Merge tag 'v3.3.7' into 3.3/master
[zen-stable.git] / arch / ia64 / kvm / kvm-ia64.c
blob8c25855f73a826635c562ba03c00fd73e4aa4998
1 /*
2 * kvm_ia64.c: Basic KVM suppport On Itanium series processors
5 * Copyright (C) 2007, Intel Corporation.
6 * Xiantao Zhang (xiantao.zhang@intel.com)
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.
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.
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.
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/fs.h>
27 #include <linux/slab.h>
28 #include <linux/smp.h>
29 #include <linux/kvm_host.h>
30 #include <linux/kvm.h>
31 #include <linux/bitops.h>
32 #include <linux/hrtimer.h>
33 #include <linux/uaccess.h>
34 #include <linux/iommu.h>
35 #include <linux/intel-iommu.h>
36 #include <linux/pci.h>
38 #include <asm/pgtable.h>
39 #include <asm/gcc_intrin.h>
40 #include <asm/pal.h>
41 #include <asm/cacheflush.h>
42 #include <asm/div64.h>
43 #include <asm/tlb.h>
44 #include <asm/elf.h>
45 #include <asm/sn/addrs.h>
46 #include <asm/sn/clksupport.h>
47 #include <asm/sn/shub_mmr.h>
49 #include "misc.h"
50 #include "vti.h"
51 #include "iodev.h"
52 #include "ioapic.h"
53 #include "lapic.h"
54 #include "irq.h"
56 static unsigned long kvm_vmm_base;
57 static unsigned long kvm_vsa_base;
58 static unsigned long kvm_vm_buffer;
59 static unsigned long kvm_vm_buffer_size;
60 unsigned long kvm_vmm_gp;
62 static long vp_env_info;
64 static struct kvm_vmm_info *kvm_vmm_info;
66 static DEFINE_PER_CPU(struct kvm_vcpu *, last_vcpu);
68 struct kvm_stats_debugfs_item debugfs_entries[] = {
69 { NULL }
72 static unsigned long kvm_get_itc(struct kvm_vcpu *vcpu)
74 #if defined(CONFIG_IA64_SGI_SN2) || defined(CONFIG_IA64_GENERIC)
75 if (vcpu->kvm->arch.is_sn2)
76 return rtc_time();
77 else
78 #endif
79 return ia64_getreg(_IA64_REG_AR_ITC);
82 static void kvm_flush_icache(unsigned long start, unsigned long len)
84 int l;
86 for (l = 0; l < (len + 32); l += 32)
87 ia64_fc((void *)(start + l));
89 ia64_sync_i();
90 ia64_srlz_i();
93 static void kvm_flush_tlb_all(void)
95 unsigned long i, j, count0, count1, stride0, stride1, addr;
96 long flags;
98 addr = local_cpu_data->ptce_base;
99 count0 = local_cpu_data->ptce_count[0];
100 count1 = local_cpu_data->ptce_count[1];
101 stride0 = local_cpu_data->ptce_stride[0];
102 stride1 = local_cpu_data->ptce_stride[1];
104 local_irq_save(flags);
105 for (i = 0; i < count0; ++i) {
106 for (j = 0; j < count1; ++j) {
107 ia64_ptce(addr);
108 addr += stride1;
110 addr += stride0;
112 local_irq_restore(flags);
113 ia64_srlz_i(); /* srlz.i implies srlz.d */
116 long ia64_pal_vp_create(u64 *vpd, u64 *host_iva, u64 *opt_handler)
118 struct ia64_pal_retval iprv;
120 PAL_CALL_STK(iprv, PAL_VP_CREATE, (u64)vpd, (u64)host_iva,
121 (u64)opt_handler);
123 return iprv.status;
126 static DEFINE_SPINLOCK(vp_lock);
128 int kvm_arch_hardware_enable(void *garbage)
130 long status;
131 long tmp_base;
132 unsigned long pte;
133 unsigned long saved_psr;
134 int slot;
136 pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
137 local_irq_save(saved_psr);
138 slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
139 local_irq_restore(saved_psr);
140 if (slot < 0)
141 return -EINVAL;
143 spin_lock(&vp_lock);
144 status = ia64_pal_vp_init_env(kvm_vsa_base ?
145 VP_INIT_ENV : VP_INIT_ENV_INITALIZE,
146 __pa(kvm_vm_buffer), KVM_VM_BUFFER_BASE, &tmp_base);
147 if (status != 0) {
148 spin_unlock(&vp_lock);
149 printk(KERN_WARNING"kvm: Failed to Enable VT Support!!!!\n");
150 return -EINVAL;
153 if (!kvm_vsa_base) {
154 kvm_vsa_base = tmp_base;
155 printk(KERN_INFO"kvm: kvm_vsa_base:0x%lx\n", kvm_vsa_base);
157 spin_unlock(&vp_lock);
158 ia64_ptr_entry(0x3, slot);
160 return 0;
163 void kvm_arch_hardware_disable(void *garbage)
166 long status;
167 int slot;
168 unsigned long pte;
169 unsigned long saved_psr;
170 unsigned long host_iva = ia64_getreg(_IA64_REG_CR_IVA);
172 pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
173 PAGE_KERNEL));
175 local_irq_save(saved_psr);
176 slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
177 local_irq_restore(saved_psr);
178 if (slot < 0)
179 return;
181 status = ia64_pal_vp_exit_env(host_iva);
182 if (status)
183 printk(KERN_DEBUG"kvm: Failed to disable VT support! :%ld\n",
184 status);
185 ia64_ptr_entry(0x3, slot);
188 void kvm_arch_check_processor_compat(void *rtn)
190 *(int *)rtn = 0;
193 int kvm_dev_ioctl_check_extension(long ext)
196 int r;
198 switch (ext) {
199 case KVM_CAP_IRQCHIP:
200 case KVM_CAP_MP_STATE:
201 case KVM_CAP_IRQ_INJECT_STATUS:
202 r = 1;
203 break;
204 case KVM_CAP_COALESCED_MMIO:
205 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
206 break;
207 case KVM_CAP_IOMMU:
208 r = iommu_present(&pci_bus_type);
209 break;
210 default:
211 r = 0;
213 return r;
217 static int handle_vm_error(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
219 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
220 kvm_run->hw.hardware_exit_reason = 1;
221 return 0;
224 static int handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
226 struct kvm_mmio_req *p;
227 struct kvm_io_device *mmio_dev;
228 int r;
230 p = kvm_get_vcpu_ioreq(vcpu);
232 if ((p->addr & PAGE_MASK) == IOAPIC_DEFAULT_BASE_ADDRESS)
233 goto mmio;
234 vcpu->mmio_needed = 1;
235 vcpu->mmio_phys_addr = kvm_run->mmio.phys_addr = p->addr;
236 vcpu->mmio_size = kvm_run->mmio.len = p->size;
237 vcpu->mmio_is_write = kvm_run->mmio.is_write = !p->dir;
239 if (vcpu->mmio_is_write)
240 memcpy(vcpu->mmio_data, &p->data, p->size);
241 memcpy(kvm_run->mmio.data, &p->data, p->size);
242 kvm_run->exit_reason = KVM_EXIT_MMIO;
243 return 0;
244 mmio:
245 if (p->dir)
246 r = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, p->addr,
247 p->size, &p->data);
248 else
249 r = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, p->addr,
250 p->size, &p->data);
251 if (r)
252 printk(KERN_ERR"kvm: No iodevice found! addr:%lx\n", p->addr);
253 p->state = STATE_IORESP_READY;
255 return 1;
258 static int handle_pal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
260 struct exit_ctl_data *p;
262 p = kvm_get_exit_data(vcpu);
264 if (p->exit_reason == EXIT_REASON_PAL_CALL)
265 return kvm_pal_emul(vcpu, kvm_run);
266 else {
267 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
268 kvm_run->hw.hardware_exit_reason = 2;
269 return 0;
273 static int handle_sal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
275 struct exit_ctl_data *p;
277 p = kvm_get_exit_data(vcpu);
279 if (p->exit_reason == EXIT_REASON_SAL_CALL) {
280 kvm_sal_emul(vcpu);
281 return 1;
282 } else {
283 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
284 kvm_run->hw.hardware_exit_reason = 3;
285 return 0;
290 static int __apic_accept_irq(struct kvm_vcpu *vcpu, uint64_t vector)
292 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
294 if (!test_and_set_bit(vector, &vpd->irr[0])) {
295 vcpu->arch.irq_new_pending = 1;
296 kvm_vcpu_kick(vcpu);
297 return 1;
299 return 0;
303 * offset: address offset to IPI space.
