2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpumask.h>
30 #include <linux/spinlock.h>
31 #include <linux/page-flags.h>
34 #include <asm/cputable.h>
35 #include <asm/cacheflush.h>
36 #include <asm/tlbflush.h>
37 #include <asm/uaccess.h>
39 #include <asm/kvm_ppc.h>
40 #include <asm/kvm_book3s.h>
41 #include <asm/mmu_context.h>
42 #include <asm/lppaca.h>
43 #include <asm/processor.h>
44 #include <asm/cputhreads.h>
46 #include <linux/gfp.h>
47 #include <linux/sched.h>
48 #include <linux/vmalloc.h>
49 #include <linux/highmem.h>
52 * For now, limit memory to 64GB and require it to be large pages.
53 * This value is chosen because it makes the ram_pginfo array be
54 * 64kB in size, which is about as large as we want to be trying
55 * to allocate with kmalloc.
57 #define MAX_MEM_ORDER 36
59 #define LARGE_PAGE_ORDER 24 /* 16MB pages */
61 /* #define EXIT_DEBUG */
62 /* #define EXIT_DEBUG_SIMPLE */
63 /* #define EXIT_DEBUG_INT */
65 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
67 local_paca->kvm_hstate.kvm_vcpu = vcpu;
68 local_paca->kvm_hstate.kvm_vcore = vcpu->arch.vcore;
71 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
75 static void kvmppc_vcpu_blocked(struct kvm_vcpu *vcpu);
76 static void kvmppc_vcpu_unblocked(struct kvm_vcpu *vcpu);
78 void kvmppc_vcpu_block(struct kvm_vcpu *vcpu)
81 unsigned long dec_nsec;
84 if (now >= vcpu->arch.dec_expires && !kvmppc_core_pending_dec(vcpu))
85 kvmppc_core_queue_dec(vcpu);
86 if (vcpu->arch.pending_exceptions)
88 if (vcpu->arch.dec_expires != ~(u64)0) {
89 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC /
91 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
95 kvmppc_vcpu_blocked(vcpu);
98 vcpu->stat.halt_wakeup++;
100 if (vcpu->arch.dec_expires != ~(u64)0)
101 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
103 kvmppc_vcpu_unblocked(vcpu);
106 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
108 vcpu->arch.shregs.msr = msr;
111 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
113 vcpu->arch.pvr = pvr;
116 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
120 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
121 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
122 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
123 for (r = 0; r < 16; ++r)
124 pr_err("r%2d = %.16lx r%d = %.16lx\n",
125 r, kvmppc_get_gpr(vcpu, r),
126 r+16, kvmppc_get_gpr(vcpu, r+16));
127 pr_err("ctr = %.16lx lr = %.16lx\n",
128 vcpu->arch.ctr, vcpu->arch.lr);
129 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
130 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
131 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
132 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
133 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
134 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
135 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
136 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
137 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
138 pr_err("fault dar = %.16lx dsisr = %.8x\n",
139 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
140 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
141 for (r = 0; r < vcpu->arch.slb_max; ++r)
142 pr_err(" ESID = %.16llx VSID = %.16llx\n",
143 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
144 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
145 vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
146 vcpu->arch.last_inst);
149 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
152 struct kvm_vcpu *v, *ret = NULL;
154 mutex_lock(&kvm->lock);
155 kvm_for_each_vcpu(r, v, kvm) {
156 if (v->vcpu_id == id) {
161 mutex_unlock(&kvm->lock);
165 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
167 vpa->shared_proc = 1;
168 vpa->yield_count = 1;
171 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
173 unsigned long vcpuid, unsigned long vpa)
175 struct kvm *kvm = vcpu->kvm;
