Merge branch 'kvm-updates/2.6.36' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-2.6/next.git] / arch / ia64 / kvm / kvm-ia64.c
blobf56a6316e134dc6b1b48f6828e06ee1aefc30a49
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>
37 #include <asm/pgtable.h>
38 #include <asm/gcc_intrin.h>
39 #include <asm/pal.h>
40 #include <asm/cacheflush.h>
41 #include <asm/div64.h>
42 #include <asm/tlb.h>
43 #include <asm/elf.h>
44 #include <asm/sn/addrs.h>
45 #include <asm/sn/clksupport.h>
46 #include <asm/sn/shub_mmr.h>
48 #include "misc.h"
49 #include "vti.h"
50 #include "iodev.h"
51 #include "ioapic.h"
52 #include "lapic.h"
53 #include "irq.h"
55 static unsigned long kvm_vmm_base;
56 static unsigned long kvm_vsa_base;
57 static unsigned long kvm_vm_buffer;
58 static unsigned long kvm_vm_buffer_size;
59 unsigned long kvm_vmm_gp;
61 static long vp_env_info;
63 static struct kvm_vmm_info *kvm_vmm_info;
65 static DEFINE_PER_CPU(struct kvm_vcpu *, last_vcpu);
67 struct kvm_stats_debugfs_item debugfs_entries[] = {
68 { NULL }
71 static unsigned long kvm_get_itc(struct kvm_vcpu *vcpu)
73 #if defined(CONFIG_IA64_SGI_SN2) || defined(CONFIG_IA64_GENERIC)
74 if (vcpu->kvm->arch.is_sn2)
75 return rtc_time();
76 else
77 #endif
78 return ia64_getreg(_IA64_REG_AR_ITC);
81 static void kvm_flush_icache(unsigned long start, unsigned long len)
83 int l;
85 for (l = 0; l < (len + 32); l += 32)
86 ia64_fc((void *)(start + l));
88 ia64_sync_i();
89 ia64_srlz_i();
92 static void kvm_flush_tlb_all(void)
94 unsigned long i, j, count0, count1, stride0, stride1, addr;
95 long flags;
97 addr = local_cpu_data->ptce_base;
98 count0 = local_cpu_data->ptce_count[0];
99 count1 = local_cpu_data->ptce_count[1];
100 stride0 = local_cpu_data->ptce_stride[0];
101 stride1 = local_cpu_data->ptce_stride[1];
103 local_irq_save(flags);
104 for (i = 0; i < count0; ++i) {
105 for (j = 0; j < count1; ++j) {
106 ia64_ptce(addr);
107 addr += stride1;
109 addr += stride0;
111 local_irq_restore(flags);
112 ia64_srlz_i(); /* srlz.i implies srlz.d */
115 long ia64_pal_vp_create(u64 *vpd, u64 *host_iva, u64 *opt_handler)
117 struct ia64_pal_retval iprv;
119 PAL_CALL_STK(iprv, PAL_VP_CREATE, (u64)vpd, (u64)host_iva,
120 (u64)opt_handler);
122 return iprv.status;
125 static DEFINE_SPINLOCK(vp_lock);
127 int kvm_arch_hardware_enable(void *garbage)
129 long status;
130 long tmp_base;
131 unsigned long pte;
132 unsigned long saved_psr;
133 int slot;
135 pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
136 local_irq_save(saved_psr);
137 slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
138 local_irq_restore(saved_psr);
139 if (slot < 0)
140 return -EINVAL;
142 spin_lock(&vp_lock);
143 status = ia64_pal_vp_init_env(kvm_vsa_base ?
144 VP_INIT_ENV : VP_INIT_ENV_INITALIZE,
145 __pa(kvm_vm_buffer), KVM_VM_BUFFER_BASE, &tmp_base);
146 if (status != 0) {
147 spin_unlock(&vp_lock);
148 printk(KERN_WARNING"kvm: Failed to Enable VT Support!!!!\n");
149 return -EINVAL;
152 if (!kvm_vsa_base) {
153 kvm_vsa_base = tmp_base;
154 printk(KERN_INFO"kvm: kvm_vsa_base:0x%lx\n", kvm_vsa_base);
156 spin_unlock(&vp_lock);
157 ia64_ptr_entry(0x3, slot);
159 return 0;
162 void kvm_arch_hardware_disable(void *garbage)
165 long status;
166 int slot;
167 unsigned long pte;
168 unsigned long saved_psr;
169 unsigned long host_iva = ia64_getreg(_IA64_REG_CR_IVA);
171 pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base),
172 PAGE_KERNEL));
174 local_irq_save(saved_psr);
175 slot = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
176 local_irq_restore(saved_psr);
177 if (slot < 0)
178 return;
180 status = ia64_pal_vp_exit_env(host_iva);
181 if (status)
182 printk(KERN_DEBUG"kvm: Failed to disable VT support! :%ld\n",
183 status);
184 ia64_ptr_entry(0x3, slot);
187 void kvm_arch_check_processor_compat(void *rtn)
189 *(int *)rtn = 0;
192 int kvm_dev_ioctl_check_extension(long ext)
195 int r;
197 switch (ext) {
198 case KVM_CAP_IRQCHIP:
199 case KVM_CAP_MP_STATE:
200 case KVM_CAP_IRQ_INJECT_STATUS:
201 r = 1;
202 break;
203 case KVM_CAP_COALESCED_MMIO:
204 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
205 break;
206 case KVM_CAP_IOMMU:
207 r = iommu_found();
208 break;
209 default:
210 r = 0;
212 return r;
216 static int handle_vm_error(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
218 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
219 kvm_run->hw.hardware_exit_reason = 1;
220 return 0;
223 static int handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
225 struct kvm_mmio_req *p;
226 struct kvm_io_device *mmio_dev;
227 int r;
229 p = kvm_get_vcpu_ioreq(vcpu);
231 if ((p->addr & PAGE_MASK) == IOAPIC_DEFAULT_BASE_ADDRESS)
232 goto mmio;
233 vcpu->mmio_needed = 1;
234 vcpu->mmio_phys_addr = kvm_run->mmio.phys_addr = p->addr;
235 vcpu->mmio_size = kvm_run->mmio.len = p->size;
236 vcpu->mmio_is_write = kvm_run->mmio.is_write = !p->dir;
238 if (vcpu->mmio_is_write)
239 memcpy(vcpu->mmio_data, &p->data, p->size);
240 memcpy(kvm_run->mmio.data, &p->data, p->size);
241 kvm_run->exit_reason = KVM_EXIT_MMIO;
242 return 0;
243 mmio:
244 if (p->dir)
245 r = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, p->addr,
246 p->size, &p->data);
247 else
248 r = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, p->addr,
249 p->size, &p->data);
250 if (r)
251 printk(KERN_ERR"kvm: No iodevice found! addr:%lx\n", p->addr);
252 p->state = STATE_IORESP_READY;
254 return 1;
257 static int handle_pal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
259 struct exit_ctl_data *p;
261 p = kvm_get_exit_data(vcpu);
263 if (p->exit_reason == EXIT_REASON_PAL_CALL)
264 return kvm_pal_emul(vcpu, kvm_run);
265 else {
266 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
267 kvm_run->hw.hardware_exit_reason = 2;
268 return 0;
272 static int handle_sal_call(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
274 struct exit_ctl_data *p;
276 p = kvm_get_exit_data(vcpu);
278 if (p->exit_reason == EXIT_REASON_SAL_CALL) {
279 kvm_sal_emul(vcpu);
280 return 1;
281 } else {
282 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
283 kvm_run->hw.hardware_exit_reason = 3;
284 return 0;
289 static int __apic_accept_irq(struct kvm_vcpu *vcpu, uint64_t vector)
291 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
293 if (!test_and_set_bit(vector, &vpd->irr[0])) {
294 vcpu->arch.irq_new_pending = 1;
295 kvm_vcpu_kick(vcpu);
296 return 1;
298 return 0;
302 * offset: address offset to IPI space.