304 * value: deliver value.
306 static void vcpu_deliver_ipi(struct kvm_vcpu *vcpu, uint64_t dm,
307 uint64_t vector)
309 switch (dm) {
310 case SAPIC_FIXED:
311 break;
312 case SAPIC_NMI:
313 vector = 2;
314 break;
315 case SAPIC_EXTINT:
316 vector = 0;
317 break;
318 case SAPIC_INIT:
319 case SAPIC_PMI:
320 default:
321 printk(KERN_ERR"kvm: Unimplemented Deliver reserved IPI!\n");
322 return;
324 __apic_accept_irq(vcpu, vector);
327 static struct kvm_vcpu *lid_to_vcpu(struct kvm *kvm, unsigned long id,
328 unsigned long eid)
330 union ia64_lid lid;
331 int i;
332 struct kvm_vcpu *vcpu;
334 kvm_for_each_vcpu(i, vcpu, kvm) {
335 lid.val = VCPU_LID(vcpu);
336 if (lid.id == id && lid.eid == eid)
337 return vcpu;
340 return NULL;
343 static int handle_ipi(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
345 struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
346 struct kvm_vcpu *target_vcpu;
347 struct kvm_pt_regs *regs;
348 union ia64_ipi_a addr = p->u.ipi_data.addr;
349 union ia64_ipi_d data = p->u.ipi_data.data;
351 target_vcpu = lid_to_vcpu(vcpu->kvm, addr.id, addr.eid);
352 if (!target_vcpu)
353 return handle_vm_error(vcpu, kvm_run);
355 if (!target_vcpu->arch.launched) {
356 regs = vcpu_regs(target_vcpu);
358 regs->cr_iip = vcpu->kvm->arch.rdv_sal_data.boot_ip;
359 regs->r1 = vcpu->kvm->arch.rdv_sal_data.boot_gp;
361 target_vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
362 if (waitqueue_active(&target_vcpu->wq))
363 wake_up_interruptible(&target_vcpu->wq);
364 } else {
365 vcpu_deliver_ipi(target_vcpu, data.dm, data.vector);
366 if (target_vcpu != vcpu)
367 kvm_vcpu_kick(target_vcpu);
370 return 1;
373 struct call_data {
374 struct kvm_ptc_g ptc_g_data;
375 struct kvm_vcpu *vcpu;
378 static void vcpu_global_purge(void *info)
380 struct call_data *p = (struct call_data *)info;
381 struct kvm_vcpu *vcpu = p->vcpu;
383 if (test_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
384 return;
386 set_bit(KVM_REQ_PTC_G, &vcpu->requests);
387 if (vcpu->arch.ptc_g_count < MAX_PTC_G_NUM) {
388 vcpu->arch.ptc_g_data[vcpu->arch.ptc_g_count++] =
389 p->ptc_g_data;
390 } else {
391 clear_bit(KVM_REQ_PTC_G, &vcpu->requests);
392 vcpu->arch.ptc_g_count = 0;
393 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
397 static int handle_global_purge(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
399 struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
400 struct kvm *kvm = vcpu->kvm;
401 struct call_data call_data;
402 int i;
403 struct kvm_vcpu *vcpui;
405 call_data.ptc_g_data = p->u.ptc_g_data;
407 kvm_for_each_vcpu(i, vcpui, kvm) {
408 if (vcpui->arch.mp_state == KVM_MP_STATE_UNINITIALIZED ||
409 vcpu == vcpui)
410 continue;
412 if (waitqueue_active(&vcpui->wq))
413 wake_up_interruptible(&vcpui->wq);
415 if (vcpui->cpu != -1) {
416 call_data.vcpu = vcpui;
417 smp_call_function_single(vcpui->cpu,
418 vcpu_global_purge, &call_data, 1);
419 } else
420 printk(KERN_WARNING"kvm: Uninit vcpu received ipi!\n");
423 return 1;
426 static int handle_switch_rr6(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
428 return 1;
431 static int kvm_sn2_setup_mappings(struct kvm_vcpu *vcpu)
433 unsigned long pte, rtc_phys_addr, map_addr;
434 int slot;
436 map_addr = KVM_VMM_BASE + (1UL << KVM_VMM_SHIFT);
437 rtc_phys_addr = LOCAL_MMR_OFFSET | SH_RTC;
438 pte = pte_val(mk_pte_phys(rtc_phys_addr, PAGE_KERNEL_UC));
439 slot = ia64_itr_entry(0x3, map_addr, pte, PAGE_SHIFT);
440 vcpu->arch.sn_rtc_tr_slot = slot;
441 if (slot < 0) {
442 printk(KERN_ERR "Mayday mayday! RTC mapping failed!\n");
443 slot = 0;
445 return slot;
448 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
451 ktime_t kt;
452 long itc_diff;
453 unsigned long vcpu_now_itc;
454 unsigned long expires;
455 struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
456 unsigned long cyc_per_usec = local_cpu_data->cyc_per_usec;
457 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
459 if (irqchip_in_kernel(vcpu->kvm)) {
461 vcpu_now_itc = kvm_get_itc(vcpu) + vcpu->arch.itc_offset;
463 if (time_after(vcpu_now_itc, vpd->itm)) {
464 vcpu->arch.timer_check = 1;
465 return 1;
467 itc_diff = vpd->itm - vcpu_now_itc;
468 if (itc_diff < 0)
469 itc_diff = -itc_diff;
471 expires = div64_u64(itc_diff, cyc_per_usec);
472 kt = ktime_set(0, 1000 * expires);
474 vcpu->arch.ht_active = 1;
475 hrtimer_start(p_ht, kt, HRTIMER_MODE_ABS);
477 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
478 kvm_vcpu_block(vcpu);
479 hrtimer_cancel(p_ht);
480 vcpu->arch.ht_active = 0;
482 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests) ||
483 kvm_cpu_has_pending_timer(vcpu))
484 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
485 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
487 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
488 return -EINTR;
489 return 1;
490 } else {
491 printk(KERN_ERR"kvm: Unsupported userspace halt!");
492 return 0;
496 static int handle_vm_shutdown(struct kvm_vcpu *vcpu,
497 struct kvm_run *kvm_run)
499 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
500 return 0;
503 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
504 struct kvm_run *kvm_run)
506 return 1;
509 static int handle_vcpu_debug(struct kvm_vcpu *vcpu,
510 struct kvm_run *kvm_run)
512 printk("VMM: %s", vcpu->arch.log_buf);
513 return 1;
516 static int (*kvm_vti_exit_handlers[])(struct kvm_vcpu *vcpu,
517 struct kvm_run *kvm_run) = {
518 [EXIT_REASON_VM_PANIC] = handle_vm_error,
519 [EXIT_REASON_MMIO_INSTRUCTION] = handle_mmio,
520 [EXIT_REASON_PAL_CALL] = handle_pal_call,
521 [EXIT_REASON_SAL_CALL] = handle_sal_call,
522 [EXIT_REASON_SWITCH_RR6] = handle_switch_rr6,
523 [EXIT_REASON_VM_DESTROY] = handle_vm_shutdown,
524 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
525 [EXIT_REASON_IPI] = handle_ipi,
526 [EXIT_REASON_PTC_G] = handle_global_purge,
527 [EXIT_REASON_DEBUG] = handle_vcpu_debug,
531 static const int kvm_vti_max_exit_handlers =
532 sizeof(kvm_vti_exit_handlers)/sizeof(*kvm_vti_exit_handlers);
534 static uint32_t kvm_get_exit_reason(struct kvm_vcpu *vcpu)
536 struct exit_ctl_data *p_exit_data;
538 p_exit_data = kvm_get_exit_data(vcpu);
539 return p_exit_data->exit_reason;
543 * The guest has exited. See if we can fix it or if we need userspace
544 * assistance.