176 unsigned long pg_index, ra, len;
177 unsigned long pg_offset;
179 struct kvm_vcpu *tvcpu;
181 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
187 if (flags == 0 || flags == 4)
192 /* registering new area; convert logical addr to real */
193 pg_index = vpa >> kvm->arch.ram_porder;
194 pg_offset = vpa & (kvm->arch.ram_psize - 1);
195 if (pg_index >= kvm->arch.ram_npages)
197 if (kvm->arch.ram_pginfo[pg_index].pfn == 0)
199 ra = kvm->arch.ram_pginfo[pg_index].pfn << PAGE_SHIFT;
203 len = *(unsigned short *)(va + 4);
205 len = *(unsigned int *)(va + 4);
206 if (pg_offset + len > kvm->arch.ram_psize)
209 case 1: /* register VPA */
212 tvcpu->arch.vpa = va;
215 case 2: /* register DTL */
218 if (!tvcpu->arch.vpa)
221 tvcpu->arch.dtl = va;
222 tvcpu->arch.dtl_end = va + len;
224 case 3: /* register SLB shadow buffer */
227 if (!tvcpu->arch.vpa)
229 tvcpu->arch.slb_shadow = va;
230 len = (len - 16) / 16;
231 tvcpu->arch.slb_shadow = va;
236 case 5: /* unregister VPA */
237 if (tvcpu->arch.slb_shadow || tvcpu->arch.dtl)
239 tvcpu->arch.vpa = NULL;
241 case 6: /* unregister DTL */
242 tvcpu->arch.dtl = NULL;
244 case 7: /* unregister SLB shadow buffer */
245 tvcpu->arch.slb_shadow = NULL;
252 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
254 unsigned long req = kvmppc_get_gpr(vcpu, 3);
255 unsigned long target, ret = H_SUCCESS;
256 struct kvm_vcpu *tvcpu;
260 vcpu->arch.shregs.msr |= MSR_EE;
261 vcpu->arch.ceded = 1;
263 if (!vcpu->arch.prodded)
264 kvmppc_vcpu_block(vcpu);
266 vcpu->arch.prodded = 0;
268 vcpu->arch.ceded = 0;
271 target = kvmppc_get_gpr(vcpu, 4);
272 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
277 tvcpu->arch.prodded = 1;
279 if (vcpu->arch.ceded) {
280 if (waitqueue_active(&vcpu->wq)) {
281 wake_up_interruptible(&vcpu->wq);
282 vcpu->stat.halt_wakeup++;
289 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
290 kvmppc_get_gpr(vcpu, 5),
291 kvmppc_get_gpr(vcpu, 6));
296 kvmppc_set_gpr(vcpu, 3, ret);
297 vcpu->arch.hcall_needed = 0;
301 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
302 struct task_struct *tsk)
306 vcpu->stat.sum_exits++;
308 run->exit_reason = KVM_EXIT_UNKNOWN;
309 run->ready_for_interrupt_injection = 1;
310 switch (vcpu->arch.trap) {
311 /* We're good on these - the host merely wanted to get our attention */
312 case BOOK3S_INTERRUPT_HV_DECREMENTER:
313 vcpu->stat.dec_exits++;
316 case BOOK3S_INTERRUPT_EXTERNAL:
317 vcpu->stat.ext_intr_exits++;
320 case BOOK3S_INTERRUPT_PERFMON:
323 case BOOK3S_INTERRUPT_PROGRAM:
327 * Normally program interrupts are delivered directly
328 * to the guest by the hardware, but we can get here
329 * as a result of a hypervisor emulation interrupt
330 * (e40) getting turned into a 700 by BML RTAS.
332 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
333 kvmppc_core_queue_program(vcpu, flags);
337 case BOOK3S_INTERRUPT_SYSCALL:
339 /* hcall - punt to userspace */
342 if (vcpu->arch.shregs.msr & MSR_PR) {
343 /* sc 1 from userspace - reflect to guest syscall */
344 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
348 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
349 for (i = 0; i < 9; ++i)
350 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
351 run->exit_reason = KVM_EXIT_PAPR_HCALL;
352 vcpu->arch.hcall_needed = 1;
357 * We get these next two if the guest does a bad real-mode access,
358 * as we have enabled VRMA (virtualized real mode area) mode in the
359 * LPCR. We just generate an appropriate DSI/ISI to the guest.
361 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
362 vcpu->arch.shregs.dsisr = vcpu->arch.fault_dsisr;
363 vcpu->arch.shregs.dar = vcpu->arch.fault_dar;
364 kvmppc_inject_interrupt(vcpu, BOOK3S_INTERRUPT_DATA_STORAGE, 0);
367 case BOOK3S_INTERRUPT_H_INST_STORAGE:
368 kvmppc_inject_interrupt(vcpu, BOOK3S_INTERRUPT_INST_STORAGE,
373 * This occurs if the guest executes an illegal instruction.