303 * value: deliver value.
305 static void vcpu_deliver_ipi(struct kvm_vcpu *vcpu, uint64_t dm,
306 uint64_t vector)
308 switch (dm) {
309 case SAPIC_FIXED:
310 break;
311 case SAPIC_NMI:
312 vector = 2;
313 break;
314 case SAPIC_EXTINT:
315 vector = 0;
316 break;
317 case SAPIC_INIT:
318 case SAPIC_PMI:
319 default:
320 printk(KERN_ERR"kvm: Unimplemented Deliver reserved IPI!\n");
321 return;
323 __apic_accept_irq(vcpu, vector);
326 static struct kvm_vcpu *lid_to_vcpu(struct kvm *kvm, unsigned long id,
327 unsigned long eid)
329 union ia64_lid lid;
330 int i;
331 struct kvm_vcpu *vcpu;
333 kvm_for_each_vcpu(i, vcpu, kvm) {
334 lid.val = VCPU_LID(vcpu);
335 if (lid.id == id && lid.eid == eid)
336 return vcpu;
339 return NULL;
342 static int handle_ipi(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
344 struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
345 struct kvm_vcpu *target_vcpu;
346 struct kvm_pt_regs *regs;
347 union ia64_ipi_a addr = p->u.ipi_data.addr;
348 union ia64_ipi_d data = p->u.ipi_data.data;
350 target_vcpu = lid_to_vcpu(vcpu->kvm, addr.id, addr.eid);
351 if (!target_vcpu)
352 return handle_vm_error(vcpu, kvm_run);
354 if (!target_vcpu->arch.launched) {
355 regs = vcpu_regs(target_vcpu);
357 regs->cr_iip = vcpu->kvm->arch.rdv_sal_data.boot_ip;
358 regs->r1 = vcpu->kvm->arch.rdv_sal_data.boot_gp;
360 target_vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
361 if (waitqueue_active(&target_vcpu->wq))
362 wake_up_interruptible(&target_vcpu->wq);
363 } else {
364 vcpu_deliver_ipi(target_vcpu, data.dm, data.vector);
365 if (target_vcpu != vcpu)
366 kvm_vcpu_kick(target_vcpu);
369 return 1;
372 struct call_data {
373 struct kvm_ptc_g ptc_g_data;
374 struct kvm_vcpu *vcpu;
377 static void vcpu_global_purge(void *info)
379 struct call_data *p = (struct call_data *)info;
380 struct kvm_vcpu *vcpu = p->vcpu;
382 if (test_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
383 return;
385 set_bit(KVM_REQ_PTC_G, &vcpu->requests);
386 if (vcpu->arch.ptc_g_count < MAX_PTC_G_NUM) {
387 vcpu->arch.ptc_g_data[vcpu->arch.ptc_g_count++] =
388 p->ptc_g_data;
389 } else {
390 clear_bit(KVM_REQ_PTC_G, &vcpu->requests);
391 vcpu->arch.ptc_g_count = 0;
392 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
396 static int handle_global_purge(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
398 struct exit_ctl_data *p = kvm_get_exit_data(vcpu);
399 struct kvm *kvm = vcpu->kvm;
400 struct call_data call_data;
401 int i;
402 struct kvm_vcpu *vcpui;
404 call_data.ptc_g_data = p->u.ptc_g_data;
406 kvm_for_each_vcpu(i, vcpui, kvm) {
407 if (vcpui->arch.mp_state == KVM_MP_STATE_UNINITIALIZED ||
408 vcpu == vcpui)
409 continue;
411 if (waitqueue_active(&vcpui->wq))
412 wake_up_interruptible(&vcpui->wq);
414 if (vcpui->cpu != -1) {
415 call_data.vcpu = vcpui;
416 smp_call_function_single(vcpui->cpu,
417 vcpu_global_purge, &call_data, 1);
418 } else
419 printk(KERN_WARNING"kvm: Uninit vcpu received ipi!\n");
422 return 1;
425 static int handle_switch_rr6(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
427 return 1;
430 static int kvm_sn2_setup_mappings(struct kvm_vcpu *vcpu)
432 unsigned long pte, rtc_phys_addr, map_addr;
433 int slot;
435 map_addr = KVM_VMM_BASE + (1UL << KVM_VMM_SHIFT);
436 rtc_phys_addr = LOCAL_MMR_OFFSET | SH_RTC;
437 pte = pte_val(mk_pte_phys(rtc_phys_addr, PAGE_KERNEL_UC));
438 slot = ia64_itr_entry(0x3, map_addr, pte, PAGE_SHIFT);
439 vcpu->arch.sn_rtc_tr_slot = slot;
440 if (slot < 0) {
441 printk(KERN_ERR "Mayday mayday! RTC mapping failed!\n");
442 slot = 0;
444 return slot;
447 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
450 ktime_t kt;
451 long itc_diff;
452 unsigned long vcpu_now_itc;
453 unsigned long expires;
454 struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
455 unsigned long cyc_per_usec = local_cpu_data->cyc_per_usec;
456 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
458 if (irqchip_in_kernel(vcpu->kvm)) {
460 vcpu_now_itc = kvm_get_itc(vcpu) + vcpu->arch.itc_offset;
462 if (time_after(vcpu_now_itc, vpd->itm)) {
463 vcpu->arch.timer_check = 1;
464 return 1;
466 itc_diff = vpd->itm - vcpu_now_itc;
467 if (itc_diff < 0)
468 itc_diff = -itc_diff;
470 expires = div64_u64(itc_diff, cyc_per_usec);
471 kt = ktime_set(0, 1000 * expires);
473 vcpu->arch.ht_active = 1;
474 hrtimer_start(p_ht, kt, HRTIMER_MODE_ABS);
476 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
477 kvm_vcpu_block(vcpu);
478 hrtimer_cancel(p_ht);
479 vcpu->arch.ht_active = 0;
481 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests) ||
482 kvm_cpu_has_pending_timer(vcpu))
483 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
484 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
486 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
487 return -EINTR;
488 return 1;
489 } else {
490 printk(KERN_ERR"kvm: Unsupported userspace halt!");
491 return 0;
495 static int handle_vm_shutdown(struct kvm_vcpu *vcpu,
496 struct kvm_run *kvm_run)
498 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
499 return 0;
502 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
503 struct kvm_run *kvm_run)
505 return 1;
508 static int handle_vcpu_debug(struct kvm_vcpu *vcpu,
509 struct kvm_run *kvm_run)
511 printk("VMM: %s", vcpu->arch.log_buf);
512 return 1;
515 static int (*kvm_vti_exit_handlers[])(struct kvm_vcpu *vcpu,
516 struct kvm_run *kvm_run) = {
517 [EXIT_REASON_VM_PANIC] = handle_vm_error,
518 [EXIT_REASON_MMIO_INSTRUCTION] = handle_mmio,
519 [EXIT_REASON_PAL_CALL] = handle_pal_call,
520 [EXIT_REASON_SAL_CALL] = handle_sal_call,
521 [EXIT_REASON_SWITCH_RR6] = handle_switch_rr6,
522 [EXIT_REASON_VM_DESTROY] = handle_vm_shutdown,
523 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
524 [EXIT_REASON_IPI] = handle_ipi,
525 [EXIT_REASON_PTC_G] = handle_global_purge,
526 [EXIT_REASON_DEBUG] = handle_vcpu_debug,
530 static const int kvm_vti_max_exit_handlers =
531 sizeof(kvm_vti_exit_handlers)/sizeof(*kvm_vti_exit_handlers);
533 static uint32_t kvm_get_exit_reason(struct kvm_vcpu *vcpu)
535 struct exit_ctl_data *p_exit_data;
537 p_exit_data = kvm_get_exit_data(vcpu);
538 return p_exit_data->exit_reason;
542 * The guest has exited. See if we can fix it or if we need userspace
543 * assistance.
