0f5ebd94843725e11277603dcb87924fd0bd2df0
[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/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         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
835         int i;
836
837         vcpu_load(vcpu);
838
839         for (i = 0; i < 16; i++) {
840                 vpd->vgr[i] = regs->vpd.vgr[i];
841                 vpd->vbgr[i] = regs->vpd.vbgr[i];
842         }
843         for (i = 0; i < 128; i++)
844                 vpd->vcr[i] = regs->vpd.vcr[i];
845         vpd->vhpi = regs->vpd.vhpi;
846         vpd->vnat = regs->vpd.vnat;
847         vpd->vbnat = regs->vpd.vbnat;
848         vpd->vpsr = regs->vpd.vpsr;
849
850         vpd->vpr = regs->vpd.vpr;
851
852         memcpy(&vcpu->arch.guest, &regs->saved_guest, sizeof(union context));
853
854         RESTORE_REGS(mp_state);
855         RESTORE_REGS(vmm_rr);
856         memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
857         memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
858         RESTORE_REGS(itr_regions);
859         RESTORE_REGS(dtr_regions);
860         RESTORE_REGS(tc_regions);
861         RESTORE_REGS(irq_check);
862         RESTORE_REGS(itc_check);
863         RESTORE_REGS(timer_check);
864         RESTORE_REGS(timer_pending);
865         RESTORE_REGS(last_itc);
866         for (i = 0; i < 8; i++) {
867                 vcpu->arch.vrr[i] = regs->vrr[i];
868                 vcpu->arch.ibr[i] = regs->ibr[i];
869                 vcpu->arch.dbr[i] = regs->dbr[i];
870         }
871         for (i = 0; i < 4; i++)
872                 vcpu->arch.insvc[i] = regs->insvc[i];
873         RESTORE_REGS(xtp);
874         RESTORE_REGS(metaphysical_rr0);
875         RESTORE_REGS(metaphysical_rr4);
876         RESTORE_REGS(metaphysical_saved_rr0);
877         RESTORE_REGS(metaphysical_saved_rr4);
878         RESTORE_REGS(fp_psr);
879         RESTORE_REGS(saved_gp);
880
881         vcpu->arch.irq_new_pending = 1;
882         vcpu->arch.itc_offset = regs->saved_itc - ia64_getreg(_IA64_REG_AR_ITC);
883         set_bit(KVM_REQ_RESUME, &vcpu->requests);
884
885         vcpu_put(vcpu);
886
887         return 0;
888 }
889
890 long kvm_arch_vm_ioctl(struct file *filp,
891                 unsigned int ioctl, unsigned long arg)
892 {
893         struct kvm *kvm = filp->private_data;
894         void __user *argp = (void __user *)arg;
895         int r = -EINVAL;
896
897         switch (ioctl) {
898         case KVM_SET_MEMORY_REGION: {
899                 struct kvm_memory_region kvm_mem;
900                 struct kvm_userspace_memory_region kvm_userspace_mem;
901
902                 r = -EFAULT;
903                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
904                         goto out;
905                 kvm_userspace_mem.slot = kvm_mem.slot;
906                 kvm_userspace_mem.flags = kvm_mem.flags;
907                 kvm_userspace_mem.guest_phys_addr =
908                                         kvm_mem.guest_phys_addr;
909                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
910                 r = kvm_vm_ioctl_set_memory_region(kvm,
911                                         &kvm_userspace_mem, 0);
912                 if (r)
913                         goto out;
914                 break;
915                 }
916         case KVM_CREATE_IRQCHIP:
917                 r = -EFAULT;
918                 r = kvm_ioapic_init(kvm);
919                 if (r)
920                         goto out;
921                 break;
922         case KVM_IRQ_LINE: {
923                 struct kvm_irq_level irq_event;
924
925                 r = -EFAULT;
926                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
927                         goto out;
928                 if (irqchip_in_kernel(kvm)) {
929                         mutex_lock(&kvm->lock);
930                         kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
931                                     irq_event.irq, irq_event.level);
932                         mutex_unlock(&kvm->lock);
933                         r = 0;
934                 }
935                 break;
936                 }
937         case KVM_GET_IRQCHIP: {
938                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
939                 struct kvm_irqchip chip;
940
941                 r = -EFAULT;
942                 if (copy_from_user(&chip, argp, sizeof chip))
943                                 goto out;
944                 r = -ENXIO;
945                 if (!irqchip_in_kernel(kvm))
946                         goto out;
947                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
948                 if (r)
949                         goto out;
950                 r = -EFAULT;
951                 if (copy_to_user(argp, &chip, sizeof chip))
