cciss: factor out cciss_enter_performant_mode
[linux-2.6/next.git] / arch / ia64 / kvm / kvm-ia64.c
blob21b701374f72335905ee9bfde866975e60800229
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 vcpu_load(vcpu);
730 if (vcpu->sigset_active)
731 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
733 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
734 kvm_vcpu_block(vcpu);
735 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
736 r = -EAGAIN;
737 goto out;
740 if (vcpu->mmio_needed) {
741 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
742 kvm_set_mmio_data(vcpu);
743 vcpu->mmio_read_completed = 1;
744 vcpu->mmio_needed = 0;
746 r = __vcpu_run(vcpu, kvm_run);
747 out:
748 if (vcpu->sigset_active)
749 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
751 vcpu_put(vcpu);
752 return r;
755 static struct kvm *kvm_alloc_kvm(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 ERR_PTR(-ENOMEM);
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_io_range {
778 unsigned long start;
779 unsigned long size;
780 unsigned long type;
783 static const struct kvm_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 static void kvm_init_vm(struct kvm *kvm)
814 BUG_ON(!kvm);
816 kvm->arch.metaphysical_rr0 = GUEST_PHYSICAL_RR0;
817 kvm->arch.metaphysical_rr4 = GUEST_PHYSICAL_RR4;
818 kvm->arch.vmm_init_rr = VMM_INIT_RR;
821 *Fill P2M entries for MMIO/IO ranges
823 kvm_build_io_pmt(kvm);
825 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
827 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
828 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
831 struct kvm *kvm_arch_create_vm(void)
833 struct kvm *kvm = kvm_alloc_kvm();
835 if (IS_ERR(kvm))
836 return ERR_PTR(-ENOMEM);
838 kvm->arch.is_sn2 = ia64_platform_is("sn2");
840 kvm_init_vm(kvm);
842 return kvm;
846 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm,
847 struct kvm_irqchip *chip)
849 int r;
851 r = 0;
852 switch (chip->chip_id) {
853 case KVM_IRQCHIP_IOAPIC:
854 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
855 break;
856 default:
857 r = -EINVAL;
858 break;
860 return r;
863 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
865 int r;
867 r = 0;
868 switch (chip->chip_id) {
869 case KVM_IRQCHIP_IOAPIC:
870 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
871 break;
872 default:
873 r = -EINVAL;
874 break;
876 return r;
879 #define RESTORE_REGS(_x) vcpu->arch._x = regs->_x
881 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
883 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
884 int i;
886 vcpu_load(vcpu);
888 for (i = 0; i < 16; i++) {
889 vpd->vgr[i] = regs->vpd.vgr[i];
890 vpd->vbgr[i] = regs->vpd.vbgr[i];
892 for (i = 0; i < 128; i++)
893 vpd->vcr[i] = regs->vpd.vcr[i];
894 vpd->vhpi = regs->vpd.vhpi;
895 vpd->vnat = regs->vpd.vnat;
896 vpd->vbnat = regs->vpd.vbnat;
897 vpd->vpsr = regs->vpd.vpsr;
899 vpd->vpr = regs->vpd.vpr;
901 memcpy(&vcpu->arch.guest, &regs->saved_guest, sizeof(union context));
903 RESTORE_REGS(mp_state);
904 RESTORE_REGS(vmm_rr);
905 memcpy(vcpu->arch.itrs, regs->itrs, sizeof(struct thash_data) * NITRS);
906 memcpy(vcpu->arch.dtrs, regs->dtrs, sizeof(struct thash_data) * NDTRS);
907 RESTORE_REGS(itr_regions);
908 RESTORE_REGS(dtr_regions);
909 RESTORE_REGS(tc_regions);
910 RESTORE_REGS(irq_check);
911 RESTORE_REGS(itc_check);
912 RESTORE_REGS(timer_check);
913 RESTORE_REGS(timer_pending);
914 RESTORE_REGS(last_itc);
915 for (i = 0; i < 8; i++) {
916 vcpu->arch.vrr[i] = regs->vrr[i];
917 vcpu->arch.ibr[i] = regs->ibr[i];
918 vcpu->arch.dbr[i] = regs->dbr[i];
920 for (i = 0; i < 4; i++)
921 vcpu->arch.insvc[i] = regs->insvc[i];
922 RESTORE_REGS(xtp);
923 RESTORE_REGS(metaphysical_rr0);
924 RESTORE_REGS(metaphysical_rr4);