546 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
548 u32 exit_reason = kvm_get_exit_reason(vcpu);
549 vcpu->arch.last_exit = exit_reason;
551 if (exit_reason < kvm_vti_max_exit_handlers
552 && kvm_vti_exit_handlers[exit_reason])
553 return kvm_vti_exit_handlers[exit_reason](vcpu, kvm_run);
554 else {
555 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
556 kvm_run->hw.hardware_exit_reason = exit_reason;
558 return 0;
561 static inline void vti_set_rr6(unsigned long rr6)
563 ia64_set_rr(RR6, rr6);
564 ia64_srlz_i();
567 static int kvm_insert_vmm_mapping(struct kvm_vcpu *vcpu)
569 unsigned long pte;
570 struct kvm *kvm = vcpu->kvm;
571 int r;
573 /*Insert a pair of tr to map vmm*/
574 pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
575 r = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
576 if (r < 0)
577 goto out;
578 vcpu->arch.vmm_tr_slot = r;
579 /*Insert a pairt of tr to map data of vm*/
580 pte = pte_val(mk_pte_phys(__pa(kvm->arch.vm_base), PAGE_KERNEL));
581 r = ia64_itr_entry(0x3, KVM_VM_DATA_BASE,
582 pte, KVM_VM_DATA_SHIFT);
583 if (r < 0)
584 goto out;
585 vcpu->arch.vm_tr_slot = r;
587 #if defined(CONFIG_IA64_SGI_SN2) || defined(CONFIG_IA64_GENERIC)
588 if (kvm->arch.is_sn2) {
589 r = kvm_sn2_setup_mappings(vcpu);
590 if (r < 0)
591 goto out;
593 #endif
595 r = 0;
596 out:
597 return r;
600 static void kvm_purge_vmm_mapping(struct kvm_vcpu *vcpu)
602 struct kvm *kvm = vcpu->kvm;
603 ia64_ptr_entry(0x3, vcpu->arch.vmm_tr_slot);
604 ia64_ptr_entry(0x3, vcpu->arch.vm_tr_slot);
605 #if defined(CONFIG_IA64_SGI_SN2) || defined(CONFIG_IA64_GENERIC)
606 if (kvm->arch.is_sn2)
607 ia64_ptr_entry(0x3, vcpu->arch.sn_rtc_tr_slot);
608 #endif
611 static int kvm_vcpu_pre_transition(struct kvm_vcpu *vcpu)
613 unsigned long psr;
614 int r;
615 int cpu = smp_processor_id();
617 if (vcpu->arch.last_run_cpu != cpu ||
618 per_cpu(last_vcpu, cpu) != vcpu) {
619 per_cpu(last_vcpu, cpu) = vcpu;
620 vcpu->arch.last_run_cpu = cpu;
621 kvm_flush_tlb_all();
624 vcpu->arch.host_rr6 = ia64_get_rr(RR6);
625 vti_set_rr6(vcpu->arch.vmm_rr);
626 local_irq_save(psr);
627 r = kvm_insert_vmm_mapping(vcpu);
628 local_irq_restore(psr);
629 return r;
632 static void kvm_vcpu_post_transition(struct kvm_vcpu *vcpu)
634 kvm_purge_vmm_mapping(vcpu);
635 vti_set_rr6(vcpu->arch.host_rr6);
638 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
640 union context *host_ctx, *guest_ctx;
641 int r, idx;
643 idx = srcu_read_lock(&vcpu->kvm->srcu);
645 again:
646 if (signal_pending(current)) {
647 r = -EINTR;
648 kvm_run->exit_reason = KVM_EXIT_INTR;
649 goto out;
652 preempt_disable();
653 local_irq_disable();
655 /*Get host and guest context with guest address space.*/
656 host_ctx = kvm_get_host_context(vcpu);
657 guest_ctx = kvm_get_guest_context(vcpu);
659 clear_bit(KVM_REQ_KICK, &vcpu->requests);
661 r = kvm_vcpu_pre_transition(vcpu);
662 if (r < 0)
663 goto vcpu_run_fail;
665 srcu_read_unlock(&vcpu->kvm->srcu, idx);
666 vcpu->mode = IN_GUEST_MODE;
667 kvm_guest_enter();
670 * Transition to the guest
672 kvm_vmm_info->tramp_entry(host_ctx, guest_ctx);
674 kvm_vcpu_post_transition(vcpu);
676 vcpu->arch.launched = 1;
677 set_bit(KVM_REQ_KICK, &vcpu->requests);
678 local_irq_enable();
681 * We must have an instruction between local_irq_enable() and
682 * kvm_guest_exit(), so the timer interrupt isn't delayed by
683 * the interrupt shadow. The stat.exits increment will do nicely.