952                                 goto out;
953                 r = 0;
954                 break;
955                 }
956         case KVM_SET_IRQCHIP: {
957                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
958                 struct kvm_irqchip chip;
959
960                 r = -EFAULT;
961                 if (copy_from_user(&chip, argp, sizeof chip))
962                                 goto out;
963                 r = -ENXIO;
964                 if (!irqchip_in_kernel(kvm))
965                         goto out;
966                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
967                 if (r)
968                         goto out;
969                 r = 0;
970                 break;
971                 }
972         default:
973                 ;
974         }
975 out:
976         return r;
977 }
978
979 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
980                 struct kvm_sregs *sregs)
981 {
982         return -EINVAL;
983 }
984
985 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
986                 struct kvm_sregs *sregs)
987 {
988         return -EINVAL;
989
990 }
991 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
992                 struct kvm_translation *tr)
993 {
994
995         return -EINVAL;
996 }
997
998 static int kvm_alloc_vmm_area(void)
999 {
1000         if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1001                 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1002                                 get_order(KVM_VMM_SIZE));
1003                 if (!kvm_vmm_base)
1004                         return -ENOMEM;
1005
1006                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1007                 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1008
1009                 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1010                                 kvm_vmm_base, kvm_vm_buffer);
1011         }
1012
1013         return 0;
1014 }
1015
1016 static void kvm_free_vmm_area(void)
1017 {
1018         if (kvm_vmm_base) {
1019                 /*Zero this area before free to avoid bits leak!!*/
1020                 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1021                 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1022                 kvm_vmm_base  = 0;
1023                 kvm_vm_buffer = 0;
1024                 kvm_vsa_base = 0;
1025         }
1026 }
1027
1028 static void vti_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1029 {
1030 }
1031
1032 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1033 {
1034         int i;
1035         union cpuid3_t cpuid3;
1036         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1037
1038         if (IS_ERR(vpd))
1039                 return PTR_ERR(vpd);
1040
1041         /* CPUID init */
1042         for (i = 0; i < 5; i++)
1043                 vpd->vcpuid[i] = ia64_get_cpuid(i);
1044
1045         /* Limit the CPUID number to 5 */
1046         cpuid3.value = vpd->vcpuid[3];
1047         cpuid3.number = 4;      /* 5 - 1 */
1048         vpd->vcpuid[3] = cpuid3.value;
1049
1050         /*Set vac and vdc fields*/
1051         vpd->vac.a_from_int_cr = 1;
1052         vpd->vac.a_to_int_cr = 1;
1053         vpd->vac.a_from_psr = 1;
1054         vpd->vac.a_from_cpuid = 1;
1055         vpd->vac.a_cover = 1;
1056         vpd->vac.a_bsw = 1;
1057         vpd->vac.a_int = 1;
1058         vpd->vdc.d_vmsw = 1;
1059
1060         /*Set virtual buffer*/
1061         vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1062
1063         return 0;
1064 }
1065
1066 static int vti_create_vp(struct kvm_vcpu *vcpu)
1067 {
1068         long ret;
1069         struct vpd *vpd = vcpu->arch.vpd;
1070         unsigned long  vmm_ivt;
1071
1072         vmm_ivt = kvm_vmm_info->vmm_ivt;
1073
1074         printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1075
1076         ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1077
1078         if (ret) {
1079                 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1080                 return -EINVAL;
1081         }
1082         return 0;
1083 }
1084
1085 static void init_ptce_info(struct kvm_vcpu *vcpu)
1086 {
1087         ia64_ptce_info_t ptce = {0};
1088
1089         ia64_get_ptce(&ptce);
1090         vcpu->arch.ptce_base = ptce.base;
1091         vcpu->arch.ptce_count[0] = ptce.count[0];
1092         vcpu->arch.ptce_count[1] = ptce.count[1];
1093         vcpu->arch.ptce_stride[0] = ptce.stride[0];
1094         vcpu->arch.ptce_stride[1] = ptce.stride[1];
1095 }
1096
1097 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1098 {
1099         struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1100
1101         if (hrtimer_cancel(p_ht))