925 RESTORE_REGS(metaphysical_saved_rr0);
926 RESTORE_REGS(metaphysical_saved_rr4);
927 RESTORE_REGS(fp_psr);
928 RESTORE_REGS(saved_gp);
930 vcpu->arch.irq_new_pending = 1;
931 vcpu->arch.itc_offset = regs->saved_itc - kvm_get_itc(vcpu);
932 set_bit(KVM_REQ_RESUME, &vcpu->requests);
934 vcpu_put(vcpu);
936 return 0;
939 long kvm_arch_vm_ioctl(struct file *filp,
940 unsigned int ioctl, unsigned long arg)
942 struct kvm *kvm = filp->private_data;
943 void __user *argp = (void __user *)arg;
944 int r = -ENOTTY;
946 switch (ioctl) {
947 case KVM_SET_MEMORY_REGION: {
948 struct kvm_memory_region kvm_mem;
949 struct kvm_userspace_memory_region kvm_userspace_mem;
951 r = -EFAULT;
952 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
953 goto out;
954 kvm_userspace_mem.slot = kvm_mem.slot;
955 kvm_userspace_mem.flags = kvm_mem.flags;
956 kvm_userspace_mem.guest_phys_addr =
957 kvm_mem.guest_phys_addr;
958 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
959 r = kvm_vm_ioctl_set_memory_region(kvm,
960 &kvm_userspace_mem, 0);
961 if (r)
962 goto out;
963 break;
965 case KVM_CREATE_IRQCHIP:
966 r = -EFAULT;
967 r = kvm_ioapic_init(kvm);
968 if (r)
969 goto out;
970 r = kvm_setup_default_irq_routing(kvm);
971 if (r) {
972 kvm_ioapic_destroy(kvm);
973 goto out;
975 break;
976 case KVM_IRQ_LINE_STATUS:
977 case KVM_IRQ_LINE: {
978 struct kvm_irq_level irq_event;
980 r = -EFAULT;
981 if (copy_from_user(&irq_event, argp, sizeof irq_event))
982 goto out;
983 r = -ENXIO;
984 if (irqchip_in_kernel(kvm)) {
985 __s32 status;
986 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
987 irq_event.irq, irq_event.level);
988 if (ioctl == KVM_IRQ_LINE_STATUS) {
989 r = -EFAULT;
990 irq_event.status = status;
991 if (copy_to_user(argp, &irq_event,
992 sizeof irq_event))
993 goto out;
995 r = 0;
997 break;
999 case KVM_GET_IRQCHIP: {
1000 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1001 struct kvm_irqchip chip;
1003 r = -EFAULT;
1004 if (copy_from_user(&chip, argp, sizeof chip))
1005 goto out;
1006 r = -ENXIO;
1007 if (!irqchip_in_kernel(kvm))
1008 goto out;
1009 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1010 if (r)
1011 goto out;
1012 r = -EFAULT;
1013 if (copy_to_user(argp, &chip, sizeof chip))
1014 goto out;
1015 r = 0;
1016 break;
1018 case KVM_SET_IRQCHIP: {
1019 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1020 struct kvm_irqchip chip;
1022 r = -EFAULT;
1023 if (copy_from_user(&chip, argp, sizeof chip))
1024 goto out;
1025 r = -ENXIO;
1026 if (!irqchip_in_kernel(kvm))
1027 goto out;
1028 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1029 if (r)
1030 goto out;
1031 r = 0;
1032 break;
1034 default:
1037 out:
1038 return r;
1041 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1042 struct kvm_sregs *sregs)
1044 return -EINVAL;
1047 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1048 struct kvm_sregs *sregs)
1050 return -EINVAL;
1053 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1054 struct kvm_translation *tr)
1057 return -EINVAL;
1060 static int kvm_alloc_vmm_area(void)
1062 if (!kvm_vmm_base && (kvm_vm_buffer_size < KVM_VM_BUFFER_SIZE)) {
1063 kvm_vmm_base = __get_free_pages(GFP_KERNEL,
1064 get_order(KVM_VMM_SIZE));
1065 if (!kvm_vmm_base)
1066 return -ENOMEM;
1068 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1069 kvm_vm_buffer = kvm_vmm_base + VMM_SIZE;
1071 printk(KERN_DEBUG"kvm:VMM's Base Addr:0x%lx, vm_buffer:0x%lx\n",
1072 kvm_vmm_base, kvm_vm_buffer);
1075 return 0;
1078 static void kvm_free_vmm_area(void)
1080 if (kvm_vmm_base) {
1081 /*Zero this area before free to avoid bits leak!!*/
1082 memset((void *)kvm_vmm_base, 0, KVM_VMM_SIZE);
1083 free_pages(kvm_vmm_base, get_order(KVM_VMM_SIZE));