374 * We just generate a program interrupt to the guest, since
375 * we don't emulate any guest instructions at this stage.
377 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
378 kvmppc_core_queue_program(vcpu, 0x80000);
382 kvmppc_dump_regs(vcpu);
383 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
384 vcpu->arch.trap, kvmppc_get_pc(vcpu),
385 vcpu->arch.shregs.msr);
392 if (!(r & RESUME_HOST)) {
393 /* To avoid clobbering exit_reason, only check for signals if
394 * we aren't already exiting to userspace for some other
396 if (signal_pending(tsk)) {
397 vcpu->stat.signal_exits++;
398 run->exit_reason = KVM_EXIT_INTR;
401 kvmppc_core_deliver_interrupts(vcpu);
408 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
409 struct kvm_sregs *sregs)
413 sregs->pvr = vcpu->arch.pvr;
415 memset(sregs, 0, sizeof(struct kvm_sregs));
416 for (i = 0; i < vcpu->arch.slb_max; i++) {
417 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
418 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
424 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
425 struct kvm_sregs *sregs)
429 kvmppc_set_pvr(vcpu, sregs->pvr);
432 for (i = 0; i < vcpu->arch.slb_nr; i++) {
433 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
434 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
435 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
439 vcpu->arch.slb_max = j;
444 int kvmppc_core_check_processor_compat(void)
446 if (cpu_has_feature(CPU_FTR_HVMODE))
451 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
453 struct kvm_vcpu *vcpu;
456 struct kvmppc_vcore *vcore;
458 core = id / threads_per_core;
459 if (core >= KVM_MAX_VCORES)
463 vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL);
467 err = kvm_vcpu_init(vcpu, kvm, id);
471 vcpu->arch.shared = &vcpu->arch.shregs;
472 vcpu->arch.last_cpu = -1;
473 vcpu->arch.mmcr[0] = MMCR0_FC;
474 vcpu->arch.ctrl = CTRL_RUNLATCH;
475 /* default to host PVR, since we can't spoof it */
476 vcpu->arch.pvr = mfspr(SPRN_PVR);
477 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
479 kvmppc_mmu_book3s_hv_init(vcpu);
482 * Some vcpus may start out in stopped state. If we initialize
483 * them to busy-in-host state they will stop other vcpus in the
484 * vcore from running. Instead we initialize them to blocked
485 * state, effectively considering them to be stopped until we
486 * see the first run ioctl for them.
488 vcpu->arch.state = KVMPPC_VCPU_BLOCKED;
490 init_waitqueue_head(&vcpu->arch.cpu_run);
492 mutex_lock(&kvm->lock);
493 vcore = kvm->arch.vcores[core];
495 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
497 INIT_LIST_HEAD(&vcore->runnable_threads);
498 spin_lock_init(&vcore->lock);
500 kvm->arch.vcores[core] = vcore;
502 mutex_unlock(&kvm->lock);
507 spin_lock(&vcore->lock);
508 ++vcore->num_threads;
510 spin_unlock(&vcore->lock);
511 vcpu->arch.vcore = vcore;
521 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
523 kvm_vcpu_uninit(vcpu);
527 static void kvmppc_vcpu_blocked(struct kvm_vcpu *vcpu)
529 struct kvmppc_vcore *vc = vcpu->arch.vcore;
531 spin_lock(&vc->lock);
532 vcpu->arch.state = KVMPPC_VCPU_BLOCKED;
534 if (vc->n_runnable > 0 &&
535 vc->n_runnable + vc->n_blocked == vc->num_threads) {
536 vcpu = list_first_entry(&vc->runnable_threads, struct kvm_vcpu,
538 wake_up(&vcpu->arch.cpu_run);
540 spin_unlock(&vc->lock);
543 static void kvmppc_vcpu_unblocked(struct kvm_vcpu *vcpu)
545 struct kvmppc_vcore *vc = vcpu->arch.vcore;
547 spin_lock(&vc->lock);
548 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
550 spin_unlock(&vc->lock);
553 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
554 extern void xics_wake_cpu(int cpu);
556 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
557 struct kvm_vcpu *vcpu)
561 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
563 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
565 /* decrement the physical thread id of each following vcpu */
567 list_for_each_entry_continue(v, &vc->runnable_threads, arch.run_list)
569 list_del(&vcpu->arch.run_list);
572 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
575 struct paca_struct *tpaca;
576 struct kvmppc_vcore *vc = vcpu->arch.vcore;
578 cpu = vc->pcpu + vcpu->arch.ptid;
580 tpaca->kvm_hstate.kvm_vcpu = vcpu;
581 tpaca->kvm_hstate.kvm_vcore = vc;
583 #ifdef CONFIG_PPC_ICP_NATIVE
584 if (vcpu->arch.ptid) {
585 tpaca->cpu_start = 0x80;
586 tpaca->kvm_hstate.in_guest = KVM_GUEST_MODE_GUEST;
594 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
600 while (vc->nap_count < vc->n_woken) {
601 if (++i >= 1000000) {
602 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
603 vc->nap_count, vc->n_woken);
612 * Check that we are on thread 0 and that any other threads in
613 * this core are off-line.