1102                 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1103 }
1104
1105 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1106 {
1107         struct kvm_vcpu *vcpu;
1108         wait_queue_head_t *q;
1109
1110         vcpu  = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1111         q = &vcpu->wq;
1112
1113         if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1114                 goto out;
1115
1116         if (waitqueue_active(q))
1117                 wake_up_interruptible(q);
1118
1119 out:
1120         vcpu->arch.timer_fired = 1;
1121         vcpu->arch.timer_check = 1;
1122         return HRTIMER_NORESTART;
1123 }
1124
1125 #define PALE_RESET_ENTRY    0x80000000ffffffb0UL
1126
1127 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1128 {
1129         struct kvm_vcpu *v;
1130         int r;
1131         int i;
1132         long itc_offset;
1133         struct kvm *kvm = vcpu->kvm;
1134         struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1135
1136         union context *p_ctx = &vcpu->arch.guest;
1137         struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1138
1139         /*Init vcpu context for first run.*/
1140         if (IS_ERR(vmm_vcpu))
1141                 return PTR_ERR(vmm_vcpu);
1142
1143         if (vcpu->vcpu_id == 0) {
1144                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1145
1146                 /*Set entry address for first run.*/
1147                 regs->cr_iip = PALE_RESET_ENTRY;
1148
1149                 /*Initialize itc offset for vcpus*/
1150                 itc_offset = 0UL - ia64_getreg(_IA64_REG_AR_ITC);
1151                 for (i = 0; i < KVM_MAX_VCPUS; i++) {
1152                         v = (struct kvm_vcpu *)((char *)vcpu +
1153                                         sizeof(struct kvm_vcpu_data) * i);
1154                         v->arch.itc_offset = itc_offset;
1155                         v->arch.last_itc = 0;
1156                 }
1157         } else
1158                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1159
1160         r = -ENOMEM;
1161         vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1162         if (!vcpu->arch.apic)
1163                 goto out;
1164         vcpu->arch.apic->vcpu = vcpu;
1165
1166         p_ctx->gr[1] = 0;
1167         p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + KVM_STK_OFFSET);
1168         p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1169         p_ctx->psr = 0x1008522000UL;
1170         p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1171         p_ctx->caller_unat = 0;
1172         p_ctx->pr = 0x0;
1173         p_ctx->ar[36] = 0x0; /*unat*/
1174         p_ctx->ar[19] = 0x0; /*rnat*/
1175         p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1176                                 ((sizeof(struct kvm_vcpu)+15) & ~15);
1177         p_ctx->ar[64] = 0x0; /*pfs*/
1178         p_ctx->cr[0] = 0x7e04UL;
1179         p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1180         p_ctx->cr[8] = 0x3c;
1181
1182         /*Initilize region register*/
1183         p_ctx->rr[0] = 0x30;
1184         p_ctx->rr[1] = 0x30;
1185         p_ctx->rr[2] = 0x30;
1186         p_ctx->rr[3] = 0x30;
1187         p_ctx->rr[4] = 0x30;
1188         p_ctx->rr[5] = 0x30;
1189         p_ctx->rr[7] = 0x30;
1190
1191         /*Initilize branch register 0*/
1192         p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1193
1194         vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1195         vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1196         vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1197
1198         hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1199         vcpu->arch.hlt_timer.function = hlt_timer_fn;
1200
1201         vcpu->arch.last_run_cpu = -1;
1202         vcpu->arch.vpd = (struct vpd *)VPD_BASE(vcpu->vcpu_id);
1203         vcpu->arch.vsa_base = kvm_vsa_base;
1204         vcpu->arch.__gp = kvm_vmm_gp;
1205         vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1206         vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_BASE(vcpu->vcpu_id);
1207         vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_BASE(vcpu->vcpu_id);
1208         init_ptce_info(vcpu);
1209
1210         r = 0;
1211 out:
1212         return r;
1213 }
1214
1215 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1216 {
1217         unsigned long psr;
1218         int r;
1219
1220         local_irq_save(psr);
1221         r = kvm_insert_vmm_mapping(vcpu);
1222         if (r)
1223                 goto fail;
1224         r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1225         if (r)
1226                 goto fail;
1227
1228         r = vti_init_vpd(vcpu);
1229         if (r) {
1230                 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1231                 goto uninit;