684 * But we need to prevent reordering, hence this barrier():
686 barrier();
687 kvm_guest_exit();
688 vcpu->mode = OUTSIDE_GUEST_MODE;
689 preempt_enable();
691 idx = srcu_read_lock(&vcpu->kvm->srcu);
693 r = kvm_handle_exit(kvm_run, vcpu);
695 if (r > 0) {
696 if (!need_resched())
697 goto again;
700 out:
701 srcu_read_unlock(&vcpu->kvm->srcu, idx);
702 if (r > 0) {
703 kvm_resched(vcpu);
704 idx = srcu_read_lock(&vcpu->kvm->srcu);
705 goto again;
708 return r;
710 vcpu_run_fail:
711 local_irq_enable();
712 preempt_enable();
713 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
714 goto out;
717 static void kvm_set_mmio_data(struct kvm_vcpu *vcpu)
719 struct kvm_mmio_req *p = kvm_get_vcpu_ioreq(vcpu);
721 if (!vcpu->mmio_is_write)
722 memcpy(&p->data, vcpu->mmio_data, 8);
723 p->state = STATE_IORESP_READY;
726 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
728 int r;
729 sigset_t sigsaved;
731 if (vcpu->sigset_active)
732 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
734 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
735 kvm_vcpu_block(vcpu);
736 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
737 r = -EAGAIN;
738 goto out;
741 if (vcpu->mmio_needed) {
742 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
743 kvm_set_mmio_data(vcpu);
744 vcpu->mmio_read_completed = 1;
745 vcpu->mmio_needed = 0;
747 r = __vcpu_run(vcpu, kvm_run);
748 out:
749 if (vcpu->sigset_active)
750 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
752 return r;
755 struct kvm *kvm_arch_alloc_vm(void)
758 struct kvm *kvm;
759 uint64_t vm_base;
761 BUG_ON(sizeof(struct kvm) > KVM_VM_STRUCT_SIZE);
763 vm_base = __get_free_pages(GFP_KERNEL, get_order(KVM_VM_DATA_SIZE));
765 if (!vm_base)
766 return NULL;
768 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
769 kvm = (struct kvm *)(vm_base +
770 offsetof(struct kvm_vm_data, kvm_vm_struct));
771 kvm->arch.vm_base = vm_base;
772 printk(KERN_DEBUG"kvm: vm's data area:0x%lx\n", vm_base);
774 return kvm;
777 struct kvm_ia64_io_range {
778 unsigned long start;
779 unsigned long size;
780 unsigned long type;
783 static const struct kvm_ia64_io_range io_ranges[] = {
784 {VGA_IO_START, VGA_IO_SIZE, GPFN_FRAME_BUFFER},
785 {MMIO_START, MMIO_SIZE, GPFN_LOW_MMIO},
786 {LEGACY_IO_START, LEGACY_IO_SIZE, GPFN_LEGACY_IO},
787 {IO_SAPIC_START, IO_SAPIC_SIZE, GPFN_IOSAPIC},
788 {PIB_START, PIB_SIZE, GPFN_PIB},
791 static void kvm_build_io_pmt(struct kvm *kvm)
793 unsigned long i, j;
795 /* Mark I/O ranges */
796 for (i = 0; i < (sizeof(io_ranges) / sizeof(struct kvm_io_range));
797 i++) {
798 for (j = io_ranges[i].start;
799 j < io_ranges[i].start + io_ranges[i].size;
800 j += PAGE_SIZE)
801 kvm_set_pmt_entry(kvm, j >> PAGE_SHIFT,
802 io_ranges[i].type, 0);
807 /*Use unused rids to virtualize guest rid.*/
808 #define GUEST_PHYSICAL_RR0 0x1739
809 #define GUEST_PHYSICAL_RR4 0x2739
810 #define VMM_INIT_RR 0x1660
812 int kvm_arch_init_vm(struct kvm *kvm)
814 BUG_ON(!kvm);
816 kvm->arch.is_sn2 = ia64_platform_is("sn2");
818 kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
819 kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
820 kvm->arch.vmm_init_rr = VMM_INIT_RR;
823 *Fill P2M entries for MMIO/IO ranges
825 kvm_build_io_pmt(kvm);
827 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
829 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
830 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
832 return 0;
835 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
836 struct kvm_irqchip *chip)
838 int r;
840 r = 0;
841 switch (chip->chip_id) {
842 case KVM_IRQCHIP_IOAPIC:
843 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
844 break;
845 default:
846 r = -EINVAL;
847 break;
849 return r;
852 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
854 int r;
856 r = 0;
857 switch (chip->chip_id) {
858 case KVM_IRQCHIP_IOAPIC:
859 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
860 break;
861 default:
862 r = -EINVAL;
863 break;
865 return r;
868 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
870 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
872 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
873 int i;
875 for (i = 0; i < 16; i++) {
876 vpd->vgr[i] = regs->vpd.vgr[i];
877 vpd->vbgr[i] = regs->vpd.vbgr[i];
879 for (i = 0; i < 128; i++)
880 vpd->vcr[i] = regs->vpd.vcr[i];
881 vpd->vhpi = regs->vpd.vhpi;
882 vpd->vnat = regs->vpd.vnat;
883 vpd->vbnat = regs->vpd.vbnat;
884 vpd->vpsr = regs->vpd.vpsr;
886 vpd->vpr = regs->vpd.vpr;
888 memcpy(&vcpu->arch.guest, &regs->saved_guest, sizeof(union context));
890 RESTORE_REGS(mp_state);
891 RESTORE_REGS(vmm_rr);
892 memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
893 memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
894 RESTORE_REGS(itr_regions);
895 RESTORE_REGS(dtr_regions);
896 RESTORE_REGS(tc_regions);
897 RESTORE_REGS(irq_check);
898 RESTORE_REGS(itc_check);
899 RESTORE_REGS(timer_check);
900 RESTORE_REGS(timer_pending);
901 RESTORE_REGS(last_itc);
902 for (i = 0; i < 8; i++) {
903 vcpu->arch.vrr[i] = regs->vrr[i];
904 vcpu->arch.ibr[i] = regs->ibr[i];
905 vcpu->arch.dbr[i] = regs->dbr[i];
907 for (i = 0; i < 4; i++)
908 vcpu->arch.insvc[i] = regs->insvc[i];
909 RESTORE_REGS(xtp);
910 RESTORE_REGS(metaphysical_rr0);
911 RESTORE_REGS(metaphysical_rr4);
912 RESTORE_REGS(metaphysical_saved_rr0);
913 RESTORE_REGS(metaphysical_saved_rr4);
914 RESTORE_REGS(fp_psr);
915 RESTORE_REGS(saved_gp);
917 vcpu->arch.irq_new_pending = 1;
918 vcpu->arch.itc_offset = regs->saved_itc - kvm_get_itc(vcpu);
919 set_bit(KVM_REQ_RESUME, &vcpu->requests);
921 return 0;
924 long kvm_arch_vm_ioctl(struct file *filp,
925 unsigned int ioctl, unsigned long arg)
927 struct kvm *kvm = filp->private_data;
928 void __user *argp = (void __user *)arg;
929 int r = -ENOTTY;
931 switch (ioctl) {
932 case KVM_SET_MEMORY_REGION: {
933 struct kvm_memory_region kvm_mem;
934 struct kvm_userspace_memory_region kvm_userspace_mem;
936 r = -EFAULT;