1084 kvm_vmm_base = 0;
1085 kvm_vm_buffer = 0;
1086 kvm_vsa_base = 0;
1090 static int vti_init_vpd(struct kvm_vcpu *vcpu)
1092 int i;
1093 union cpuid3_t cpuid3;
1094 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1096 if (IS_ERR(vpd))
1097 return PTR_ERR(vpd);
1099 /* CPUID init */
1100 for (i = 0; i < 5; i++)
1101 vpd->vcpuid[i] = ia64_get_cpuid(i);
1103 /* Limit the CPUID number to 5 */
1104 cpuid3.value = vpd->vcpuid[3];
1105 cpuid3.number = 4; /* 5 - 1 */
1106 vpd->vcpuid[3] = cpuid3.value;
1108 /*Set vac and vdc fields*/
1109 vpd->vac.a_from_int_cr = 1;
1110 vpd->vac.a_to_int_cr = 1;
1111 vpd->vac.a_from_psr = 1;
1112 vpd->vac.a_from_cpuid = 1;
1113 vpd->vac.a_cover = 1;
1114 vpd->vac.a_bsw = 1;
1115 vpd->vac.a_int = 1;
1116 vpd->vdc.d_vmsw = 1;
1118 /*Set virtual buffer*/
1119 vpd->virt_env_vaddr = KVM_VM_BUFFER_BASE;
1121 return 0;
1124 static int vti_create_vp(struct kvm_vcpu *vcpu)
1126 long ret;
1127 struct vpd *vpd = vcpu->arch.vpd;
1128 unsigned long vmm_ivt;
1130 vmm_ivt = kvm_vmm_info->vmm_ivt;
1132 printk(KERN_DEBUG "kvm: vcpu:%p,ivt: 0x%lx\n", vcpu, vmm_ivt);
1134 ret = ia64_pal_vp_create((u64 *)vpd, (u64 *)vmm_ivt, 0);
1136 if (ret) {
1137 printk(KERN_ERR"kvm: ia64_pal_vp_create failed!\n");
1138 return -EINVAL;
1140 return 0;
1143 static void init_ptce_info(struct kvm_vcpu *vcpu)
1145 ia64_ptce_info_t ptce = {0};
1147 ia64_get_ptce(&ptce);
1148 vcpu->arch.ptce_base = ptce.base;
1149 vcpu->arch.ptce_count[0] = ptce.count[0];
1150 vcpu->arch.ptce_count[1] = ptce.count[1];
1151 vcpu->arch.ptce_stride[0] = ptce.stride[0];
1152 vcpu->arch.ptce_stride[1] = ptce.stride[1];
1155 static void kvm_migrate_hlt_timer(struct kvm_vcpu *vcpu)
1157 struct hrtimer *p_ht = &vcpu->arch.hlt_timer;
1159 if (hrtimer_cancel(p_ht))
1160 hrtimer_start_expires(p_ht, HRTIMER_MODE_ABS);
1163 static enum hrtimer_restart hlt_timer_fn(struct hrtimer *data)
1165 struct kvm_vcpu *vcpu;
1166 wait_queue_head_t *q;
1168 vcpu = container_of(data, struct kvm_vcpu, arch.hlt_timer);
1169 q = &vcpu->wq;
1171 if (vcpu->arch.mp_state != KVM_MP_STATE_HALTED)
1172 goto out;
1174 if (waitqueue_active(q))
1175 wake_up_interruptible(q);
1177 out:
1178 vcpu->arch.timer_fired = 1;
1179 vcpu->arch.timer_check = 1;
1180 return HRTIMER_NORESTART;
1183 #define PALE_RESET_ENTRY 0x80000000ffffffb0UL
1185 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1187 struct kvm_vcpu *v;
1188 int r;
1189 int i;
1190 long itc_offset;
1191 struct kvm *kvm = vcpu->kvm;
1192 struct kvm_pt_regs *regs = vcpu_regs(vcpu);
1194 union context *p_ctx = &vcpu->arch.guest;
1195 struct kvm_vcpu *vmm_vcpu = to_guest(vcpu->kvm, vcpu);
1197 /*Init vcpu context for first run.*/
1198 if (IS_ERR(vmm_vcpu))
1199 return PTR_ERR(vmm_vcpu);
1201 if (kvm_vcpu_is_bsp(vcpu)) {
1202 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1204 /*Set entry address for first run.*/
1205 regs->cr_iip = PALE_RESET_ENTRY;
1207 /*Initialize itc offset for vcpus*/
1208 itc_offset = 0UL - kvm_get_itc(vcpu);
1209 for (i = 0; i < KVM_MAX_VCPUS; i++) {
1210 v = (struct kvm_vcpu *)((char *)vcpu +
1211 sizeof(struct kvm_vcpu_data) * i);
1212 v->arch.itc_offset = itc_offset;
1213 v->arch.last_itc = 0;
1215 } else
1216 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
1218 r = -ENOMEM;
1219 vcpu->arch.apic = kzalloc(sizeof(struct kvm_lapic), GFP_KERNEL);
1220 if (!vcpu->arch.apic)
1221 goto out;
1222 vcpu->arch.apic->vcpu = vcpu;
1224 p_ctx->gr[1] = 0;
1225 p_ctx->gr[12] = (unsigned long)((char *)vmm_vcpu + KVM_STK_OFFSET);
1226 p_ctx->gr[13] = (unsigned long)vmm_vcpu;
1227 p_ctx->psr = 0x1008522000UL;
1228 p_ctx->ar[40] = FPSR_DEFAULT; /*fpsr*/
1229 p_ctx->caller_unat = 0;
1230 p_ctx->pr = 0x0;