1232         }
1233
1234         r = vti_create_vp(vcpu);
1235         if (r)
1236                 goto uninit;
1237
1238         kvm_purge_vmm_mapping(vcpu);
1239         local_irq_restore(psr);
1240
1241         return 0;
1242 uninit:
1243         kvm_vcpu_uninit(vcpu);
1244 fail:
1245         local_irq_restore(psr);
1246         return r;
1247 }
1248
1249 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1250                 unsigned int id)
1251 {
1252         struct kvm_vcpu *vcpu;
1253         unsigned long vm_base = kvm->arch.vm_base;
1254         int r;
1255         int cpu;
1256
1257         BUG_ON(sizeof(struct kvm_vcpu) > VCPU_STRUCT_SIZE/2);
1258
1259         r = -EINVAL;
1260         if (id >= KVM_MAX_VCPUS) {
1261                 printk(KERN_ERR"kvm: Can't configure vcpus > %ld",
1262                                 KVM_MAX_VCPUS);
1263                 goto fail;
1264         }
1265
1266         r = -ENOMEM;
1267         if (!vm_base) {
1268                 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1269                 goto fail;
1270         }
1271         vcpu = (struct kvm_vcpu *)(vm_base + offsetof(struct kvm_vm_data,
1272                                         vcpu_data[id].vcpu_struct));
1273         vcpu->kvm = kvm;
1274
1275         cpu = get_cpu();
1276         vti_vcpu_load(vcpu, cpu);
1277         r = vti_vcpu_setup(vcpu, id);
1278         put_cpu();
1279
1280         if (r) {
1281                 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1282                 goto fail;
1283         }
1284
1285         return vcpu;
1286 fail:
1287         return ERR_PTR(r);
1288 }
1289
1290 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1291 {
1292         return 0;
1293 }
1294
1295 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1296 {
1297         return -EINVAL;
1298 }
1299
1300 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1301 {
1302         return -EINVAL;
1303 }
1304
1305 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1306                 struct kvm_debug_guest *dbg)
1307 {
1308         return -EINVAL;
1309 }
1310
1311 static void free_kvm(struct kvm *kvm)
1312 {
1313         unsigned long vm_base = kvm->arch.vm_base;
1314
1315         if (vm_base) {
1316                 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1317                 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1318         }
1319
1320 }
1321
1322 static void kvm_release_vm_pages(struct kvm *kvm)
1323 {
1324         struct kvm_memory_slot *memslot;
1325         int i, j;
1326         unsigned long base_gfn;
1327
1328         for (i = 0; i < kvm->nmemslots; i++) {
1329                 memslot = &kvm->memslots[i];
1330                 base_gfn = memslot->base_gfn;
1331
1332                 for (j = 0; j < memslot->npages; j++) {
1333                         if (memslot->rmap[j])
1334                                 put_page((struct page *)memslot->rmap[j]);
1335                 }
1336         }
1337 }
1338
1339 void kvm_arch_destroy_vm(struct kvm *kvm)
1340 {
1341         kvm_iommu_unmap_guest(kvm);
1342 #ifdef  KVM_CAP_DEVICE_ASSIGNMENT
1343         kvm_free_all_assigned_devices(kvm);
1344 #endif
1345         kfree(kvm->arch.vioapic);
1346         kvm_release_vm_pages(kvm);
1347         kvm_free_physmem(kvm);
1348         free_kvm(kvm);
1349 }
1350
1351 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1352 {
1353 }
1354
1355 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1356 {
1357         if (cpu != vcpu->cpu) {
1358                 vcpu->cpu = cpu;
1359                 if (vcpu->arch.ht_active)
1360                         kvm_migrate_hlt_timer(vcpu);
1361         }
1362 }
1363
1364 #define SAVE_REGS(_x)   regs->_x = vcpu->arch._x
1365
1366 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1367 {
1368         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1369         int i;
1370
1371         vcpu_load(vcpu);
1372
1373         for (i = 0; i < 16; i++) {
1374                 regs->vpd.vgr[i] = vpd->vgr[i];
1375                 regs->vpd.vbgr[i] = vpd->vbgr[i];
1376         }
1377         for (i = 0; i < 128; i++)
1378                 regs->vpd.vcr[i] = vpd->vcr[i];
1379         regs->vpd.vhpi = vpd->vhpi;
1380         regs->vpd.vnat = vpd->vnat;
1381         regs->vpd.vbnat = vpd->vbnat;
1382         regs->vpd.vpsr = vpd->vpsr;
1383         regs->vpd.vpr = vpd->vpr;
1384
1385         memcpy(&regs->saved_guest, &vcpu->arch.guest, sizeof(union context));
1386
1387         SAVE_REGS(mp_state);
1388         SAVE_REGS(vmm_rr);
1389         memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1390         memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1391         SAVE_REGS(itr_regions);