937 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
938 goto out;
939 kvm_userspace_mem.slot = kvm_mem.slot;
940 kvm_userspace_mem.flags = kvm_mem.flags;
941 kvm_userspace_mem.guest_phys_addr =
942 kvm_mem.guest_phys_addr;
943 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
944 r = kvm_vm_ioctl_set_memory_region(kvm,
945 &kvm_userspace_mem, 0);
946 if (r)
947 goto out;
948 break;
950 case KVM_CREATE_IRQCHIP:
951 r = -EFAULT;
952 r = kvm_ioapic_init(kvm);
953 if (r)
954 goto out;
955 r = kvm_setup_default_irq_routing(kvm);
956 if (r) {
957 mutex_lock(&kvm->slots_lock);
958 kvm_ioapic_destroy(kvm);
959 mutex_unlock(&kvm->slots_lock);
960 goto out;
962 break;
963 case KVM_IRQ_LINE_STATUS:
964 case KVM_IRQ_LINE: {
965 struct kvm_irq_level irq_event;
967 r = -EFAULT;
968 if (copy_from_user(&irq_event, argp, sizeof irq_event))
969 goto out;
970 r = -ENXIO;
971 if (irqchip_in_kernel(kvm)) {
972 __s32 status;
973 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
974 irq_event.irq, irq_event.level);
975 if (ioctl == KVM_IRQ_LINE_STATUS) {
976 r = -EFAULT;
977 irq_event.status = status;
978 if (copy_to_user(argp, &irq_event,
979 sizeof irq_event))
980 goto out;
982 r = 0;
984 break;
986 case KVM_GET_IRQCHIP: {
987 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
988 struct kvm_irqchip chip;
990 r = -EFAULT;
991 if (copy_from_user(&chip, argp, sizeof chip))
992 goto out;
993 r = -ENXIO;
994 if (!irqchip_in_kernel(kvm))
995 goto out;
996 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
997 if (r)
998 goto out;
999 r = -EFAULT;
1000 if (copy_to_user(argp, &chip, sizeof chip))
1001 goto out;
1002 r = 0;
1003 break;
1005 case KVM_SET_IRQCHIP: {
1006 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1007 struct kvm_irqchip chip;
1009 r = -EFAULT;
1010 if (copy_from_user(&chip, argp, sizeof chip))
1011 goto out;
1012 r = -ENXIO;
1013 if (!irqchip_in_kernel(kvm))
1014 goto out;
1015 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1016 if (r)
1017 goto out;
1018 r = 0;
1019 break;
1021 default:
1024 out:
1025 return r;
1028 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1029 struct kvm_sregs *sregs)
1031 return -EINVAL;
1034 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1035 struct kvm_sregs *sregs)
1037 return -EINVAL;
1040 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1041 struct kvm_translation *tr)
1044 return -EINVAL;
1047 static int kvm_alloc_vmm_area(void)
1049 if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1050 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1051 get_order(KVM_VMM_SIZE));
1052 if (!kvm_vmm_base)
1053 return -ENOMEM;
1055 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1056 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1058 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1059 kvm_vmm_base, kvm_vm_buffer);
1062 return 0;
1065 static void kvm_free_vmm_area(void)
1067 if (kvm_vmm_base) {
1068 /*Zero this area before free to avoid bits leak!!*/
1069 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1070 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1071 kvm_vmm_base = 0;
1072 kvm_vm_buffer = 0;
1073 kvm_vsa_base = 0;
1077 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1079 int i;
1080 union cpuid3_t cpuid3;
1081 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1083 if (IS_ERR(vpd))
1084 return PTR_ERR(vpd);
1086 /* CPUID init */
1087 for (i = 0; i < 5; i++)
1088 vpd->vcpuid[i] = ia64_get_cpuid(i);
1090 /* Limit the CPUID number to 5 */
1091 cpuid3.value = vpd->vcpuid[3];
1092 cpuid3.number = 4; /* 5 - 1 */
1093 vpd->vcpuid[3] = cpuid3.value;
1095 /*Set vac and vdc fields*/
1096 vpd->vac.a_from_int_cr = 1;
1097 vpd->vac.a_to_int_cr = 1;
1098 vpd->vac.a_from_psr = 1;
1099 vpd->vac.a_from_cpuid = 1;
1100 vpd->vac.a_cover = 1;
1101 vpd->vac.a_bsw = 1;
1102 vpd->vac.a_int = 1;
1103 vpd->vdc.d_vmsw = 1;
1105 /*Set virtual buffer*/
1106 vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1108 return 0;
1111 static int vti_create_vp(struct kvm_vcpu *vcpu)
1113 long ret;
1114 struct vpd *vpd = vcpu->arch.vpd;
1115 unsigned long vmm_ivt;
1117 vmm_ivt = kvm_vmm_info->vmm_ivt;
1119 printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1121 ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1123 if (ret) {
1124 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1125 return -EINVAL;
1127 return 0;
1130 static void init_ptce_info(struct kvm_vcpu *vcpu)
1132 ia64_ptce_info_t ptce = {0};
1134 ia64_get_ptce(&ptce);
1135 vcpu->arch.ptce_base = ptce.base;
1136 vcpu->arch.ptce_count[0] = ptce.count[0];
1137 vcpu->arch.ptce_count[1] = ptce.count[1];
1138 vcpu->arch.ptce_stride[0] = ptce.stride[0];
1139 vcpu->arch.ptce_stride[1] = ptce.stride[1];
1142 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1144 struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1146 if (hrtimer_cancel(p_ht))
1147 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1150 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1152 struct kvm_vcpu *vcpu;
1153 wait_queue_head_t *q;
1155 vcpu = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1156 q = &vcpu->wq;
1158 if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1159 goto out;
1161 if (waitqueue_active(q))
1162 wake_up_interruptible(q);
1164 out:
1165 vcpu->arch.timer_fired = 1;
1166 vcpu->arch.timer_check = 1;
1167 return HRTIMER_NORESTART;
1170 #define PALE_RESET_ENTRY 0x80000000ffffffb0UL
1172 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
1174 return irqchip_in_kernel(vcpu->kcm) == (vcpu->arch.apic != NULL);
1177 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1179 struct kvm_vcpu *v;
1180 int r;
1181 int i;
1182 long itc_offset;
1183 struct kvm *kvm = vcpu->kvm;
1184 struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1186 union context *p_ctx = &vcpu->arch.guest;
1187 struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1189 /*Init vcpu context for first run.*/
1190 if (IS_ERR(vmm_vcpu))
1191 return PTR_ERR(vmm_vcpu);
1193 if (kvm_vcpu_is_bsp(vcpu)) {