1231 p_ctx->ar[36] = 0x0; /*unat*/
1232 p_ctx->ar[19] = 0x0; /*rnat*/
1233 p_ctx->ar[18] = (unsigned long)vmm_vcpu +
1234 ((sizeof(struct kvm_vcpu)+15) & ~15);
1235 p_ctx->ar[64] = 0x0; /*pfs*/
1236 p_ctx->cr[0] = 0x7e04UL;
1237 p_ctx->cr[2] = (unsigned long)kvm_vmm_info->vmm_ivt;
1238 p_ctx->cr[8] = 0x3c;
1240 /*Initilize region register*/
1241 p_ctx->rr[0] = 0x30;
1242 p_ctx->rr[1] = 0x30;
1243 p_ctx->rr[2] = 0x30;
1244 p_ctx->rr[3] = 0x30;
1245 p_ctx->rr[4] = 0x30;
1246 p_ctx->rr[5] = 0x30;
1247 p_ctx->rr[7] = 0x30;
1249 /*Initilize branch register 0*/
1250 p_ctx->br[0] = *(unsigned long *)kvm_vmm_info->vmm_entry;
1252 vcpu->arch.vmm_rr = kvm->arch.vmm_init_rr;
1253 vcpu->arch.metaphysical_rr0 = kvm->arch.metaphysical_rr0;
1254 vcpu->arch.metaphysical_rr4 = kvm->arch.metaphysical_rr4;
1256 hrtimer_init(&vcpu->arch.hlt_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1257 vcpu->arch.hlt_timer.function = hlt_timer_fn;
1259 vcpu->arch.last_run_cpu = -1;
1260 vcpu->arch.vpd = (struct vpd *)VPD_BASE(vcpu->vcpu_id);
1261 vcpu->arch.vsa_base = kvm_vsa_base;
1262 vcpu->arch.__gp = kvm_vmm_gp;
1263 vcpu->arch.dirty_log_lock_pa = __pa(&kvm->arch.dirty_log_lock);
1264 vcpu->arch.vhpt.hash = (struct thash_data *)VHPT_BASE(vcpu->vcpu_id);
1265 vcpu->arch.vtlb.hash = (struct thash_data *)VTLB_BASE(vcpu->vcpu_id);
1266 init_ptce_info(vcpu);
1268 r = 0;
1269 out:
1270 return r;
1273 static int vti_vcpu_setup(struct kvm_vcpu *vcpu, int id)
1275 unsigned long psr;
1276 int r;
1278 local_irq_save(psr);
1279 r = kvm_insert_vmm_mapping(vcpu);
1280 local_irq_restore(psr);
1281 if (r)
1282 goto fail;
1283 r = kvm_vcpu_init(vcpu, vcpu->kvm, id);
1284 if (r)
1285 goto fail;
1287 r = vti_init_vpd(vcpu);
1288 if (r) {
1289 printk(KERN_DEBUG"kvm: vpd init error!!\n");
1290 goto uninit;
1293 r = vti_create_vp(vcpu);
1294 if (r)
1295 goto uninit;
1297 kvm_purge_vmm_mapping(vcpu);
1299 return 0;
1300 uninit:
1301 kvm_vcpu_uninit(vcpu);
1302 fail:
1303 return r;
1306 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1307 unsigned int id)
1309 struct kvm_vcpu *vcpu;
1310 unsigned long vm_base = kvm->arch.vm_base;
1311 int r;
1312 int cpu;
1314 BUG_ON(sizeof(struct kvm_vcpu) > VCPU_STRUCT_SIZE/2);
1316 r = -EINVAL;
1317 if (id >= KVM_MAX_VCPUS) {
1318 printk(KERN_ERR"kvm: Can't configure vcpus > %ld",
1319 KVM_MAX_VCPUS);
1320 goto fail;
1323 r = -ENOMEM;
1324 if (!vm_base) {
1325 printk(KERN_ERR"kvm: Create vcpu[%d] error!\n", id);
1326 goto fail;
1328 vcpu = (struct kvm_vcpu *)(vm_base + offsetof(struct kvm_vm_data,
1329 vcpu_data[id].vcpu_struct));
1330 vcpu->kvm = kvm;
1332 cpu = get_cpu();
1333 r = vti_vcpu_setup(vcpu, id);
1334 put_cpu();
1336 if (r) {
1337 printk(KERN_DEBUG"kvm: vcpu_setup error!!\n");
1338 goto fail;
1341 return vcpu;
1342 fail:
1343 return ERR_PTR(r);
1346 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1348 return 0;
1351 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1353 return -EINVAL;
1356 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1358 return -EINVAL;
1361 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1362 struct kvm_guest_debug *dbg)
1364 return -EINVAL;
1367 static void free_kvm(struct kvm *kvm)
1369 unsigned long vm_base = kvm->arch.vm_base;
1371 if (vm_base) {
1372 memset((void *)vm_base, 0, KVM_VM_DATA_SIZE);
1373 free_pages(vm_base, get_order(KVM_VM_DATA_SIZE));
1378 static void kvm_release_vm_pages(struct kvm *kvm)
1380 struct kvm_memslots *slots;
1381 struct kvm_memory_slot *memslot;
1382 int i, j;
1383 unsigned long base_gfn;
1385 slots = kvm_memslots(kvm);
1386 for (i = 0; i < slots->nmemslots; i++) {
1387 memslot = &slots->memslots[i];
1388 base_gfn = memslot->base_gfn;
1390 for (j = 0; j < memslot->npages; j++) {