615 static int on_primary_thread(void)
617 int cpu = smp_processor_id();
618 int thr = cpu_thread_in_core(cpu);
622 while (++thr < threads_per_core)
623 if (cpu_online(cpu + thr))
629 * Run a set of guest threads on a physical core.
630 * Called with vc->lock held.
632 static int kvmppc_run_core(struct kvmppc_vcore *vc)
634 struct kvm_vcpu *vcpu, *vnext;
638 /* don't start if any threads have a signal pending */
639 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
640 if (signal_pending(vcpu->arch.run_task))
644 * Make sure we are running on thread 0, and that
645 * secondary threads are offline.
646 * XXX we should also block attempts to bring any
647 * secondary threads online.
649 if (threads_per_core > 1 && !on_primary_thread()) {
650 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
651 vcpu->arch.ret = -EBUSY;
657 vc->entry_exit_count = 0;
658 vc->vcore_running = 1;
660 vc->pcpu = smp_processor_id();
661 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
662 kvmppc_start_thread(vcpu);
663 vcpu = list_first_entry(&vc->runnable_threads, struct kvm_vcpu,
666 spin_unlock(&vc->lock);
670 __kvmppc_vcore_entry(NULL, vcpu);
672 /* wait for secondary threads to finish writing their state to memory */
673 spin_lock(&vc->lock);
674 if (vc->nap_count < vc->n_woken)
675 kvmppc_wait_for_nap(vc);
676 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
677 vc->vcore_running = 2;
678 spin_unlock(&vc->lock);
680 /* make sure updates to secondary vcpu structs are visible now */
688 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
689 /* cancel pending dec exception if dec is positive */
690 if (now < vcpu->arch.dec_expires &&
691 kvmppc_core_pending_dec(vcpu))
692 kvmppc_core_dequeue_dec(vcpu);
693 if (!vcpu->arch.trap) {
694 if (signal_pending(vcpu->arch.run_task)) {
695 vcpu->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
696 vcpu->arch.ret = -EINTR;
698 continue; /* didn't get to run */
700 ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
701 vcpu->arch.run_task);
702 vcpu->arch.ret = ret;
706 spin_lock(&vc->lock);
708 vc->vcore_running = 0;
709 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
711 if (vcpu->arch.ret != RESUME_GUEST) {
712 kvmppc_remove_runnable(vc, vcpu);
713 wake_up(&vcpu->arch.cpu_run);
720 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
724 struct kvmppc_vcore *vc;
727 /* No need to go into the guest when all we do is going out */
728 if (signal_pending(current)) {
729 kvm_run->exit_reason = KVM_EXIT_INTR;
733 /* On PPC970, check that we have an RMA region */
734 if (!vcpu->kvm->arch.rma && cpu_has_feature(CPU_FTR_ARCH_201))
737 kvm_run->exit_reason = 0;
738 vcpu->arch.ret = RESUME_GUEST;
741 flush_fp_to_thread(current);
742 flush_altivec_to_thread(current);
743 flush_vsx_to_thread(current);
746 * Synchronize with other threads in this virtual core
748 vc = vcpu->arch.vcore;
749 spin_lock(&vc->lock);
750 /* This happens the first time this is called for a vcpu */
751 if (vcpu->arch.state == KVMPPC_VCPU_BLOCKED)
753 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
754 ptid = vc->n_runnable;
755 vcpu->arch.run_task = current;
756 vcpu->arch.kvm_run = kvm_run;
757 vcpu->arch.ptid = ptid;
758 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
761 wait_state = TASK_INTERRUPTIBLE;
762 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
763 if (signal_pending(current)) {
764 if (!vc->vcore_running) {
765 kvm_run->exit_reason = KVM_EXIT_INTR;
766 vcpu->arch.ret = -EINTR;
769 /* have to wait for vcore to stop executing guest */