1194 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1196 /*Set entry address for first run.*/
1197 regs->cr_iip = PALE_RESET_ENTRY;
1199 /*Initialize itc offset for vcpus*/
1200 itc_offset = 0UL - kvm_get_itc(vcpu);
1201 for (i = 0; i < KVM_MAX_VCPUS; i++) {
1202 v = (struct kvm_vcpu *)((char *)vcpu +
1203 sizeof(struct kvm_vcpu_data) * i);
1204 v->arch.itc_offset = itc_offset;
1205 v->arch.last_itc = 0;
1207 } else
1208 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1210 r = -ENOMEM;
1211 vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1212 if (!vcpu->arch.apic)
1213 goto out;
1214 vcpu->arch.apic->vcpu = vcpu;
1216 p_ctx->gr[1] = 0;
1217 p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + KVM_STK_OFFSET);
1218 p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1219 p_ctx->psr = 0x1008522000UL;
1220 p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1221 p_ctx->caller_unat = 0;
1222 p_ctx->pr = 0x0;
1223 p_ctx->ar[36] = 0x0; /*unat*/
1224 p_ctx->ar[19] = 0x0; /*rnat*/
1225 p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1226 ((sizeof(struct kvm_vcpu)+15) & ~15);
1227 p_ctx->ar[64] = 0x0; /*pfs*/
1228 p_ctx->cr[0] = 0x7e04UL;
1229 p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1230 p_ctx->cr[8] = 0x3c;
1232 /*Initialize region register*/
1233 p_ctx->rr[0] = 0x30;
1234 p_ctx->rr[1] = 0x30;
1235 p_ctx->rr[2] = 0x30;
1236 p_ctx->rr[3] = 0x30;
1237 p_ctx->rr[4] = 0x30;
1238 p_ctx->rr[5] = 0x30;
1239 p_ctx->rr[7] = 0x30;
1241 /*Initialize branch register 0*/
1242 p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1244 vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1245 vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1246 vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1248 hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1249 vcpu->arch.hlt_timer.function = hlt_timer_fn;
1251 vcpu->arch.last_run_cpu = -1;
1252 vcpu->arch.vpd = (struct vpd *)VPD_BASE(vcpu->vcpu_id);
1253 vcpu->arch.vsa_base = kvm_vsa_base;
1254 vcpu->arch.__gp = kvm_vmm_gp;
1255 vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1256 vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_BASE(vcpu->vcpu_id);
1257 vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_BASE(vcpu->vcpu_id);
1258 init_ptce_info(vcpu);
1260 r = 0;
1261 out:
1262 return r;
1265 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1267 unsigned long psr;
1268 int r;
1270 local_irq_save(psr);
1271 r = kvm_insert_vmm_mapping(vcpu);
1272 local_irq_restore(psr);
1273 if (r)
1274 goto fail;
1275 r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1276 if (r)
1277 goto fail;
1279 r = vti_init_vpd(vcpu);
1280 if (r) {
1281 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1282 goto uninit;
1285 r = vti_create_vp(vcpu);
1286 if (r)
1287 goto uninit;
1289 kvm_purge_vmm_mapping(vcpu);
1291 return 0;
1292 uninit:
1293 kvm_vcpu_uninit(vcpu);
1294 fail:
1295 return r;
1298 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1299 unsigned int id)
1301 struct kvm_vcpu *vcpu;
1302 unsigned long vm_base = kvm->arch.vm_base;
1303 int r;
1304 int cpu;
1306 BUG_ON(sizeof(struct kvm_vcpu) > VCPU_STRUCT_SIZE/2);
1308 r = -EINVAL;
1309 if (id >= KVM_MAX_VCPUS) {
1310 printk(KERN_ERR"kvm: Can't configure vcpus > %ld",
1311 KVM_MAX_VCPUS);
1312 goto fail;
1315 r = -ENOMEM;
1316 if (!vm_base) {
1317 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1318 goto fail;
1320 vcpu = (struct kvm_vcpu *)(vm_base + offsetof(struct kvm_vm_data,
1321 vcpu_data[id].vcpu_struct));
1322 vcpu->kvm = kvm;
1324 cpu = get_cpu();
1325 r = vti_vcpu_setup(vcpu, id);
1326 put_cpu();
1328 if (r) {
1329 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1330 goto fail;
1333 return vcpu;
1334 fail:
1335 return ERR_PTR(r);
1338 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1340 return 0;
1343 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1345 return -EINVAL;
1348 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1350 return -EINVAL;
1353 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1354 struct kvm_guest_debug *dbg)
1356 return -EINVAL;
1359 void kvm_arch_free_vm(struct kvm *kvm)
1361 unsigned long vm_base = kvm->arch.vm_base;
1363 if (vm_base) {
1364 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1365 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1370 static void kvm_release_vm_pages(struct kvm *kvm)
1372 struct kvm_memslots *slots;
1373 struct kvm_memory_slot *memslot;
1374 int j;
1375 unsigned long base_gfn;
1377 slots = kvm_memslots(kvm);
1378 kvm_for_each_memslot(memslot, slots) {
1379 base_gfn = memslot->base_gfn;
1380 for (j = 0; j < memslot->npages; j++) {
1381 if (memslot->rmap[j])
1382 put_page((struct page *)memslot->rmap[j]);
1387 void kvm_arch_sync_events(struct kvm *kvm)
1391 void kvm_arch_destroy_vm(struct kvm *kvm)
1393 kvm_iommu_unmap_guest(kvm);
1394 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1395 kvm_free_all_assigned_devices(kvm);
1396 #endif
1397 kfree(kvm->arch.vioapic);
1398 kvm_release_vm_pages(kvm);
1401 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1405 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1407 if (cpu != vcpu->cpu) {
1408 vcpu->cpu = cpu;
1409 if (vcpu->arch.ht_active)
1410 kvm_migrate_hlt_timer(vcpu);
1414 #define SAVE_REGS(_x) regs->_x = vcpu->arch._x
1416 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1418 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1419 int i;
1421 vcpu_load(vcpu);
1423 for (i = 0; i < 16; i++) {
1424 regs->vpd.vgr[i] = vpd->vgr[i];
1425 regs->vpd.vbgr[i] = vpd->vbgr[i];
1427 for (i = 0; i < 128; i++)
1428 regs->vpd.vcr[i] = vpd->vcr[i];
1429 regs->vpd.vhpi = vpd->vhpi;
1430 regs->vpd.vnat = vpd->vnat;
1431 regs->vpd.vbnat = vpd->vbnat;
1432 regs->vpd.vpsr = vpd->vpsr;
1433 regs->vpd.vpr = vpd->vpr;
1435 memcpy(&regs->saved_guest, &vcpu->arch.guest, sizeof(union context));
1437 SAVE_REGS(mp_state);
1438 SAVE_REGS(vmm_rr);