1391 if (memslot->rmap[j])
1392 put_page((struct page *)memslot->rmap[j]);
1397 void kvm_arch_sync_events(struct kvm *kvm)
1401 void kvm_arch_destroy_vm(struct kvm *kvm)
1403 kvm_iommu_unmap_guest(kvm);
1404 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1405 kvm_free_all_assigned_devices(kvm);
1406 #endif
1407 kfree(kvm->arch.vioapic);
1408 kvm_release_vm_pages(kvm);
1409 kvm_free_physmem(kvm);
1410 cleanup_srcu_struct(&kvm->srcu);
1411 free_kvm(kvm);
1414 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1418 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1420 if (cpu != vcpu->cpu) {
1421 vcpu->cpu = cpu;
1422 if (vcpu->arch.ht_active)
1423 kvm_migrate_hlt_timer(vcpu);
1427 #define SAVE_REGS(_x) regs->_x = vcpu->arch._x
1429 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1431 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1432 int i;
1434 vcpu_load(vcpu);
1436 for (i = 0; i < 16; i++) {
1437 regs->vpd.vgr[i] = vpd->vgr[i];
1438 regs->vpd.vbgr[i] = vpd->vbgr[i];
1440 for (i = 0; i < 128; i++)
1441 regs->vpd.vcr[i] = vpd->vcr[i];
1442 regs->vpd.vhpi = vpd->vhpi;
1443 regs->vpd.vnat = vpd->vnat;
1444 regs->vpd.vbnat = vpd->vbnat;
1445 regs->vpd.vpsr = vpd->vpsr;
1446 regs->vpd.vpr = vpd->vpr;
1448 memcpy(&regs->saved_guest, &vcpu->arch.guest, sizeof(union context));
1450 SAVE_REGS(mp_state);
1451 SAVE_REGS(vmm_rr);
1452 memcpy(regs->itrs, vcpu->arch.itrs, sizeof(struct thash_data) * NITRS);
1453 memcpy(regs->dtrs, vcpu->arch.dtrs, sizeof(struct thash_data) * NDTRS);
1454 SAVE_REGS(itr_regions);
1455 SAVE_REGS(dtr_regions);
1456 SAVE_REGS(tc_regions);
1457 SAVE_REGS(irq_check);
1458 SAVE_REGS(itc_check);
1459 SAVE_REGS(timer_check);
1460 SAVE_REGS(timer_pending);
1461 SAVE_REGS(last_itc);
1462 for (i = 0; i < 8; i++) {
1463 regs->vrr[i] = vcpu->arch.vrr[i];
1464 regs->ibr[i] = vcpu->arch.ibr[i];
1465 regs->dbr[i] = vcpu->arch.dbr[i];
1467 for (i = 0; i < 4; i++)
1468 regs->insvc[i] = vcpu->arch.insvc[i];
1469 regs->saved_itc = vcpu->arch.itc_offset + kvm_get_itc(vcpu);
1470 SAVE_REGS(xtp);
1471 SAVE_REGS(metaphysical_rr0);
1472 SAVE_REGS(metaphysical_rr4);
1473 SAVE_REGS(metaphysical_saved_rr0);
1474 SAVE_REGS(metaphysical_saved_rr4);
1475 SAVE_REGS(fp_psr);
1476 SAVE_REGS(saved_gp);
1478 vcpu_put(vcpu);
1479 return 0;
1482 int kvm_arch_vcpu_ioctl_get_stack(struct kvm_vcpu *vcpu,
1483 struct kvm_ia64_vcpu_stack *stack)
1485 memcpy(stack, vcpu, sizeof(struct kvm_ia64_vcpu_stack));
1486 return 0;
1489 int kvm_arch_vcpu_ioctl_set_stack(struct kvm_vcpu *vcpu,
1490 struct kvm_ia64_vcpu_stack *stack)
1492 memcpy(vcpu + 1, &stack->stack[0] + sizeof(struct kvm_vcpu),
1493 sizeof(struct kvm_ia64_vcpu_stack) - sizeof(struct kvm_vcpu));
1495 vcpu->arch.exit_data = ((struct kvm_vcpu *)stack)->arch.exit_data;
1496 return 0;
1499 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1502 hrtimer_cancel(&vcpu->arch.hlt_timer);
1503 kfree(vcpu->arch.apic);
1507 long kvm_arch_vcpu_ioctl(struct file *filp,
1508 unsigned int ioctl, unsigned long arg)
1510 struct kvm_vcpu *vcpu = filp->private_data;
1511 void __user *argp = (void __user *)arg;
1512 struct kvm_ia64_vcpu_stack *stack = NULL;
1513 long r;
1515 switch (ioctl) {
1516 case KVM_IA64_VCPU_GET_STACK: {
1517 struct kvm_ia64_vcpu_stack __user *user_stack;
1518 void __user *first_p = argp;
1520 r = -EFAULT;
1521 if (copy_from_user(&user_stack, first_p, sizeof(void *)))
1522 goto out;
1524 if (!access_ok(VERIFY_WRITE, user_stack,
1525 sizeof(struct kvm_ia64_vcpu_stack))) {
1526 printk(KERN_INFO "KVM_IA64_VCPU_GET_STACK: "
1527 "Illegal user destination address for stack\n");
1528 goto out;
1530 stack = kzalloc(sizeof(struct kvm_ia64_vcpu_stack), GFP_KERNEL);
1531 if (!stack) {
1532 r = -ENOMEM;