545 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
547 u32 exit_reason = kvm_get_exit_reason(vcpu);
548 vcpu->arch.last_exit = exit_reason;
550 if (exit_reason < kvm_vti_max_exit_handlers
551 && kvm_vti_exit_handlers[exit_reason])
552 return kvm_vti_exit_handlers[exit_reason](vcpu, kvm_run);
553 else {
554 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
555 kvm_run->hw.hardware_exit_reason = exit_reason;
557 return 0;
560 static inline void vti_set_rr6(unsigned long rr6)
562 ia64_set_rr(RR6, rr6);
563 ia64_srlz_i();
566 static int kvm_insert_vmm_mapping(struct kvm_vcpu *vcpu)
568 unsigned long pte;
569 struct kvm *kvm = vcpu->kvm;
570 int r;
572 /*Insert a pair of tr to map vmm*/
573 pte = pte_val(mk_pte_phys(__pa(kvm_vmm_base), PAGE_KERNEL));
574 r = ia64_itr_entry(0x3, KVM_VMM_BASE, pte, KVM_VMM_SHIFT);
575 if (r < 0)
576 goto out;
577 vcpu->arch.vmm_tr_slot = r;
578 /*Insert a pairt of tr to map data of vm*/
579 pte = pte_val(mk_pte_phys(__pa(kvm->arch.vm_base), PAGE_KERNEL));
580 r = ia64_itr_entry(0x3, KVM_VM_DATA_BASE,
581 pte, KVM_VM_DATA_SHIFT);
582 if (r < 0)
583 goto out;
584 vcpu->arch.vm_tr_slot = r;
586 #if defined(CONFIG_IA64_SGI_SN2) || defined(CONFIG_IA64_GENERIC)
587 if (kvm->arch.is_sn2) {
588 r = kvm_sn2_setup_mappings(vcpu);
589 if (r < 0)
590 goto out;
592 #endif
594 r = 0;
595 out:
596 return r;
599 static void kvm_purge_vmm_mapping(struct kvm_vcpu *vcpu)
601 struct kvm *kvm = vcpu->kvm;
602 ia64_ptr_entry(0x3, vcpu->arch.vmm_tr_slot);
603 ia64_ptr_entry(0x3, vcpu->arch.vm_tr_slot);
604 #if defined(CONFIG_IA64_SGI_SN2) || defined(CONFIG_IA64_GENERIC)
605 if (kvm->arch.is_sn2)
606 ia64_ptr_entry(0x3, vcpu->arch.sn_rtc_tr_slot);
607 #endif
610 static int kvm_vcpu_pre_transition(struct kvm_vcpu *vcpu)
612 unsigned long psr;
613 int r;
614 int cpu = smp_processor_id();
616 if (vcpu->arch.last_run_cpu != cpu ||
617 per_cpu(last_vcpu, cpu) != vcpu) {
618 per_cpu(last_vcpu, cpu) = vcpu;
619 vcpu->arch.last_run_cpu = cpu;
620 kvm_flush_tlb_all();
623 vcpu->arch.host_rr6 = ia64_get_rr(RR6);
624 vti_set_rr6(vcpu->arch.vmm_rr);
625 local_irq_save(psr);
626 r = kvm_insert_vmm_mapping(vcpu);
627 local_irq_restore(psr);
628 return r;
631 static void kvm_vcpu_post_transition(struct kvm_vcpu *vcpu)
633 kvm_purge_vmm_mapping(vcpu);
634 vti_set_rr6(vcpu->arch.host_rr6);
637 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
639 union context *host_ctx, *guest_ctx;
640 int r, idx;
642 idx = srcu_read_lock(&vcpu->kvm->srcu);
644 again:
645 if (signal_pending(current)) {
646 r = -EINTR;
647 kvm_run->exit_reason = KVM_EXIT_INTR;
648 goto out;
651 preempt_disable();
652 local_irq_disable();
654 /*Get host and guest context with guest address space.*/
655 host_ctx = kvm_get_host_context(vcpu);
656 guest_ctx = kvm_get_guest_context(vcpu);
658 clear_bit(KVM_REQ_KICK, &vcpu->requests);
660 r = kvm_vcpu_pre_transition(vcpu);
661 if (r < 0)
662 goto vcpu_run_fail;
664 srcu_read_unlock(&vcpu->kvm->srcu, idx);
665 kvm_guest_enter();
668 * Transition to the guest
670 kvm_vmm_info->tramp_entry(host_ctx, guest_ctx);
672 kvm_vcpu_post_transition(vcpu);
674 vcpu->arch.launched = 1;
675 set_bit(KVM_REQ_KICK, &vcpu->requests);
676 local_irq_enable();
679 * We must have an instruction between local_irq_enable() and
680 * kvm_guest_exit(), so the timer interrupt isn't delayed by
681 * the interrupt shadow. The stat.exits increment will do nicely.