1392         SAVE_REGS(dtr_regions);
1393         SAVE_REGS(tc_regions);
1394         SAVE_REGS(irq_check);
1395         SAVE_REGS(itc_check);
1396         SAVE_REGS(timer_check);
1397         SAVE_REGS(timer_pending);
1398         SAVE_REGS(last_itc);
1399         for (i = 0; i < 8; i++) {
1400                 regs->vrr[i] = vcpu->arch.vrr[i];
1401                 regs->ibr[i] = vcpu->arch.ibr[i];
1402                 regs->dbr[i] = vcpu->arch.dbr[i];
1403         }
1404         for (i = 0; i < 4; i++)
1405                 regs->insvc[i] = vcpu->arch.insvc[i];
1406         regs->saved_itc = vcpu->arch.itc_offset + ia64_getreg(_IA64_REG_AR_ITC);
1407         SAVE_REGS(xtp);
1408         SAVE_REGS(metaphysical_rr0);
1409         SAVE_REGS(metaphysical_rr4);
1410         SAVE_REGS(metaphysical_saved_rr0);
1411         SAVE_REGS(metaphysical_saved_rr4);
1412         SAVE_REGS(fp_psr);
1413         SAVE_REGS(saved_gp);
1414
1415         vcpu_put(vcpu);
1416         return 0;
1417 }
1418
1419 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1420 {
1421
1422         hrtimer_cancel(&vcpu->arch.hlt_timer);
1423         kfree(vcpu->arch.apic);
1424 }
1425
1426
1427 long kvm_arch_vcpu_ioctl(struct file *filp,
1428                 unsigned int ioctl, unsigned long arg)
1429 {
1430         return -EINVAL;
1431 }
1432
1433 int kvm_arch_set_memory_region(struct kvm *kvm,
1434                 struct kvm_userspace_memory_region *mem,
1435                 struct kvm_memory_slot old,
1436                 int user_alloc)
1437 {
1438         unsigned long i;
1439         unsigned long pfn;
1440         int npages = mem->memory_size >> PAGE_SHIFT;
1441         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
1442         unsigned long base_gfn = memslot->base_gfn;
1443
1444         if (base_gfn + npages > (KVM_MAX_MEM_SIZE >> PAGE_SHIFT))
1445                 return -ENOMEM;
1446
1447         for (i = 0; i < npages; i++) {
1448                 pfn = gfn_to_pfn(kvm, base_gfn + i);
1449                 if (!kvm_is_mmio_pfn(pfn)) {
1450                         kvm_set_pmt_entry(kvm, base_gfn + i,
1451                                         pfn << PAGE_SHIFT,
1452                                 _PAGE_AR_RWX | _PAGE_MA_WB);
1453                         memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1454                 } else {
1455                         kvm_set_pmt_entry(kvm, base_gfn + i,
1456                                         GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1457                                         _PAGE_MA_UC);
1458                         memslot->rmap[i] = 0;
1459                         }
1460         }
1461
1462         return 0;
1463 }
1464
1465 void kvm_arch_flush_shadow(struct kvm *kvm)
1466 {
1467 }
1468
1469 long kvm_arch_dev_ioctl(struct file *filp,
1470                 unsigned int ioctl, unsigned long arg)
1471 {
1472         return -EINVAL;
1473 }
1474
1475 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1476 {
1477         kvm_vcpu_uninit(vcpu);
1478 }
1479
1480 static int vti_cpu_has_kvm_support(void)
1481 {
1482         long  avail = 1, status = 1, control = 1;
1483         long ret;
1484
1485         ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1486         if (ret)
1487                 goto out;
1488
1489         if (!(avail & PAL_PROC_VM_BIT))
1490                 goto out;
1491
1492         printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1493
1494         ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1495         if (ret)
1496                 goto out;
1497         printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1498
1499         if (!(vp_env_info & VP_OPCODE)) {
1500                 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1501                                 "vm_env_info:0x%lx\n", vp_env_info);
1502         }
1503
1504         return 1;
1505 out:
1506         return 0;
1507 }
1508
1509 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1510                                                 struct module *module)
1511 {
1512         unsigned long module_base;
1513         unsigned long vmm_size;
1514
1515         unsigned long vmm_offset, func_offset, fdesc_offset;
1516         struct fdesc *p_fdesc;
1517
1518         BUG_ON(!module);
1519
1520         if (!kvm_vmm_base) {
1521                 printk("kvm: kvm area hasn't been initilized yet!!\n");
1522                 return -EFAULT;
1523         }
1524
1525         /*Calculate new position of relocated vmm module.*/
1526         module_base = (unsigned long)module->module_core;
1527         vmm_size = module->core_size;
1528         if (unlikely(vmm_size > KVM_VMM_SIZE))
1529                 return -EFAULT;
1530
1531         memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1532         kvm_flush_icache(kvm_vmm_base, vmm_size);