1533 goto out;
1536 r = kvm_arch_vcpu_ioctl_get_stack(vcpu, stack);
1537 if (r)
1538 goto out;
1540 if (copy_to_user(user_stack, stack,
1541 sizeof(struct kvm_ia64_vcpu_stack))) {
1542 r = -EFAULT;
1543 goto out;
1546 break;
1548 case KVM_IA64_VCPU_SET_STACK: {
1549 struct kvm_ia64_vcpu_stack __user *user_stack;
1550 void __user *first_p = argp;
1552 r = -EFAULT;
1553 if (copy_from_user(&user_stack, first_p, sizeof(void *)))
1554 goto out;
1556 if (!access_ok(VERIFY_READ, user_stack,
1557 sizeof(struct kvm_ia64_vcpu_stack))) {
1558 printk(KERN_INFO "KVM_IA64_VCPU_SET_STACK: "
1559 "Illegal user address for stack\n");
1560 goto out;
1562 stack = kmalloc(sizeof(struct kvm_ia64_vcpu_stack), GFP_KERNEL);
1563 if (!stack) {
1564 r = -ENOMEM;
1565 goto out;
1567 if (copy_from_user(stack, user_stack,
1568 sizeof(struct kvm_ia64_vcpu_stack)))
1569 goto out;
1571 r = kvm_arch_vcpu_ioctl_set_stack(vcpu, stack);
1572 break;
1575 default:
1576 r = -EINVAL;
1579 out:
1580 kfree(stack);
1581 return r;
1584 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1585 struct kvm_memory_slot *memslot,
1586 struct kvm_memory_slot old,
1587 struct kvm_userspace_memory_region *mem,
1588 int user_alloc)
1590 unsigned long i;
1591 unsigned long pfn;
1592 int npages = memslot->npages;
1593 unsigned long base_gfn = memslot->base_gfn;
1595 if (base_gfn + npages > (KVM_MAX_MEM_SIZE >> PAGE_SHIFT))
1596 return -ENOMEM;
1598 for (i = 0; i < npages; i++) {
1599 pfn = gfn_to_pfn(kvm, base_gfn + i);
1600 if (!kvm_is_mmio_pfn(pfn)) {
1601 kvm_set_pmt_entry(kvm, base_gfn + i,
1602 pfn << PAGE_SHIFT,
1603 _PAGE_AR_RWX | _PAGE_MA_WB);
1604 memslot->rmap[i] = (unsigned long)pfn_to_page(pfn);
1605 } else {
1606 kvm_set_pmt_entry(kvm, base_gfn + i,
1607 GPFN_PHYS_MMIO | (pfn << PAGE_SHIFT),
1608 _PAGE_MA_UC);
1609 memslot->rmap[i] = 0;
1613 return 0;
1616 void kvm_arch_commit_memory_region(struct kvm *kvm,
1617 struct kvm_userspace_memory_region *mem,
1618 struct kvm_memory_slot old,
1619 int user_alloc)
1621 return;
1624 void kvm_arch_flush_shadow(struct kvm *kvm)
1626 kvm_flush_remote_tlbs(kvm);
1629 long kvm_arch_dev_ioctl(struct file *filp,
1630 unsigned int ioctl, unsigned long arg)
1632 return -EINVAL;
1635 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1637 kvm_vcpu_uninit(vcpu);
1640 static int vti_cpu_has_kvm_support(void)
1642 long avail = 1, status = 1, control = 1;
1643 long ret;
1645 ret = ia64_pal_proc_get_features(&avail, &status, &control, 0);
1646 if (ret)
1647 goto out;
1649 if (!(avail & PAL_PROC_VM_BIT))
1650 goto out;
1652 printk(KERN_DEBUG"kvm: Hardware Supports VT\n");
1654 ret = ia64_pal_vp_env_info(&kvm_vm_buffer_size, &vp_env_info);
1655 if (ret)
1656 goto out;
1657 printk(KERN_DEBUG"kvm: VM Buffer Size:0x%lx\n", kvm_vm_buffer_size);
1659 if (!(vp_env_info & VP_OPCODE)) {
1660 printk(KERN_WARNING"kvm: No opcode ability on hardware, "
1661 "vm_env_info:0x%lx\n", vp_env_info);
1664 return 1;
1665 out:
1666 return 0;
1671 * On SN2, the ITC isn't stable, so copy in fast path code to use the
1672 * SN2 RTC, replacing the ITC based default verion.
1674 static void kvm_patch_vmm(struct kvm_vmm_info *vmm_info,
1675 struct module *module)
1677 unsigned long new_ar, new_ar_sn2;
1678 unsigned long module_base;
1680 if (!ia64_platform_is("sn2"))
1681 return;
1683 module_base = (unsigned long)module->module_core;
1685 new_ar = kvm_vmm_base + vmm_info->patch_mov_ar - module_base;
1686 new_ar_sn2 = kvm_vmm_base + vmm_info->patch_mov_ar_sn2 - module_base;
1688 printk(KERN_INFO "kvm: Patching ITC emulation to use SGI SN2 RTC "
1689 "as source\n");
1692 * Copy the SN2 version of mov_ar into place. They are both
1693 * the same size, so 6 bundles is sufficient (6 * 0x10).