682 * But we need to prevent reordering, hence this barrier():
684 barrier();
685 kvm_guest_exit();
686 preempt_enable();
688 idx = srcu_read_lock(&vcpu->kvm->srcu);
690 r = kvm_handle_exit(kvm_run, vcpu);
692 if (r > 0) {
693 if (!need_resched())
694 goto again;
697 out:
698 srcu_read_unlock(&vcpu->kvm->srcu, idx);
699 if (r > 0) {
700 kvm_resched(vcpu);
701 idx = srcu_read_lock(&vcpu->kvm->srcu);
702 goto again;
705 return r;
707 vcpu_run_fail:
708 local_irq_enable();
709 preempt_enable();
710 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
711 goto out;
714 static void kvm_set_mmio_data(struct kvm_vcpu *vcpu)
716 struct kvm_mmio_req *p = kvm_get_vcpu_ioreq(vcpu);
718 if (!vcpu->mmio_is_write)
719 memcpy(&p->data, vcpu->mmio_data, 8);
720 p->state = STATE_IORESP_READY;
723 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
725 int r;
726 sigset_t sigsaved;
728 if (vcpu->sigset_active)
729 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
731 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
732 kvm_vcpu_block(vcpu);
733 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
734 r = -EAGAIN;
735 goto out;
738 if (vcpu->mmio_needed) {
739 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
740 kvm_set_mmio_data(vcpu);
741 vcpu->mmio_read_completed = 1;
742 vcpu->mmio_needed = 0;
744 r = __vcpu_run(vcpu, kvm_run);
745 out:
746 if (vcpu->sigset_active)
747 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
749 return r;
752 static struct kvm *kvm_alloc_kvm(void)
755 struct kvm *kvm;
756 uint64_t vm_base;
758 BUG_ON(sizeof(struct kvm) > KVM_VM_STRUCT_SIZE);
760 vm_base = __get_free_pages(GFP_KERNEL, get_order(KVM_VM_DATA_SIZE));
762 if (!vm_base)
763 return ERR_PTR(-ENOMEM);
765 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
766 kvm = (struct kvm *)(vm_base +
767 offsetof(struct kvm_vm_data, kvm_vm_struct));
768 kvm->arch.vm_base = vm_base;
769 printk(KERN_DEBUG"kvm: vm's data area:0x%lx\n", vm_base);
771 return kvm;
774 struct kvm_io_range {
775 unsigned long start;
776 unsigned long size;
777 unsigned long type;
780 static const struct kvm_io_range io_ranges[] = {
781 {VGA_IO_START, VGA_IO_SIZE, GPFN_FRAME_BUFFER},
782 {MMIO_START, MMIO_SIZE, GPFN_LOW_MMIO},
783 {LEGACY_IO_START, LEGACY_IO_SIZE, GPFN_LEGACY_IO},
784 {IO_SAPIC_START, IO_SAPIC_SIZE, GPFN_IOSAPIC},
785 {PIB_START, PIB_SIZE, GPFN_PIB},
788 static void kvm_build_io_pmt(struct kvm *kvm)
790 unsigned long i, j;
792 /* Mark I/O ranges */
793 for (i = 0; i < (sizeof(io_ranges) / sizeof(struct kvm_io_range));
794 i++) {
795 for (j = io_ranges[i].start;
796 j < io_ranges[i].start + io_ranges[i].size;
797 j += PAGE_SIZE)
798 kvm_set_pmt_entry(kvm, j >> PAGE_SHIFT,
799 io_ranges[i].type, 0);
804 /*Use unused rids to virtualize guest rid.*/
805 #define GUEST_PHYSICAL_RR0 0x1739
806 #define GUEST_PHYSICAL_RR4 0x2739
807 #define VMM_INIT_RR 0x1660
809 static void kvm_init_vm(struct kvm *kvm)
811 BUG_ON(!kvm);
813 kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
814 kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
815 kvm->arch.vmm_init_rr = VMM_INIT_RR;
818 *Fill P2M entries for MMIO/IO ranges
820 kvm_build_io_pmt(kvm);
822 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
824 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
825 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
828 struct kvm *kvm_arch_create_vm(void)
830 struct kvm *kvm = kvm_alloc_kvm();
832 if (IS_ERR(kvm))
833 return ERR_PTR(-ENOMEM);
835 kvm->arch.is_sn2 = ia64_platform_is("sn2");
837 kvm_init_vm(kvm);
839 return kvm;
843 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
844 struct kvm_irqchip *chip)
846 int r;
848 r = 0;
849 switch (chip->chip_id) {
850 case KVM_IRQCHIP_IOAPIC:
851 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
852 break;
853 default:
854 r = -EINVAL;
855 break;
857 return r;
860 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
862 int r;
864 r = 0;
865 switch (chip->chip_id) {
866 case KVM_IRQCHIP_IOAPIC:
867 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
868 break;
869 default:
870 r = -EINVAL;
871 break;
873 return r;
876 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
878 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
880 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
881 int i;
883 for (i = 0; i < 16; i++) {
884 vpd->vgr[i] = regs->vpd.vgr[i];
885 vpd->vbgr[i] = regs->vpd.vbgr[i];
887 for (i = 0; i < 128; i++)
888 vpd->vcr[i] = regs->vpd.vcr[i];
889 vpd->vhpi = regs->vpd.vhpi;
890 vpd->vnat = regs->vpd.vnat;
891 vpd->vbnat = regs->vpd.vbnat;
892 vpd->vpsr = regs->vpd.vpsr;
894 vpd->vpr = regs->vpd.vpr;
896 memcpy(&vcpu->arch.guest, &regs->saved_guest, sizeof(union context));
898 RESTORE_REGS(mp_state);
899 RESTORE_REGS(vmm_rr);
900 memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
901 memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
902 RESTORE_REGS(itr_regions);
903 RESTORE_REGS(dtr_regions);
904 RESTORE_REGS(tc_regions);
905 RESTORE_REGS(irq_check);
906 RESTORE_REGS(itc_check);
907 RESTORE_REGS(timer_check);
908 RESTORE_REGS(timer_pending);
909 RESTORE_REGS(last_itc);
910 for (i = 0; i < 8; i++) {
911 vcpu->arch.vrr[i] = regs->vrr[i];
912 vcpu->arch.ibr[i] = regs->ibr[i];
913 vcpu->arch.dbr[i] = regs->dbr[i];
915 for (i = 0; i < 4; i++)
916 vcpu->arch.insvc[i] = regs->insvc[i];
917 RESTORE_REGS(xtp);
918 RESTORE_REGS(metaphysical_rr0);
919 RESTORE_REGS(metaphysical_rr4);
920 RESTORE_REGS(metaphysical_saved_rr0);
921 RESTORE_REGS(metaphysical_saved_rr4);
922 RESTORE_REGS(fp_psr);
923 RESTORE_REGS(saved_gp);
925 vcpu->arch.irq_new_pending = 1;
926 vcpu->arch.itc_offset = regs->saved_itc - kvm_get_itc(vcpu);
927 set_bit(KVM_REQ_RESUME, &vcpu->requests);
929 return 0;
932 long kvm_arch_vm_ioctl(struct file *filp,
933 unsigned int ioctl, unsigned long arg)
935 struct kvm *kvm = filp->private_data;
936 void __user *argp = (void __user *)arg;
937 int r = -ENOTTY;
939 switch (ioctl) {
940 case KVM_SET_MEMORY_REGION: {
941 struct kvm_memory_region kvm_mem;