1533
1534         /*Recalculate kvm_vmm_info based on new VMM*/
1535         vmm_offset = vmm_info->vmm_ivt - module_base;
1536         kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1537         printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1538                         kvm_vmm_info->vmm_ivt);
1539
1540         fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1541         kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1542                                                         fdesc_offset);
1543         func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1544         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1545         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1546         p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1547
1548         printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1549                         KVM_VMM_BASE+func_offset);
1550
1551         fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1552         kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1553                         fdesc_offset);
1554         func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1555         p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1556         p_fdesc->ip = KVM_VMM_BASE + func_offset;
1557         p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1558
1559         kvm_vmm_gp = p_fdesc->gp;
1560
1561         printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1562                                                 kvm_vmm_info->vmm_entry);
1563         printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1564                                                 KVM_VMM_BASE + func_offset);
1565
1566         return 0;
1567 }
1568
1569 int kvm_arch_init(void *opaque)
1570 {
1571         int r;
1572         struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1573
1574         if (!vti_cpu_has_kvm_support()) {
1575                 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1576                 r = -EOPNOTSUPP;
1577                 goto out;
1578         }
1579
1580         if (kvm_vmm_info) {
1581                 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1582                 r = -EEXIST;
1583                 goto out;
1584         }
1585
1586         r = -ENOMEM;
1587         kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1588         if (!kvm_vmm_info)
1589                 goto out;
1590
1591         if (kvm_alloc_vmm_area())
1592                 goto out_free0;
1593
1594         r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1595         if (r)
1596                 goto out_free1;
1597
1598         return 0;
1599
1600 out_free1:
1601         kvm_free_vmm_area();
1602 out_free0:
1603         kfree(kvm_vmm_info);
1604 out:
1605         return r;
1606 }
1607
1608 void kvm_arch_exit(void)
1609 {
1610         kvm_free_vmm_area();
1611         kfree(kvm_vmm_info);
1612         kvm_vmm_info = NULL;
1613 }
1614
1615 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1616                 struct kvm_dirty_log *log)
1617 {
1618         struct kvm_memory_slot *memslot;
1619         int r, i;
1620         long n, base;
1621         unsigned long *dirty_bitmap = (unsigned long *)(kvm->arch.vm_base +
1622                         offsetof(struct kvm_vm_data, kvm_mem_dirty_log));
1623
1624         r = -EINVAL;
1625         if (log->slot >= KVM_MEMORY_SLOTS)
1626                 goto out;
1627
1628         memslot = &kvm->memslots[log->slot];
1629         r = -ENOENT;
1630         if (!memslot->dirty_bitmap)
1631                 goto out;
1632
1633         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1634         base = memslot->base_gfn / BITS_PER_LONG;
1635
1636         for (i = 0; i < n/sizeof(long); ++i) {
1637                 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1638                 dirty_bitmap[base + i] = 0;
1639         }
1640         r = 0;
1641 out:
1642         return r;
1643 }
1644
1645 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1646                 struct kvm_dirty_log *log)
1647 {
1648         int r;
1649         int n;
1650         struct kvm_memory_slot *memslot;
1651         int is_dirty = 0;
1652
1653         spin_lock(&kvm->arch.dirty_log_lock);
1654
1655         r = kvm_ia64_sync_dirty_log(kvm, log);
1656         if (r)
1657                 goto out;
1658
1659         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1660         if (r)
1661                 goto out;
1662
1663         /* If nothing is dirty, don't bother messing with page tables. */
1664         if (is_dirty) {
1665                 kvm_flush_remote_tlbs(kvm);
1666                 memslot = &kvm->memslots[log->slot];