1695 memcpy((void *)new_ar, (void *)new_ar_sn2, 0x60);
1698 static int kvm_relocate_vmm(struct kvm_vmm_info *vmm_info,
1699 struct module *module)
1701 unsigned long module_base;
1702 unsigned long vmm_size;
1704 unsigned long vmm_offset, func_offset, fdesc_offset;
1705 struct fdesc *p_fdesc;
1707 BUG_ON(!module);
1709 if (!kvm_vmm_base) {
1710 printk("kvm: kvm area hasn't been initilized yet!!\n");
1711 return -EFAULT;
1714 /*Calculate new position of relocated vmm module.*/
1715 module_base = (unsigned long)module->module_core;
1716 vmm_size = module->core_size;
1717 if (unlikely(vmm_size > KVM_VMM_SIZE))
1718 return -EFAULT;
1720 memcpy((void *)kvm_vmm_base, (void *)module_base, vmm_size);
1721 kvm_patch_vmm(vmm_info, module);
1722 kvm_flush_icache(kvm_vmm_base, vmm_size);
1724 /*Recalculate kvm_vmm_info based on new VMM*/
1725 vmm_offset = vmm_info->vmm_ivt - module_base;
1726 kvm_vmm_info->vmm_ivt = KVM_VMM_BASE + vmm_offset;
1727 printk(KERN_DEBUG"kvm: Relocated VMM's IVT Base Addr:%lx\n",
1728 kvm_vmm_info->vmm_ivt);
1730 fdesc_offset = (unsigned long)vmm_info->vmm_entry - module_base;
1731 kvm_vmm_info->vmm_entry = (kvm_vmm_entry *)(KVM_VMM_BASE +
1732 fdesc_offset);
1733 func_offset = *(unsigned long *)vmm_info->vmm_entry - module_base;
1734 p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1735 p_fdesc->ip = KVM_VMM_BASE + func_offset;
1736 p_fdesc->gp = KVM_VMM_BASE+(p_fdesc->gp - module_base);
1738 printk(KERN_DEBUG"kvm: Relocated VMM's Init Entry Addr:%lx\n",
1739 KVM_VMM_BASE+func_offset);
1741 fdesc_offset = (unsigned long)vmm_info->tramp_entry - module_base;
1742 kvm_vmm_info->tramp_entry = (kvm_tramp_entry *)(KVM_VMM_BASE +
1743 fdesc_offset);
1744 func_offset = *(unsigned long *)vmm_info->tramp_entry - module_base;
1745 p_fdesc = (struct fdesc *)(kvm_vmm_base + fdesc_offset);
1746 p_fdesc->ip = KVM_VMM_BASE + func_offset;
1747 p_fdesc->gp = KVM_VMM_BASE + (p_fdesc->gp - module_base);
1749 kvm_vmm_gp = p_fdesc->gp;
1751 printk(KERN_DEBUG"kvm: Relocated VMM's Entry IP:%p\n",
1752 kvm_vmm_info->vmm_entry);
1753 printk(KERN_DEBUG"kvm: Relocated VMM's Trampoline Entry IP:0x%lx\n",
1754 KVM_VMM_BASE + func_offset);
1756 return 0;
1759 int kvm_arch_init(void *opaque)
1761 int r;
1762 struct kvm_vmm_info *vmm_info = (struct kvm_vmm_info *)opaque;
1764 if (!vti_cpu_has_kvm_support()) {
1765 printk(KERN_ERR "kvm: No Hardware Virtualization Support!\n");
1766 r = -EOPNOTSUPP;
1767 goto out;
1770 if (kvm_vmm_info) {
1771 printk(KERN_ERR "kvm: Already loaded VMM module!\n");
1772 r = -EEXIST;
1773 goto out;
1776 r = -ENOMEM;
1777 kvm_vmm_info = kzalloc(sizeof(struct kvm_vmm_info), GFP_KERNEL);
1778 if (!kvm_vmm_info)
1779 goto out;
1781 if (kvm_alloc_vmm_area())
1782 goto out_free0;
1784 r = kvm_relocate_vmm(vmm_info, vmm_info->module);
1785 if (r)
1786 goto out_free1;
1788 return 0;
1790 out_free1:
1791 kvm_free_vmm_area();
1792 out_free0:
1793 kfree(kvm_vmm_info);
1794 out:
1795 return r;
1798 void kvm_arch_exit(void)
1800 kvm_free_vmm_area();
1801 kfree(kvm_vmm_info);
1802 kvm_vmm_info = NULL;
1805 static int kvm_ia64_sync_dirty_log(struct kvm *kvm,
1806 struct kvm_dirty_log *log)
1808 struct kvm_memory_slot *memslot;
1809 int r, i;
1810 long base;
1811 unsigned long n;
1812 unsigned long *dirty_bitmap = (unsigned long *)(kvm->arch.vm_base +
1813 offsetof(struct kvm_vm_data, kvm_mem_dirty_log));
1815 r = -EINVAL;
1816 if (log->slot >= KVM_MEMORY_SLOTS)
1817 goto out;
1819 memslot = &kvm->memslots->memslots[log->slot];
1820 r = -ENOENT;
1821 if (!memslot->dirty_bitmap)
1822 goto out;
1824 n = kvm_dirty_bitmap_bytes(memslot);
1825 base = memslot->base_gfn / BITS_PER_LONG;
1827 for (i = 0; i < n/sizeof(long); ++i) {
1828 memslot->dirty_bitmap[i] = dirty_bitmap[base + i];
1829 dirty_bitmap[base + i] = 0;