942 struct kvm_userspace_memory_region kvm_userspace_mem;
944 r = -EFAULT;
945 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
946 goto out;
947 kvm_userspace_mem.slot = kvm_mem.slot;
948 kvm_userspace_mem.flags = kvm_mem.flags;
949 kvm_userspace_mem.guest_phys_addr =
950 kvm_mem.guest_phys_addr;
951 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
952 r = kvm_vm_ioctl_set_memory_region(kvm,
953 &kvm_userspace_mem, 0);
954 if (r)
955 goto out;
956 break;
958 case KVM_CREATE_IRQCHIP:
959 r = -EFAULT;
960 r = kvm_ioapic_init(kvm);
961 if (r)
962 goto out;
963 r = kvm_setup_default_irq_routing(kvm);
964 if (r) {
965 kvm_ioapic_destroy(kvm);
966 goto out;
968 break;
969 case KVM_IRQ_LINE_STATUS:
970 case KVM_IRQ_LINE: {
971 struct kvm_irq_level irq_event;
973 r = -EFAULT;
974 if (copy_from_user(&irq_event, argp, sizeof irq_event))
975 goto out;
976 r = -ENXIO;
977 if (irqchip_in_kernel(kvm)) {
978 __s32 status;
979 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
980 irq_event.irq, irq_event.level);
981 if (ioctl == KVM_IRQ_LINE_STATUS) {
982 r = -EFAULT;
983 irq_event.status = status;
984 if (copy_to_user(argp, &irq_event,
985 sizeof irq_event))
986 goto out;
988 r = 0;
990 break;
992 case KVM_GET_IRQCHIP: {
993 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
994 struct kvm_irqchip chip;
996 r = -EFAULT;
997 if (copy_from_user(&chip, argp, sizeof chip))
998 goto out;
999 r = -ENXIO;
1000 if (!irqchip_in_kernel(kvm))
1001 goto out;
1002 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1003 if (r)
1004 goto out;
1005 r = -EFAULT;
1006 if (copy_to_user(argp, &chip, sizeof chip))
1007 goto out;
1008 r = 0;
1009 break;
1011 case KVM_SET_IRQCHIP: {
1012 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1013 struct kvm_irqchip chip;
1015 r = -EFAULT;
1016 if (copy_from_user(&chip, argp, sizeof chip))
1017 goto out;
1018 r = -ENXIO;
1019 if (!irqchip_in_kernel(kvm))
1020 goto out;
1021 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1022 if (r)
1023 goto out;
1024 r = 0;
1025 break;
1027 default:
1030 out:
1031 return r;
1034 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1035 struct kvm_sregs *sregs)
1037 return -EINVAL;
1040 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1041 struct kvm_sregs *sregs)
1043 return -EINVAL;
1046 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1047 struct kvm_translation *tr)
1050 return -EINVAL;
1053 static int kvm_alloc_vmm_area(void)
1055 if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1056 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1057 get_order(KVM_VMM_SIZE));
1058 if (!kvm_vmm_base)
1059 return -ENOMEM;
1061 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1062 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1064 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1065 kvm_vmm_base, kvm_vm_buffer);
1068 return 0;
1071 static void kvm_free_vmm_area(void)
1073 if (kvm_vmm_base) {
1074 /*Zero this area before free to avoid bits leak!!*/
1075 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1076 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1077 kvm_vmm_base = 0;
1078 kvm_vm_buffer = 0;
1079 kvm_vsa_base = 0;
1083 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1085 int i;
1086 union cpuid3_t cpuid3;
1087 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1089 if (IS_ERR(vpd))
1090 return PTR_ERR(vpd);
1092 /* CPUID init */
1093 for (i = 0; i < 5; i++)
1094 vpd->vcpuid[i] = ia64_get_cpuid(i);
1096 /* Limit the CPUID number to 5 */
1097 cpuid3.value = vpd->vcpuid[3];
1098 cpuid3.number = 4; /* 5 - 1 */
1099 vpd->vcpuid[3] = cpuid3.value;
1101 /*Set vac and vdc fields*/
1102 vpd->vac.a_from_int_cr = 1;
1103 vpd->vac.a_to_int_cr = 1;
1104 vpd->vac.a_from_psr = 1;
1105 vpd->vac.a_from_cpuid = 1;
1106 vpd->vac.a_cover = 1;
1107 vpd->vac.a_bsw = 1;
1108 vpd->vac.a_int = 1;
1109 vpd->vdc.d_vmsw = 1;
1111 /*Set virtual buffer*/
1112 vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1114 return 0;
1117 static int vti_create_vp(struct kvm_vcpu *vcpu)
1119 long ret;
1120 struct vpd *vpd = vcpu->arch.vpd;
1121 unsigned long vmm_ivt;
1123 vmm_ivt = kvm_vmm_info->vmm_ivt;
1125 printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1127 ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1129 if (ret) {
1130 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1131 return -EINVAL;
1133 return 0;
1136 static void init_ptce_info(struct kvm_vcpu *vcpu)
1138 ia64_ptce_info_t ptce = {0};
1140 ia64_get_ptce(&ptce);
1141 vcpu->arch.ptce_base = ptce.base;
1142 vcpu->arch.ptce_count[0] = ptce.count[0];
1143 vcpu->arch.ptce_count[1] = ptce.count[1];
1144 vcpu->arch.ptce_stride[0] = ptce.stride[0];
1145 vcpu->arch.ptce_stride[1] = ptce.stride[1];
1148 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1150 struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1152 if (hrtimer_cancel(p_ht))
1153 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1156 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1158 struct kvm_vcpu *vcpu;
1159 wait_queue_head_t *q;
1161 vcpu = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1162 q = &vcpu->wq;
1164 if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1165 goto out;
1167 if (waitqueue_active(q))
1168 wake_up_interruptible(q);
1170 out:
1171 vcpu->arch.timer_fired = 1;
1172 vcpu->arch.timer_check = 1;
1173 return HRTIMER_NORESTART;
1176 #define PALE_RESET_ENTRY 0x80000000ffffffb0UL
1178 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1180 struct kvm_vcpu *v;
1181 int r;
1182 int i;
1183 long itc_offset;
1184 struct kvm *kvm = vcpu->kvm;
1185 struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1187 union context *p_ctx = &vcpu->arch.guest;
1188 struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1190 /*Init vcpu context for first run.*/
1191 if (IS_ERR(vmm_vcpu))
1192 return PTR_ERR(vmm_vcpu);
1194 if (kvm_vcpu_is_bsp(vcpu)) {
1195 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1197 /*Set entry address for first run.*/