770 wait_state = TASK_UNINTERRUPTIBLE;
771 smp_send_reschedule(vc->pcpu);
774 if (!vc->vcore_running &&
775 vc->n_runnable + vc->n_blocked == vc->num_threads) {
777 if (kvmppc_run_core(vc))
781 if (vc->vcore_running == 1 && VCORE_EXIT_COUNT(vc) == 0)
782 kvmppc_start_thread(vcpu);
784 /* wait for other threads to come in, or wait for vcore */
785 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
786 spin_unlock(&vc->lock);
788 finish_wait(&vcpu->arch.cpu_run, &wait);
789 spin_lock(&vc->lock);
792 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
793 kvmppc_remove_runnable(vc, vcpu);
794 spin_unlock(&vc->lock);
796 return vcpu->arch.ret;
799 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
804 r = kvmppc_run_vcpu(run, vcpu);
806 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
807 !(vcpu->arch.shregs.msr & MSR_PR)) {
808 r = kvmppc_pseries_do_hcall(vcpu);
809 kvmppc_core_deliver_interrupts(vcpu);
811 } while (r == RESUME_GUEST);
815 static long kvmppc_stt_npages(unsigned long window_size)
817 return ALIGN((window_size >> SPAPR_TCE_SHIFT)
818 * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
821 static void release_spapr_tce_table(struct kvmppc_spapr_tce_table *stt)
823 struct kvm *kvm = stt->kvm;
826 mutex_lock(&kvm->lock);
827 list_del(&stt->list);
828 for (i = 0; i < kvmppc_stt_npages(stt->window_size); i++)
829 __free_page(stt->pages[i]);
831 mutex_unlock(&kvm->lock);
836 static int kvm_spapr_tce_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
838 struct kvmppc_spapr_tce_table *stt = vma->vm_file->private_data;
841 if (vmf->pgoff >= kvmppc_stt_npages(stt->window_size))
842 return VM_FAULT_SIGBUS;
844 page = stt->pages[vmf->pgoff];
850 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
851 .fault = kvm_spapr_tce_fault,
854 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
856 vma->vm_ops = &kvm_spapr_tce_vm_ops;
860 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
862 struct kvmppc_spapr_tce_table *stt = filp->private_data;
864 release_spapr_tce_table(stt);
868 static struct file_operations kvm_spapr_tce_fops = {
869 .mmap = kvm_spapr_tce_mmap,
870 .release = kvm_spapr_tce_release,
873 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
874 struct kvm_create_spapr_tce *args)
876 struct kvmppc_spapr_tce_table *stt = NULL;
881 /* Check this LIOBN hasn't been previously allocated */
882 list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) {
883 if (stt->liobn == args->liobn)
887 npages = kvmppc_stt_npages(args->window_size);
889 stt = kzalloc(sizeof(*stt) + npages* sizeof(struct page *),
894 stt->liobn = args->liobn;
895 stt->window_size = args->window_size;
898 for (i = 0; i < npages; i++) {
899 stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
906 mutex_lock(&kvm->lock);
907 list_add(&stt->list, &kvm->arch.spapr_tce_tables);
909 mutex_unlock(&kvm->lock);
911 return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
916 for (i = 0; i < npages; i++)
918 __free_page(stt->pages[i]);
925 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
926 Assumes POWER7 or PPC970. */
927 static inline int lpcr_rmls(unsigned long rma_size)
930 case 32ul << 20: /* 32 MB */
931 if (cpu_has_feature(CPU_FTR_ARCH_206))
932 return 8; /* only supported on POWER7 */
934 case 64ul << 20: /* 64 MB */
936 case 128ul << 20: /* 128 MB */
938 case 256ul << 20: /* 256 MB */
940 case 1ul << 30: /* 1 GB */
942 case 16ul << 30: /* 16 GB */
944 case 256ul << 30: /* 256 GB */
951 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