1831 r = 0;
1832 out:
1833 return r;
1836 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1837 struct kvm_dirty_log *log)
1839 int r;
1840 unsigned long n;
1841 struct kvm_memory_slot *memslot;
1842 int is_dirty = 0;
1844 mutex_lock(&kvm->slots_lock);
1845 spin_lock(&kvm->arch.dirty_log_lock);
1847 r = kvm_ia64_sync_dirty_log(kvm, log);
1848 if (r)
1849 goto out;
1851 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1852 if (r)
1853 goto out;
1855 /* If nothing is dirty, don't bother messing with page tables. */
1856 if (is_dirty) {
1857 kvm_flush_remote_tlbs(kvm);
1858 memslot = &kvm->memslots->memslots[log->slot];
1859 n = kvm_dirty_bitmap_bytes(memslot);
1860 memset(memslot->dirty_bitmap, 0, n);
1862 r = 0;
1863 out:
1864 mutex_unlock(&kvm->slots_lock);
1865 spin_unlock(&kvm->arch.dirty_log_lock);
1866 return r;
1869 int kvm_arch_hardware_setup(void)
1871 return 0;
1874 void kvm_arch_hardware_unsetup(void)
1878 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1880 int me;
1881 int cpu = vcpu->cpu;
1883 if (waitqueue_active(&vcpu->wq))
1884 wake_up_interruptible(&vcpu->wq);
1886 me = get_cpu();
1887 if (cpu != me && (unsigned) cpu < nr_cpu_ids && cpu_online(cpu))
1888 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
1889 smp_send_reschedule(cpu);
1890 put_cpu();
1893 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
1895 return __apic_accept_irq(vcpu, irq->vector);
1898 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
1900 return apic->vcpu->vcpu_id == dest;
1903 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
1905 return 0;
1908 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1910 return vcpu1->arch.xtp - vcpu2->arch.xtp;
1913 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
1914 int short_hand, int dest, int dest_mode)
1916 struct kvm_lapic *target = vcpu->arch.apic;
1917 return (dest_mode == 0) ?
1918 kvm_apic_match_physical_addr(target, dest) :
1919 kvm_apic_match_logical_addr(target, dest);
1922 static int find_highest_bits(int *dat)
1924 u32 bits, bitnum;
1925 int i;
1927 /* loop for all 256 bits */
1928 for (i = 7; i >= 0 ; i--) {
1929 bits = dat[i];
1930 if (bits) {
1931 bitnum = fls(bits);
1932 return i * 32 + bitnum - 1;
1936 return -1;
1939 int kvm_highest_pending_irq(struct kvm_vcpu *vcpu)
1941 struct vpd *vpd = to_host(vcpu->kvm, vcpu->arch.vpd);
1943 if (vpd->irr[0] & (1UL << NMI_VECTOR))
1944 return NMI_VECTOR;
1945 if (vpd->irr[0] & (1UL << ExtINT_VECTOR))
1946 return ExtINT_VECTOR;
1948 return find_highest_bits((int *)&vpd->irr[0]);
1951 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1953 return vcpu->arch.timer_fired;
1956 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1958 return gfn;
1961 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1963 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE) ||
1964 (kvm_highest_pending_irq(vcpu) != -1);
1967 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1968 struct kvm_mp_state *mp_state)
1970 vcpu_load(vcpu);
1971 mp_state->mp_state = vcpu->arch.mp_state;
1972 vcpu_put(vcpu);
1973 return 0;
1976 static int vcpu_reset(struct kvm_vcpu *vcpu)
1978 int r;
1979 long psr;
1980 local_irq_save(psr);
1981 r = kvm_insert_vmm_mapping(vcpu);
1982 local_irq_restore(psr);
1983 if (r)
1984 goto fail;
1986 vcpu->arch.launched = 0;
1987 kvm_arch_vcpu_uninit(vcpu);
1988 r = kvm_arch_vcpu_init(vcpu);
1989 if (r)
1990 goto fail;
1992 kvm_purge_vmm_mapping(vcpu);
1993 r = 0;
1994 fail:
1995 return r;
1998 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1999 struct kvm_mp_state *mp_state)
2001 int r = 0;
2003 vcpu_load(vcpu);
2004 vcpu->arch.mp_state = mp_state->mp_state;
2005 if (vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)
2006 r = vcpu_reset(vcpu);
2007 vcpu_put(vcpu);
2008 return r;