1198 regs->cr_iip = PALE_RESET_ENTRY;
1200 /*Initialize itc offset for vcpus*/
1201 itc_offset = 0UL - kvm_get_itc(vcpu);
1202 for (i = 0; i < KVM_MAX_VCPUS; i++) {
1203 v = (struct kvm_vcpu *)((char *)vcpu +
1204 sizeof(struct kvm_vcpu_data) * i);
1205 v->arch.itc_offset = itc_offset;
1206 v->arch.last_itc = 0;
1208 } else
1209 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1211 r = -ENOMEM;
1212 vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1213 if (!vcpu->arch.apic)
1214 goto out;
1215 vcpu->arch.apic->vcpu = vcpu;
1217 p_ctx->gr[1] = 0;
1218 p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + KVM_STK_OFFSET);
1219 p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1220 p_ctx->psr = 0x1008522000UL;
1221 p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1222 p_ctx->caller_unat = 0;
1223 p_ctx->pr = 0x0;
1224 p_ctx->ar[36] = 0x0; /*unat*/
1225 p_ctx->ar[19] = 0x0; /*rnat*/
1226 p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1227 ((sizeof(struct kvm_vcpu)+15) & ~15);
1228 p_ctx->ar[64] = 0x0; /*pfs*/
1229 p_ctx->cr[0] = 0x7e04UL;
1230 p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1231 p_ctx->cr[8] = 0x3c;
1233 /*Initialize region register*/
1234 p_ctx->rr[0] = 0x30;
1235 p_ctx->rr[1] = 0x30;
1236 p_ctx->rr[2] = 0x30;
1237 p_ctx->rr[3] = 0x30;
1238 p_ctx->rr[4] = 0x30;
1239 p_ctx->rr[5] = 0x30;
1240 p_ctx->rr[7] = 0x30;
1242 /*Initialize branch register 0*/
1243 p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1245 vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1246 vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1247 vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1249 hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1250 vcpu->arch.hlt_timer.function = hlt_timer_fn;
1252 vcpu->arch.last_run_cpu = -1;
1253 vcpu->arch.vpd = (struct vpd *)VPD_BASE(vcpu->vcpu_id);
1254 vcpu->arch.vsa_base = kvm_vsa_base;
1255 vcpu->arch.__gp = kvm_vmm_gp;
1256 vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1257 vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_BASE(vcpu->vcpu_id);
1258 vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_BASE(vcpu->vcpu_id);
1259 init_ptce_info(vcpu);
1261 r = 0;
1262 out:
1263 return r;
1266 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1268 unsigned long psr;
1269 int r;
1271 local_irq_save(psr);
1272 r = kvm_insert_vmm_mapping(vcpu);
1273 local_irq_restore(psr);
1274 if (r)
1275 goto fail;
1276 r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1277 if (r)
1278 goto fail;
1280 r = vti_init_vpd(vcpu);
1281 if (r) {
1282 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1283 goto uninit;
1286 r = vti_create_vp(vcpu);
1287 if (r)
1288 goto uninit;
1290 kvm_purge_vmm_mapping(vcpu);
1292 return 0;
1293 uninit:
1294 kvm_vcpu_uninit(vcpu);
1295 fail:
1296 return r;
1299 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1300 unsigned int id)
1302 struct kvm_vcpu *vcpu;
1303 unsigned long vm_base = kvm->arch.vm_base;
1304 int r;
1305 int cpu;
1307 BUG_ON(sizeof(struct kvm_vcpu) > VCPU_STRUCT_SIZE/2);
1309 r = -EINVAL;
1310 if (id >= KVM_MAX_VCPUS) {
1311 printk(KERN_ERR"kvm: Can't configure vcpus > %ld",
1312 KVM_MAX_VCPUS);
1313 goto fail;
1316 r = -ENOMEM;
1317 if (!vm_base) {
1318 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1319 goto fail;
1321 vcpu = (struct kvm_vcpu *)(vm_base + offsetof(struct kvm_vm_data,
1322 vcpu_data[id].vcpu_struct));
1323 vcpu->kvm = kvm;
1325 cpu = get_cpu();
1326 r = vti_vcpu_setup(vcpu, id);
1327 put_cpu();
1329 if (r) {
1330 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1331 goto fail;
1334 return vcpu;
1335 fail:
1336 return ERR_PTR(r);
1339 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1341 return 0;
1344 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1346 return -EINVAL;
1349 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1351 return -EINVAL;
1354 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1355 struct kvm_guest_debug *dbg)
1357 return -EINVAL;
1360 static void free_kvm(struct kvm *kvm)
1362 unsigned long vm_base = kvm->arch.vm_base;
1364 if (vm_base) {
1365 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1366 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1371 static void kvm_release_vm_pages(struct kvm *kvm)
1373 struct kvm_memslots *slots;
1374 struct kvm_memory_slot *memslot;
1375 int i, j;
1376 unsigned long base_gfn;
1378 slots = kvm_memslots(kvm);
1379 for (i = 0; i < slots->nmemslots; i++) {
1380 memslot = &slots->memslots[i];
1381 base_gfn = memslot->base_gfn;
1383 for (j = 0; j < memslot->npages; j++) {
1384 if (memslot->rmap[j])
1385 put_page((struct page *)memslot->rmap[j]);
1390 void kvm_arch_sync_events(struct kvm *kvm)
1394 void kvm_arch_destroy_vm(struct kvm *kvm)
1396 kvm_iommu_unmap_guest(kvm);
1397 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1398 kvm_free_all_assigned_devices(kvm);
1399 #endif
1400 kfree(kvm->arch.vioapic);
1401 kvm_release_vm_pages(kvm);
1402 kvm_free_physmem(kvm);
1403 cleanup_srcu_struct(&kvm->srcu);
1404 free_kvm(kvm);
1407 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1411 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1413 if (cpu != vcpu->cpu) {
1414 vcpu->cpu = cpu;
1415 if (vcpu->arch.ht_active)
1416 kvm_migrate_hlt_timer(vcpu);
1420 #define SAVE_REGS(_x) regs->_x = vcpu->arch._x
1422 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1424 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1425 int i;
1427 vcpu_load(vcpu);
1429 for (i = 0; i < 16; i++) {
1430 regs->vpd.vgr[i] = vpd->vgr[i];
1431 regs->vpd.vbgr[i] = vpd->vbgr[i];
1433 for (i = 0; i < 128; i++)
1434 regs->vpd.vcr[i] = vpd->vcr[i];
1435 regs->vpd.vhpi = vpd->vhpi;
1436 regs->vpd.vnat = vpd->vnat;
1437 regs->vpd.vbnat = vpd->vbnat;
1438 regs->vpd.vpsr = vpd->vpsr;
1439 regs->vpd.vpr = vpd->vpr;
1441 memcpy(&regs->saved_guest, &vcpu->arch.guest, sizeof(union context));
1443 SAVE_REGS(mp_state);
1444 SAVE_REGS(vmm_rr);