953 struct kvmppc_rma_info *ri = vma->vm_file->private_data;
956 if (vmf->pgoff >= ri->npages)
957 return VM_FAULT_SIGBUS;
959 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
965 static const struct vm_operations_struct kvm_rma_vm_ops = {
966 .fault = kvm_rma_fault,
969 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
971 vma->vm_flags |= VM_RESERVED;
972 vma->vm_ops = &kvm_rma_vm_ops;
976 static int kvm_rma_release(struct inode *inode, struct file *filp)
978 struct kvmppc_rma_info *ri = filp->private_data;
984 static struct file_operations kvm_rma_fops = {
985 .mmap = kvm_rma_mmap,
986 .release = kvm_rma_release,
989 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
991 struct kvmppc_rma_info *ri;
994 ri = kvm_alloc_rma();
998 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1000 kvm_release_rma(ri);
1002 ret->rma_size = ri->npages << PAGE_SHIFT;
1006 static struct page *hva_to_page(unsigned long addr)
1008 struct page *page[1];
1013 npages = get_user_pages_fast(addr, 1, 1, page);
1015 if (unlikely(npages != 1))
1021 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1022 struct kvm_userspace_memory_region *mem)
1024 unsigned long psize, porder;
1025 unsigned long i, npages, totalpages;
1026 unsigned long pg_ix;
1027 struct kvmppc_pginfo *pginfo;
1029 struct kvmppc_rma_info *ri = NULL;
1032 /* For now, only allow 16MB pages */
1033 porder = LARGE_PAGE_ORDER;
1034 psize = 1ul << porder;
1035 if ((mem->memory_size & (psize - 1)) ||
1036 (mem->guest_phys_addr & (psize - 1))) {
1037 pr_err("bad memory_size=%llx @ %llx\n",
1038 mem->memory_size, mem->guest_phys_addr);
1042 npages = mem->memory_size >> porder;
1043 totalpages = (mem->guest_phys_addr + mem->memory_size) >> porder;
1045 /* More memory than we have space to track? */
1046 if (totalpages > (1ul << (MAX_MEM_ORDER - LARGE_PAGE_ORDER)))
1049 /* Do we already have an RMA registered? */
1050 if (mem->guest_phys_addr == 0 && kvm->arch.rma)
1053 if (totalpages > kvm->arch.ram_npages)
1054 kvm->arch.ram_npages = totalpages;
1056 /* Is this one of our preallocated RMAs? */
1057 if (mem->guest_phys_addr == 0) {
1058 struct vm_area_struct *vma;
1060 down_read(¤t->mm->mmap_sem);
1061 vma = find_vma(current->mm, mem->userspace_addr);
1062 if (vma && vma->vm_file &&
1063 vma->vm_file->f_op == &kvm_rma_fops &&
1064 mem->userspace_addr == vma->vm_start)
1065 ri = vma->vm_file->private_data;
1066 up_read(¤t->mm->mmap_sem);
1067 if (!ri && cpu_has_feature(CPU_FTR_ARCH_201)) {
1068 pr_err("CPU requires an RMO\n");
1074 unsigned long rma_size;
1078 rma_size = ri->npages << PAGE_SHIFT;
1079 if (rma_size > mem->memory_size)
1080 rma_size = mem->memory_size;
1081 rmls = lpcr_rmls(rma_size);
1083 pr_err("Can't use RMA of 0x%lx bytes\n", rma_size);
1086 atomic_inc(&ri->use_count);
1088 kvm->arch.n_rma_pages = rma_size >> porder;
1090 /* Update LPCR and RMOR */
1091 lpcr = kvm->arch.lpcr;
1092 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1093 /* PPC970; insert RMLS value (split field) in HID4 */
1094 lpcr &= ~((1ul << HID4_RMLS0_SH) |
1095 (3ul << HID4_RMLS2_SH));
1096 lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1097 ((rmls & 3) << HID4_RMLS2_SH);
1098 /* RMOR is also in HID4 */
1099 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1103 lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1104 lpcr |= rmls << LPCR_RMLS_SH;
1105 kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1107 kvm->arch.lpcr = lpcr;
1108 pr_info("Using RMO at %lx size %lx (LPCR = %lx)\n",
