1795 {
1796         if (kvm_highest_pending_irq(vcpu) != -1)
1797                 return 1;
1798         return 0;
1799 }
1800
1801 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1802 {
1803         return vcpu->arch.timer_fired;
1804 }
1805
1806 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1807 {
1808         return gfn;
1809 }
1810
1811 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1812 {
1813         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE;
1814 }
1815
1816 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1817                                     struct kvm_mp_state *mp_state)
1818 {
1819         vcpu_load(vcpu);
1820         mp_state->mp_state = vcpu->arch.mp_state;
1821         vcpu_put(vcpu);
1822         return 0;
1823 }
1824
1825 static int vcpu_reset(struct kvm_vcpu *vcpu)
1826 {
1827         int r;
1828         long psr;
1829         local_irq_save(psr);
1830         r = kvm_insert_vmm_mapping(vcpu);
1831         if (r)
1832                 goto fail;
1833
1834         vcpu->arch.launched = 0;
1835         kvm_arch_vcpu_uninit(vcpu);
1836         r = kvm_arch_vcpu_init(vcpu);
1837         if (r)
1838                 goto fail;
1839
1840         kvm_purge_vmm_mapping(vcpu);
1841         r = 0;
1842 fail:
1843         local_irq_restore(psr);
1844         return r;
1845 }
1846
1847 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1848                                     struct kvm_mp_state *mp_state)
1849 {
1850         int r = 0;
1851
1852         vcpu_load(vcpu);
1853         vcpu->arch.mp_state = mp_state->mp_state;
1854         if (vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)
1855                 r = vcpu_reset(vcpu);
1856         vcpu_put(vcpu);
1857         return r;
1858 }