1445 memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1446 memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1447 SAVE_REGS(itr_regions);
1448 SAVE_REGS(dtr_regions);
1449 SAVE_REGS(tc_regions);
1450 SAVE_REGS(irq_check);
1451 SAVE_REGS(itc_check);
1452 SAVE_REGS(timer_check);
1453 SAVE_REGS(timer_pending);
1454 SAVE_REGS(last_itc);
1455 for (i = 0; i < 8; i++) {
1456 regs->vrr[i] = vcpu->arch.vrr[i];
1457 regs->ibr[i] = vcpu->arch.ibr[i];
1458 regs->dbr[i] = vcpu->arch.dbr[i];
1460 for (i = 0; i < 4; i++)
1461 regs->insvc[i] = vcpu->arch.insvc[i];
1462 regs->saved_itc = vcpu->arch.itc_offset + kvm_get_itc(vcpu);
1463 SAVE_REGS(xtp);
1464 SAVE_REGS(metaphysical_rr0);
1465 SAVE_REGS(metaphysical_rr4);
1466 SAVE_REGS(metaphysical_saved_rr0);
1467 SAVE_REGS(metaphysical_saved_rr4);
1468 SAVE_REGS(fp_psr);
1469 SAVE_REGS(saved_gp);
1471 vcpu_put(vcpu);
1472 return 0;
1475 int kvm_arch_vcpu_ioctl_get_stack(struct kvm_vcpu *vcpu,
1476 struct kvm_ia64_vcpu_stack *stack)
1478 memcpy(stack, vcpu, sizeof(struct kvm_ia64_vcpu_stack));
1479 return 0;
1482 int kvm_arch_vcpu_ioctl_set_stack(struct kvm_vcpu *vcpu,
1483 struct kvm_ia64_vcpu_stack *stack)
1485 memcpy(vcpu + 1, &stack->stack[0] + sizeof(struct kvm_vcpu),
1486 sizeof(struct kvm_ia64_vcpu_stack) - sizeof(struct kvm_vcpu));
1488 vcpu->arch.exit_data = ((struct kvm_vcpu *)stack)->arch.exit_data;
1489 return 0;
1492 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1495 hrtimer_cancel(&vcpu->arch.hlt_timer);
1496 kfree(vcpu->arch.apic);
1500 long kvm_arch_vcpu_ioctl(struct file *filp,
1501 unsigned int ioctl, unsigned long arg)
1503 struct kvm_vcpu *vcpu = filp->private_data;
1504 void __user *argp = (void __user *)arg;
1505 struct kvm_ia64_vcpu_stack *stack = NULL;
1506 long r;
1508 switch (ioctl) {
1509 case KVM_IA64_VCPU_GET_STACK: {
1510 struct kvm_ia64_vcpu_stack __user *user_stack;
1511 void __user *first_p = argp;
1513 r = -EFAULT;
1514 if (copy_from_user(&user_stack, first_p, sizeof(void *)))
1515 goto out;
1517 if (!access_ok(VERIFY_WRITE, user_stack,
1518 sizeof(struct kvm_ia64_vcpu_stack))) {
1519 printk(KERN_INFO "KVM_IA64_VCPU_GET_STACK: "
1520 "Illegal user destination address for stack\n");
1521 goto out;
1523 stack = kzalloc(sizeof(struct kvm_ia64_vcpu_stack), GFP_KERNEL);
1524 if (!stack) {
1525 r = -ENOMEM;
1526 goto out;
1529 r = kvm_arch_vcpu_ioctl_get_stack(vcpu, stack);
1530 if (r)
1531 goto out;
1533 if (copy_to_user(user_stack, stack,
1534 sizeof(struct kvm_ia64_vcpu_stack))) {
1535 r = -EFAULT;
1536 goto out;
1539 break;
1541 case KVM_IA64_VCPU_SET_STACK: {
1542 struct kvm_ia64_vcpu_stack __user *user_stack;
1543 void __user *first_p = argp;
1545 r = -EFAULT;
1546 if (copy_from_user(&user_stack, first_p, sizeof(void *)))
1547 goto out;
1549 if (!access_ok(VERIFY_READ, user_stack,
1550 sizeof(struct kvm_ia64_vcpu_stack))) {
1551 printk(KERN_INFO "KVM_IA64_VCPU_SET_STACK: "
1552 "Illegal user address for stack\n");
1553 goto out;
1555 stack = kmalloc(sizeof(struct kvm_ia64_vcpu_stack), GFP_KERNEL);
1556 if (!stack) {
1557 r = -ENOMEM;
1558 goto out;
1560 if (copy_from_user(stack, user_stack,
1561 sizeof(struct kvm_ia64_vcpu_stack)))
1562 goto out;
1564 r = kvm_arch_vcpu_ioctl_set_stack(vcpu, stack);
1565 break;
1568 default:
1569 r = -EINVAL;
1572 out:
1573 kfree(stack);
1574 return r;
1577 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1578 struct kvm_memory_slot *memslot,
1579 struct kvm_memory_slot old,
1580 struct kvm_userspace_memory_region *mem,
1581 int user_alloc)
1583 unsigned long i;
1584 unsigned long pfn;
1585 int npages = memslot->npages;
1586 unsigned long base_gfn = memslot->base_gfn;
1588 if (base_gfn + npages > (KVM_MAX_MEM_SIZE >> PAGE_SHIFT))
1589 return -ENOMEM;
1591 for (i = 0; i < npages; i++) {
1592 pfn = gfn_to_pfn(kvm, base_gfn + i);
1593 if (!kvm_is_mmio_pfn(pfn)) {
1594 kvm_set_pmt_entry(kvm, base_gfn + i,
1595 pfn << PAGE_SHIFT,
1596 _PAGE_AR_RWX | _PAGE_MA_WB);
1597 memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1598 } else {
1599 kvm_set_pmt_entry(kvm, base_gfn + i,
1600 GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1601 _PAGE_MA_UC);
1602 memslot->rmap[i] = 0;
1606 return 0;
1609 void kvm_arch_commit_memory_region(struct kvm *kvm,
1610 struct kvm_userspace_memory_region *mem,
1611 struct kvm_memory_slot old,
1612 int user_alloc)
1614 return;
1617 void kvm_arch_flush_shadow(struct kvm *kvm)
1619 kvm_flush_remote_tlbs(kvm);
1622 long kvm_arch_dev_ioctl(struct file *filp,
1623 unsigned int ioctl, unsigned long arg)
1625 return -EINVAL;
1628 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1630 kvm_vcpu_uninit(vcpu);
1633 static int vti_cpu_has_kvm_support(void)
1635 long avail = 1, status = 1, control = 1;
1636 long ret;
1638 ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1639 if (ret)
1640 goto out;
1642 if (!(avail & PAL_PROC_VM_BIT))
1643 goto out;
1645 printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1647 ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1648 if (ret)
1649 goto out;
1650 printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1652 if (!(vp_env_info & VP_OPCODE)) {
1653 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1654 "vm_env_info:0x%lx\n", vp_env_info);
1657 return 1;
1658 out:
1659 return 0;
1664 * On SN2, the ITC isn't stable, so copy in fast path code to use the
1665 * SN2 RTC, replacing the ITC based default verion.
1667 static void kvm_patch_vmm(struct kvm_vmm_info *vmm_info,
1668 struct module *module)
1670 unsigned long new_ar, new_ar_sn2;
1671 unsigned long module_base;
1673 if (!ia64_platform_is("sn2"))
1674 return;
1676 module_base = (unsigned long)module->module_core;
1678 new_ar = kvm_vmm_base + vmm_info->patch_mov_ar - module_base;
1679 new_ar_sn2 = kvm_vmm_base + vmm_info->patch_mov_ar_sn2 - module_base;
1681 printk(KERN_INFO "kvm: Patching ITC emulation to use SGI SN2 RTC "
1682 "as source\n");
1685 * Copy the SN2 version of mov_ar into place. They are both
1686 * the same size, so 6 bundles is sufficient (6 * 0x10).