1109 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1112 pg_ix = mem->guest_phys_addr >> porder;
1113 pginfo = kvm->arch.ram_pginfo + pg_ix;
1114 for (i = 0; i < npages; ++i, ++pg_ix) {
1115 if (ri && pg_ix < kvm->arch.n_rma_pages) {
1116 pginfo[i].pfn = ri->base_pfn +
1117 (pg_ix << (porder - PAGE_SHIFT));
1120 hva = mem->userspace_addr + (i << porder);
1121 page = hva_to_page(hva);
1123 pr_err("oops, no pfn for hva %lx\n", hva);
1126 /* Check it's a 16MB page */
1127 if (!PageHead(page) ||
1128 compound_order(page) != (LARGE_PAGE_ORDER - PAGE_SHIFT)) {
1129 pr_err("page at %lx isn't 16MB (o=%d)\n",
1130 hva, compound_order(page));
1133 pginfo[i].pfn = page_to_pfn(page);
1142 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1143 struct kvm_userspace_memory_region *mem)
1145 if (mem->guest_phys_addr == 0 && mem->memory_size != 0 &&
1147 kvmppc_map_vrma(kvm, mem);
1150 int kvmppc_core_init_vm(struct kvm *kvm)
1153 unsigned long npages = 1ul << (MAX_MEM_ORDER - LARGE_PAGE_ORDER);
1157 /* Allocate hashed page table */
1158 r = kvmppc_alloc_hpt(kvm);
1162 INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1164 kvm->arch.ram_pginfo = kzalloc(npages * sizeof(struct kvmppc_pginfo),
1166 if (!kvm->arch.ram_pginfo) {
1167 pr_err("kvmppc_core_init_vm: couldn't alloc %lu bytes\n",
1168 npages * sizeof(struct kvmppc_pginfo));
1172 kvm->arch.ram_npages = 0;
1173 kvm->arch.ram_psize = 1ul << LARGE_PAGE_ORDER;
1174 kvm->arch.ram_porder = LARGE_PAGE_ORDER;
1175 kvm->arch.rma = NULL;
1176 kvm->arch.n_rma_pages = 0;
1178 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1180 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1181 /* PPC970; HID4 is effectively the LPCR */
1182 unsigned long lpid = kvm->arch.lpid;
1183 kvm->arch.host_lpid = 0;
1184 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1185 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1186 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1187 ((lpid & 0xf) << HID4_LPID5_SH);
1189 /* POWER7; init LPCR for virtual RMA mode */
1190 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1191 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1192 lpcr &= LPCR_PECE | LPCR_LPES;
1193 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1194 LPCR_VPM0 | LPCR_VRMA_L;
1196 kvm->arch.lpcr = lpcr;
1201 kvmppc_free_hpt(kvm);
1205 void kvmppc_core_destroy_vm(struct kvm *kvm)
1207 struct kvmppc_pginfo *pginfo;
1210 if (kvm->arch.ram_pginfo) {
1211 pginfo = kvm->arch.ram_pginfo;
1212 kvm->arch.ram_pginfo = NULL;
1213 for (i = kvm->arch.n_rma_pages; i < kvm->arch.ram_npages; ++i)
1215 put_page(pfn_to_page(pginfo[i].pfn));
1218 if (kvm->arch.rma) {
1219 kvm_release_rma(kvm->arch.rma);
1220 kvm->arch.rma = NULL;
1223 kvmppc_free_hpt(kvm);
1224 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1227 /* These are stubs for now */
1228 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1232 /* We don't need to emulate any privileged instructions or dcbz */
1233 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1234 unsigned int inst, int *advance)
1236 return EMULATE_FAIL;
1239 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, int rs)
1241 return EMULATE_FAIL;
1244 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, int rt)
1246 return EMULATE_FAIL;
1249 static int kvmppc_book3s_hv_init(void)
1253 r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1258 r = kvmppc_mmu_hv_init();
1263 static void kvmppc_book3s_hv_exit(void)
1268 module_init(kvmppc_book3s_hv_init);
1269 module_exit(kvmppc_book3s_hv_exit);