1439 memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1440 memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1441 SAVE_REGS(itr_regions);
1442 SAVE_REGS(dtr_regions);
1443 SAVE_REGS(tc_regions);
1444 SAVE_REGS(irq_check);
1445 SAVE_REGS(itc_check);
1446 SAVE_REGS(timer_check);
1447 SAVE_REGS(timer_pending);
1448 SAVE_REGS(last_itc);
1449 for (i = 0; i < 8; i++) {
1450 regs->vrr[i] = vcpu->arch.vrr[i];
1451 regs->ibr[i] = vcpu->arch.ibr[i];
1452 regs->dbr[i] = vcpu->arch.dbr[i];
1454 for (i = 0; i < 4; i++)
1455 regs->insvc[i] = vcpu->arch.insvc[i];
1456 regs->saved_itc = vcpu->arch.itc_offset + kvm_get_itc(vcpu);
1457 SAVE_REGS(xtp);
1458 SAVE_REGS(metaphysical_rr0);
1459 SAVE_REGS(metaphysical_rr4);
1460 SAVE_REGS(metaphysical_saved_rr0);
1461 SAVE_REGS(metaphysical_saved_rr4);
1462 SAVE_REGS(fp_psr);
1463 SAVE_REGS(saved_gp);
1465 vcpu_put(vcpu);
1466 return 0;
1469 int kvm_arch_vcpu_ioctl_get_stack(struct kvm_vcpu *vcpu,
1470 struct kvm_ia64_vcpu_stack *stack)
1472 memcpy(stack, vcpu, sizeof(struct kvm_ia64_vcpu_stack));
1473 return 0;
1476 int kvm_arch_vcpu_ioctl_set_stack(struct kvm_vcpu *vcpu,
1477 struct kvm_ia64_vcpu_stack *stack)
1479 memcpy(vcpu + 1, &stack->stack[0] + sizeof(struct kvm_vcpu),
1480 sizeof(struct kvm_ia64_vcpu_stack) - sizeof(struct kvm_vcpu));
1482 vcpu->arch.exit_data = ((struct kvm_vcpu *)stack)->arch.exit_data;
1483 return 0;
1486 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1489 hrtimer_cancel(&vcpu->arch.hlt_timer);
1490 kfree(vcpu->arch.apic);
1494 long kvm_arch_vcpu_ioctl(struct file *filp,
1495 unsigned int ioctl, unsigned long arg)
1497 struct kvm_vcpu *vcpu = filp->private_data;
1498 void __user *argp = (void __user *)arg;
1499 struct kvm_ia64_vcpu_stack *stack = NULL;
1500 long r;
1502 switch (ioctl) {
1503 case KVM_IA64_VCPU_GET_STACK: {
1504 struct kvm_ia64_vcpu_stack __user *user_stack;
1505 void __user *first_p = argp;
1507 r = -EFAULT;
1508 if (copy_from_user(&user_stack, first_p, sizeof(void *)))
1509 goto out;
1511 if (!access_ok(VERIFY_WRITE, user_stack,
1512 sizeof(struct kvm_ia64_vcpu_stack))) {
1513 printk(KERN_INFO "KVM_IA64_VCPU_GET_STACK: "
1514 "Illegal user destination address for stack\n");
1515 goto out;
1517 stack = kzalloc(sizeof(struct kvm_ia64_vcpu_stack), GFP_KERNEL);
1518 if (!stack) {
1519 r = -ENOMEM;
1520 goto out;
1523 r = kvm_arch_vcpu_ioctl_get_stack(vcpu, stack);
1524 if (r)
1525 goto out;
1527 if (copy_to_user(user_stack, stack,
1528 sizeof(struct kvm_ia64_vcpu_stack))) {
1529 r = -EFAULT;
1530 goto out;
1533 break;
1535 case KVM_IA64_VCPU_SET_STACK: {
1536 struct kvm_ia64_vcpu_stack __user *user_stack;
1537 void __user *first_p = argp;
1539 r = -EFAULT;
1540 if (copy_from_user(&user_stack, first_p, sizeof(void *)))
1541 goto out;
1543 if (!access_ok(VERIFY_READ, user_stack,
1544 sizeof(struct kvm_ia64_vcpu_stack))) {
1545 printk(KERN_INFO "KVM_IA64_VCPU_SET_STACK: "
1546 "Illegal user address for stack\n");
1547 goto out;
1549 stack = kmalloc(sizeof(struct kvm_ia64_vcpu_stack), GFP_KERNEL);
1550 if (!stack) {
1551 r = -ENOMEM;
1552 goto out;
1554 if (copy_from_user(stack, user_stack,
1555 sizeof(struct kvm_ia64_vcpu_stack)))
1556 goto out;
1558 r = kvm_arch_vcpu_ioctl_set_stack(vcpu, stack);
1559 break;
1562 default:
1563 r = -EINVAL;
1566 out:
1567 kfree(stack);
1568 return r;
1571 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1572 struct kvm_memory_slot *memslot,
1573 struct kvm_memory_slot old,
1574 struct kvm_userspace_memory_region *mem,
1575 int user_alloc)
1577 unsigned long i;
1578 unsigned long pfn;
1579 int npages = memslot->npages;
1580 unsigned long base_gfn = memslot->base_gfn;
1582 if (base_gfn + npages > (KVM_MAX_MEM_SIZE >> PAGE_SHIFT))
1583 return -ENOMEM;
1585 for (i = 0; i < npages; i++) {
1586 pfn = gfn_to_pfn(kvm, base_gfn + i);
1587 if (!kvm_is_mmio_pfn(pfn)) {
1588 kvm_set_pmt_entry(kvm, base_gfn + i,
1589 pfn << PAGE_SHIFT,
1590 _PAGE_AR_RWX | _PAGE_MA_WB);
1591 memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1592 } else {
1593 kvm_set_pmt_entry(kvm, base_gfn + i,
1594 GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1595 _PAGE_MA_UC);
1596 memslot->rmap[i] = 0;
1600 return 0;
1603 void kvm_arch_commit_memory_region(struct kvm *kvm,
1604 struct kvm_userspace_memory_region *mem,
1605 struct kvm_memory_slot old,
1606 int user_alloc)
1608 return;
1611 void kvm_arch_flush_shadow(struct kvm *kvm)
1613 kvm_flush_remote_tlbs(kvm);
1616 long kvm_arch_dev_ioctl(struct file *filp,
1617 unsigned int ioctl, unsigned long arg)
1619 return -EINVAL;
1622 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1624 kvm_vcpu_uninit(vcpu);
1627 static int vti_cpu_has_kvm_support(void)
1629 long avail = 1, status = 1, control = 1;
1630 long ret;
1632 ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1633 if (ret)
1634 goto out;
1636 if (!(avail & PAL_PROC_VM_BIT))
1637 goto out;
1639 printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1641 ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1642 if (ret)
1643 goto out;
1644 printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1646 if (!(vp_env_info & VP_OPCODE)) {
1647 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1648 "vm_env_info:0x%lx\n", vp_env_info);
1651 return 1;
1652 out:
1653 return 0;
1658 * On SN2, the ITC isn't stable, so copy in fast path code to use the
1659 * SN2 RTC, replacing the ITC based default verion.
1661 static void kvm_patch_vmm(struct kvm_vmm_info *vmm_info,
1662 struct module *module)
1664 unsigned long new_ar, new_ar_sn2;
1665 unsigned long module_base;
1667 if (!ia64_platform_is("sn2"))
1668 return;
1670 module_base = (unsigned long)module->module_core;
1672 new_ar = kvm_vmm_base + vmm_info->patch_mov_ar - module_base;
1673 new_ar_sn2 = kvm_vmm_base + vmm_info->patch_mov_ar_sn2 - module_base;
1675 printk(KERN_INFO "kvm: Patching ITC emulation to use SGI SN2 RTC "
1676 "as source\n");
1679 * Copy the SN2 version of mov_ar into place. They are both
1680 * the same size, so 6 bundles is sufficient (6 * 0x10).