1667                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1668                 memset(memslot->dirty_bitmap, 0, n);
1669         }
1670         r = 0;
1671 out:
1672         spin_unlock(&kvm->arch.dirty_log_lock);
1673         return r;
1674 }
1675
1676 int kvm_arch_hardware_setup(void)
1677 {
1678         return 0;
1679 }
1680
1681 void kvm_arch_hardware_unsetup(void)
1682 {
1683 }
1684
1685 static void vcpu_kick_intr(void *info)
1686 {
1687 #ifdef DEBUG
1688         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
1689         printk(KERN_DEBUG"vcpu_kick_intr %p \n", vcpu);
1690 #endif
1691 }
1692
1693 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1694 {
1695         int ipi_pcpu = vcpu->cpu;
1696         int cpu = get_cpu();
1697
1698         if (waitqueue_active(&vcpu->wq))
1699                 wake_up_interruptible(&vcpu->wq);
1700
1701         if (vcpu->guest_mode && cpu != ipi_pcpu)
1702                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);
1703         put_cpu();
1704 }
1705
1706 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
1707 {
1708
1709         struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1710
1711         if (!test_and_set_bit(vec, &vpd->irr[0])) {
1712                 vcpu->arch.irq_new_pending = 1;
1713                 kvm_vcpu_kick(vcpu);
1714                 return 1;
1715         }
1716         return 0;
1717 }
1718
1719 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1720 {
1721         return apic->vcpu->vcpu_id == dest;
1722 }
1723
1724 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1725 {
1726         return 0;
1727 }
1728
1729 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
1730                                        unsigned long bitmap)
1731 {
1732         struct kvm_vcpu *lvcpu = kvm->vcpus[0];
1733         int i;
1734
1735         for (i = 1; i < KVM_MAX_VCPUS; i++) {
1736                 if (!kvm->vcpus[i])
1737                         continue;
1738                 if (lvcpu->arch.xtp > kvm->vcpus[i]->arch.xtp)
1739                         lvcpu = kvm->vcpus[i];
1740         }
1741
1742         return lvcpu;
1743 }
1744
1745 static int find_highest_bits(int *dat)
1746 {
1747         u32  bits, bitnum;
1748         int i;
1749
1750         /* loop for all 256 bits */
1751         for (i = 7; i >= 0 ; i--) {
1752                 bits = dat[i];
1753                 if (bits) {
1754                         bitnum = fls(bits);
1755                         return i * 32 + bitnum - 1;
1756                 }
1757         }
1758
1759         return -1;
1760 }
1761
1762 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1763 {
1764     struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1765
1766     if (vpd->irr[0] & (1UL << NMI_VECTOR))
1767                 return NMI_VECTOR;
1768     if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1769                 return ExtINT_VECTOR;
1770
1771     return find_highest_bits((int *)&vpd->irr[0]);
1772 }
1773
1774 int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
1775 {
1776         if (kvm_highest_pending_irq(vcpu) != -1)
1777                 return 1;
1778         return 0;
1779 }
1780
1781 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1782 {
1783         return vcpu->arch.timer_fired;
1784 }
1785
1786 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1787 {
1788         return gfn;
1789 }
1790
1791 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1792 {
1793         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE;
1794 }
1795
1796 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1797                                     struct kvm_mp_state *mp_state)
1798 {
1799         vcpu_load(vcpu);
1800         mp_state->mp_state = vcpu->arch.mp_state;
1801         vcpu_put(vcpu);
1802         return 0;
1803 }
1804
1805 static int vcpu_reset(struct kvm_vcpu *vcpu)
1806 {
1807         int r;
1808         long psr;
1809         local_irq_save(psr);
1810         r = kvm_insert_vmm_mapping(vcpu);
1811         if (r)
1812                 goto fail;
1813
1814         vcpu->arch.launched = 0;
1815         kvm_arch_vcpu_uninit(vcpu);
1816         r = kvm_arch_vcpu_init(vcpu);
1817         if (r)
1818                 goto fail;
1819
1820         kvm_purge_vmm_mapping(vcpu);
1821         r = 0;
1822 fail:
1823         local_irq_restore(psr);
1824         return r;
1825 }
1826
1827 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1828                                     struct kvm_mp_state *mp_state)
1829 {
1830         int r = 0;
1831
1832         vcpu_load(vcpu);
1833         vcpu->arch.mp_state = mp_state->mp_state;
1834         if (vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)
1835                 r = vcpu_reset(vcpu);
1836         vcpu_put(vcpu);
1837         return r;
1838 }