1688 memcpy((void *)new_ar, (void *)new_ar_sn2, 0x60);
1691 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1692 struct module *module)
1694 unsigned long module_base;
1695 unsigned long vmm_size;
1697 unsigned long vmm_offset, func_offset, fdesc_offset;
1698 struct fdesc *p_fdesc;
1700 BUG_ON(!module);
1702 if (!kvm_vmm_base) {
1703 printk("kvm: kvm area hasn't been initialized yet!!\n");
1704 return -EFAULT;
1707 /*Calculate new position of relocated vmm module.*/
1708 module_base = (unsigned long)module->module_core;
1709 vmm_size = module->core_size;
1710 if (unlikely(vmm_size > KVM_VMM_SIZE))
1711 return -EFAULT;
1713 memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1714 kvm_patch_vmm(vmm_info, module);
1715 kvm_flush_icache(kvm_vmm_base, vmm_size);
1717 /*Recalculate kvm_vmm_info based on new VMM*/
1718 vmm_offset = vmm_info->vmm_ivt - module_base;
1719 kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1720 printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1721 kvm_vmm_info->vmm_ivt);
1723 fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1724 kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1725 fdesc_offset);
1726 func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1727 p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1728 p_fdesc->ip = KVM_VMM_BASE + func_offset;
1729 p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1731 printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1732 KVM_VMM_BASE+func_offset);
1734 fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1735 kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1736 fdesc_offset);
1737 func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1738 p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1739 p_fdesc->ip = KVM_VMM_BASE + func_offset;
1740 p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1742 kvm_vmm_gp = p_fdesc->gp;
1744 printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1745 kvm_vmm_info->vmm_entry);
1746 printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1747 KVM_VMM_BASE + func_offset);
1749 return 0;
1752 int kvm_arch_init(void *opaque)
1754 int r;
1755 struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1757 if (!vti_cpu_has_kvm_support()) {
1758 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1759 r = -EOPNOTSUPP;
1760 goto out;
1763 if (kvm_vmm_info) {
1764 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1765 r = -EEXIST;
1766 goto out;
1769 r = -ENOMEM;
1770 kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1771 if (!kvm_vmm_info)
1772 goto out;
1774 if (kvm_alloc_vmm_area())
1775 goto out_free0;
1777 r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1778 if (r)
1779 goto out_free1;
1781 return 0;
1783 out_free1:
1784 kvm_free_vmm_area();
1785 out_free0:
1786 kfree(kvm_vmm_info);
1787 out:
1788 return r;
1791 void kvm_arch_exit(void)
1793 kvm_free_vmm_area();
1794 kfree(kvm_vmm_info);
1795 kvm_vmm_info = NULL;
1798 static void kvm_ia64_sync_dirty_log(struct kvm *kvm,
1799 struct kvm_memory_slot *memslot)
1801 int i;
1802 long base;
1803 unsigned long n;
1804 unsigned long *dirty_bitmap = (unsigned long *)(kvm->arch.vm_base +
1805 offsetof(struct kvm_vm_data, kvm_mem_dirty_log));
1807 n = kvm_dirty_bitmap_bytes(memslot);
1808 base = memslot->base_gfn / BITS_PER_LONG;
1810 spin_lock(&kvm->arch.dirty_log_lock);
1811 for (i = 0; i < n/sizeof(long); ++i) {
1812 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1813 dirty_bitmap[base + i] = 0;
1815 spin_unlock(&kvm->arch.dirty_log_lock);
1818 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1819 struct kvm_dirty_log *log)
1821 int r;
1822 unsigned long n;
1823 struct kvm_memory_slot *memslot;
1824 int is_dirty = 0;
1826 mutex_lock(&kvm->slots_lock);
1828 r = -EINVAL;
1829 if (log->slot >= KVM_MEMORY_SLOTS)
1830 goto out;
1832 memslot = &kvm->memslots->memslots[log->slot];
1833 r = -ENOENT;
1834 if (!memslot->dirty_bitmap)
1835 goto out;
1837 kvm_ia64_sync_dirty_log(kvm, memslot);
1838 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1839 if (r)
1840 goto out;
1842 /* If nothing is dirty, don't bother messing with page tables. */
1843 if (is_dirty) {
1844 kvm_flush_remote_tlbs(kvm);
1845 n = kvm_dirty_bitmap_bytes(memslot);
1846 memset(memslot->dirty_bitmap, 0, n);
1848 r = 0;
1849 out:
1850 mutex_unlock(&kvm->slots_lock);
1851 return r;
1854 int kvm_arch_hardware_setup(void)
1856 return 0;
1859 void kvm_arch_hardware_unsetup(void)
1863 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1865 int me;
1866 int cpu = vcpu->cpu;
1868 if (waitqueue_active(&vcpu->wq))
1869 wake_up_interruptible(&vcpu->wq);
1871 me = get_cpu();
1872 if (cpu != me && (unsigned) cpu < nr_cpu_ids && cpu_online(cpu))
1873 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
1874 smp_send_reschedule(cpu);
1875 put_cpu();
1878 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
1880 return __apic_accept_irq(vcpu, irq->vector);
1883 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1885 return apic->vcpu->vcpu_id == dest;
1888 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1890 return 0;
1893 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1895 return vcpu1->arch.xtp - vcpu2->arch.xtp;
1898 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
1899 int short_hand, int dest, int dest_mode)
1901 struct kvm_lapic *target = vcpu->arch.apic;
1902 return (dest_mode == 0) ?
1903 kvm_apic_match_physical_addr(target, dest) :
1904 kvm_apic_match_logical_addr(target, dest);
1907 static int find_highest_bits(int *dat)
1909 u32 bits, bitnum;
1910 int i;
1912 /* loop for all 256 bits */
1913 for (i = 7; i >= 0 ; i--) {
1914 bits = dat[i];
1915 if (bits) {
1916 bitnum = fls(bits);
1917 return i * 32 + bitnum - 1;
1921 return -1;
1924 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1926 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1928 if (vpd->irr[0] & (1UL << NMI_VECTOR))
1929 return NMI_VECTOR;
1930 if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1931 return ExtINT_VECTOR;
1933 return find_highest_bits((int *)&vpd->irr[0]);
1936 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1938 return vcpu->arch.timer_fired;
1941 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1943 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE) ||
1944 (kvm_highest_pending_irq(vcpu) != -1);
1947 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1948 struct kvm_mp_state *mp_state)
1950 mp_state->mp_state = vcpu->arch.mp_state;
1951 return 0;
1954 static int vcpu_reset(struct kvm_vcpu *vcpu)
1956 int r;
1957 long psr;
1958 local_irq_save(psr);
1959 r = kvm_insert_vmm_mapping(vcpu);
1960 local_irq_restore(psr);
1961 if (r)
1962 goto fail;
1964 vcpu->arch.launched = 0;
1965 kvm_arch_vcpu_uninit(vcpu);
1966 r = kvm_arch_vcpu_init(vcpu);
1967 if (r)
1968 goto fail;
1970 kvm_purge_vmm_mapping(vcpu);
1971 r = 0;
1972 fail:
1973 return r;
1976 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1977 struct kvm_mp_state *mp_state)
1979 int r = 0;
1981 vcpu->arch.mp_state = mp_state->mp_state;
1982 if (vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)
1983 r = vcpu_reset(vcpu);
1984 return r;