1682 memcpy((void *)new_ar, (void *)new_ar_sn2, 0x60);
1685 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1686 struct module *module)
1688 unsigned long module_base;
1689 unsigned long vmm_size;
1691 unsigned long vmm_offset, func_offset, fdesc_offset;
1692 struct fdesc *p_fdesc;
1694 BUG_ON(!module);
1696 if (!kvm_vmm_base) {
1697 printk("kvm: kvm area hasn't been initialized yet!!\n");
1698 return -EFAULT;
1701 /*Calculate new position of relocated vmm module.*/
1702 module_base = (unsigned long)module->module_core;
1703 vmm_size = module->core_size;
1704 if (unlikely(vmm_size > KVM_VMM_SIZE))
1705 return -EFAULT;
1707 memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1708 kvm_patch_vmm(vmm_info, module);
1709 kvm_flush_icache(kvm_vmm_base, vmm_size);
1711 /*Recalculate kvm_vmm_info based on new VMM*/
1712 vmm_offset = vmm_info->vmm_ivt - module_base;
1713 kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1714 printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1715 kvm_vmm_info->vmm_ivt);
1717 fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1718 kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1719 fdesc_offset);
1720 func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1721 p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1722 p_fdesc->ip = KVM_VMM_BASE + func_offset;
1723 p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1725 printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1726 KVM_VMM_BASE+func_offset);
1728 fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1729 kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1730 fdesc_offset);
1731 func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1732 p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1733 p_fdesc->ip = KVM_VMM_BASE + func_offset;
1734 p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1736 kvm_vmm_gp = p_fdesc->gp;
1738 printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1739 kvm_vmm_info->vmm_entry);
1740 printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1741 KVM_VMM_BASE + func_offset);
1743 return 0;
1746 int kvm_arch_init(void *opaque)
1748 int r;
1749 struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1751 if (!vti_cpu_has_kvm_support()) {
1752 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1753 r = -EOPNOTSUPP;
1754 goto out;
1757 if (kvm_vmm_info) {
1758 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1759 r = -EEXIST;
1760 goto out;
1763 r = -ENOMEM;
1764 kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1765 if (!kvm_vmm_info)
1766 goto out;
1768 if (kvm_alloc_vmm_area())
1769 goto out_free0;
1771 r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1772 if (r)
1773 goto out_free1;
1775 return 0;
1777 out_free1:
1778 kvm_free_vmm_area();
1779 out_free0:
1780 kfree(kvm_vmm_info);
1781 out:
1782 return r;
1785 void kvm_arch_exit(void)
1787 kvm_free_vmm_area();
1788 kfree(kvm_vmm_info);
1789 kvm_vmm_info = NULL;
1792 static void kvm_ia64_sync_dirty_log(struct kvm *kvm,
1793 struct kvm_memory_slot *memslot)
1795 int i;
1796 long base;
1797 unsigned long n;
1798 unsigned long *dirty_bitmap = (unsigned long *)(kvm->arch.vm_base +
1799 offsetof(struct kvm_vm_data, kvm_mem_dirty_log));
1801 n = kvm_dirty_bitmap_bytes(memslot);
1802 base = memslot->base_gfn / BITS_PER_LONG;
1804 spin_lock(&kvm->arch.dirty_log_lock);
1805 for (i = 0; i < n/sizeof(long); ++i) {
1806 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1807 dirty_bitmap[base + i] = 0;
1809 spin_unlock(&kvm->arch.dirty_log_lock);
1812 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1813 struct kvm_dirty_log *log)
1815 int r;
1816 unsigned long n;
1817 struct kvm_memory_slot *memslot;
1818 int is_dirty = 0;
1820 mutex_lock(&kvm->slots_lock);
1822 r = -EINVAL;
1823 if (log->slot >= KVM_MEMORY_SLOTS)
1824 goto out;
1826 memslot = id_to_memslot(kvm->memslots, log->slot);
1827 r = -ENOENT;
1828 if (!memslot->dirty_bitmap)
1829 goto out;
1831 kvm_ia64_sync_dirty_log(kvm, memslot);
1832 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1833 if (r)
1834 goto out;
1836 /* If nothing is dirty, don't bother messing with page tables. */
1837 if (is_dirty) {
1838 kvm_flush_remote_tlbs(kvm);
1839 n = kvm_dirty_bitmap_bytes(memslot);
1840 memset(memslot->dirty_bitmap, 0, n);
1842 r = 0;
1843 out:
1844 mutex_unlock(&kvm->slots_lock);
1845 return r;
1848 int kvm_arch_hardware_setup(void)
1850 return 0;
1853 void kvm_arch_hardware_unsetup(void)
1857 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1859 int me;
1860 int cpu = vcpu->cpu;
1862 if (waitqueue_active(&vcpu->wq))
1863 wake_up_interruptible(&vcpu->wq);
1865 me = get_cpu();
1866 if (cpu != me && (unsigned) cpu < nr_cpu_ids && cpu_online(cpu))
1867 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
1868 smp_send_reschedule(cpu);
1869 put_cpu();
1872 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
1874 return __apic_accept_irq(vcpu, irq->vector);
1877 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1879 return apic->vcpu->vcpu_id == dest;
1882 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1884 return 0;
1887 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1889 return vcpu1->arch.xtp - vcpu2->arch.xtp;
1892 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
1893 int short_hand, int dest, int dest_mode)
1895 struct kvm_lapic *target = vcpu->arch.apic;
1896 return (dest_mode == 0) ?
1897 kvm_apic_match_physical_addr(target, dest) :
1898 kvm_apic_match_logical_addr(target, dest);
1901 static int find_highest_bits(int *dat)
1903 u32 bits, bitnum;
1904 int i;
1906 /* loop for all 256 bits */
1907 for (i = 7; i >= 0 ; i--) {
1908 bits = dat[i];
1909 if (bits) {
1910 bitnum = fls(bits);
1911 return i * 32 + bitnum - 1;
1915 return -1;
1918 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1920 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1922 if (vpd->irr[0] & (1UL << NMI_VECTOR))
1923 return NMI_VECTOR;
1924 if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1925 return ExtINT_VECTOR;
1927 return find_highest_bits((int *)&vpd->irr[0]);
1930 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1932 return vcpu->arch.timer_fired;
1935 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1937 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE) ||
1938 (kvm_highest_pending_irq(vcpu) != -1);
1941 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1942 struct kvm_mp_state *mp_state)
1944 mp_state->mp_state = vcpu->arch.mp_state;
1945 return 0;
1948 static int vcpu_reset(struct kvm_vcpu *vcpu)
1950 int r;
1951 long psr;
1952 local_irq_save(psr);
1953 r = kvm_insert_vmm_mapping(vcpu);
1954 local_irq_restore(psr);
1955 if (r)
1956 goto fail;
1958 vcpu->arch.launched = 0;
1959 kvm_arch_vcpu_uninit(vcpu);
1960 r = kvm_arch_vcpu_init(vcpu);
1961 if (r)
1962 goto fail;
1964 kvm_purge_vmm_mapping(vcpu);
1965 r = 0;
1966 fail:
1967 return r;
1970 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1971 struct kvm_mp_state *mp_state)
1973 int r = 0;
1975 vcpu->arch.mp_state = mp_state->mp_state;
1976 if (vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)
1977 r = vcpu_reset(vcpu);
1978 return r;