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>
27 #include <linux/export.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
35 #include <asm/cputable.h>
36 #include <asm/cacheflush.h>
37 #include <asm/tlbflush.h>
38 #include <asm/uaccess.h>
40 #include <asm/kvm_ppc.h>
41 #include <asm/kvm_book3s.h>
42 #include <asm/mmu_context.h>
43 #include <asm/lppaca.h>
44 #include <asm/processor.h>
45 #include <asm/cputhreads.h>
47 #include <asm/hvcall.h>
48 #include <asm/switch_to.h>
49 #include <linux/gfp.h>
50 #include <linux/vmalloc.h>
51 #include <linux/highmem.h>
52 #include <linux/hugetlb.h>
54 /* #define EXIT_DEBUG */
55 /* #define EXIT_DEBUG_SIMPLE */
56 /* #define EXIT_DEBUG_INT */
58 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
59 static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu);
61 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
63 local_paca->kvm_hstate.kvm_vcpu = vcpu;
64 local_paca->kvm_hstate.kvm_vcore = vcpu->arch.vcore;
67 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
71 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
73 vcpu->arch.shregs.msr = msr;
74 kvmppc_end_cede(vcpu);
77 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
82 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
86 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
87 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
88 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
89 for (r = 0; r < 16; ++r)
90 pr_err("r%2d = %.16lx r%d = %.16lx\n",
91 r, kvmppc_get_gpr(vcpu, r),
92 r+16, kvmppc_get_gpr(vcpu, r+16));
93 pr_err("ctr = %.16lx lr = %.16lx\n",
94 vcpu->arch.ctr, vcpu->arch.lr);
95 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
96 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
97 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
98 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
99 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
100 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
101 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
102 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
103 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
104 pr_err("fault dar = %.16lx dsisr = %.8x\n",
105 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
106 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
107 for (r = 0; r < vcpu->arch.slb_max; ++r)
108 pr_err(" ESID = %.16llx VSID = %.16llx\n",
109 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
110 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
111 vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
112 vcpu->arch.last_inst);
115 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
118 struct kvm_vcpu *v, *ret = NULL;
120 mutex_lock(&kvm->lock);
121 kvm_for_each_vcpu(r, v, kvm) {
122 if (v->vcpu_id == id) {
127 mutex_unlock(&kvm->lock);
131 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
133 vpa->shared_proc = 1;
134 vpa->yield_count = 1;
137 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
146 static int vpa_is_registered(struct kvmppc_vpa *vpap)
148 if (vpap->update_pending)
149 return vpap->next_gpa != 0;
150 return vpap->pinned_addr != NULL;
153 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
155 unsigned long vcpuid, unsigned long vpa)
157 struct kvm *kvm = vcpu->kvm;
158 unsigned long len, nb;
160 struct kvm_vcpu *tvcpu;
163 struct kvmppc_vpa *vpap;
165 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
169 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
170 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
171 subfunc == H_VPA_REG_SLB) {
172 /* Registering new area - address must be cache-line aligned */
173 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
176 /* convert logical addr to kernel addr and read length */
177 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
180 if (subfunc == H_VPA_REG_VPA)
181 len = ((struct reg_vpa *)va)->length.hword;
183 len = ((struct reg_vpa *)va)->length.word;
184 kvmppc_unpin_guest_page(kvm, va);
187 if (len > nb || len < sizeof(struct reg_vpa))
196 spin_lock(&tvcpu->arch.vpa_update_lock);
199 case H_VPA_REG_VPA: /* register VPA */
200 if (len < sizeof(struct lppaca))
202 vpap = &tvcpu->arch.vpa;
206 case H_VPA_REG_DTL: /* register DTL */
207 if (len < sizeof(struct dtl_entry))
209 len -= len % sizeof(struct dtl_entry);
211 /* Check that they have previously registered a VPA */
213 if (!vpa_is_registered(&tvcpu->arch.vpa))
216 vpap = &tvcpu->arch.dtl;
220 case H_VPA_REG_SLB: /* register SLB shadow buffer */
221 /* Check that they have previously registered a VPA */
223 if (!vpa_is_registered(&tvcpu->arch.vpa))
226 vpap = &tvcpu->arch.slb_shadow;
230 case H_VPA_DEREG_VPA: /* deregister VPA */
231 /* Check they don't still have a DTL or SLB buf registered */
233 if (vpa_is_registered(&tvcpu->arch.dtl) ||
234 vpa_is_registered(&tvcpu->arch.slb_shadow))
237 vpap = &tvcpu->arch.vpa;
241 case H_VPA_DEREG_DTL: /* deregister DTL */
242 vpap = &tvcpu->arch.dtl;
246 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
247 vpap = &tvcpu->arch.slb_shadow;
253 vpap->next_gpa = vpa;
255 vpap->update_pending = 1;
258 spin_unlock(&tvcpu->arch.vpa_update_lock);
263 static void kvmppc_update_vpa(struct kvm *kvm, struct kvmppc_vpa *vpap)
268 vpap->update_pending = 0;
270 if (vpap->next_gpa) {
271 va = kvmppc_pin_guest_page(kvm, vpap->next_gpa, &nb);
272 if (nb < vpap->len) {
274 * If it's now too short, it must be that userspace
275 * has changed the mappings underlying guest memory,
276 * so unregister the region.
278 kvmppc_unpin_guest_page(kvm, va);
282 if (vpap->pinned_addr)
283 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr);
284 vpap->pinned_addr = va;
286 vpap->pinned_end = va + vpap->len;
289 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
291 struct kvm *kvm = vcpu->kvm;
293 spin_lock(&vcpu->arch.vpa_update_lock);
294 if (vcpu->arch.vpa.update_pending) {
295 kvmppc_update_vpa(kvm, &vcpu->arch.vpa);
296 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
298 if (vcpu->arch.dtl.update_pending) {
299 kvmppc_update_vpa(kvm, &vcpu->arch.dtl);
300 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
301 vcpu->arch.dtl_index = 0;
303 if (vcpu->arch.slb_shadow.update_pending)
304 kvmppc_update_vpa(kvm, &vcpu->arch.slb_shadow);
305 spin_unlock(&vcpu->arch.vpa_update_lock);
308 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
310 unsigned long req = kvmppc_get_gpr(vcpu, 3);
311 unsigned long target, ret = H_SUCCESS;
312 struct kvm_vcpu *tvcpu;
316 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
317 kvmppc_get_gpr(vcpu, 5),
318 kvmppc_get_gpr(vcpu, 6),
319 kvmppc_get_gpr(vcpu, 7));
324 target = kvmppc_get_gpr(vcpu, 4);
325 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
330 tvcpu->arch.prodded = 1;
332 if (vcpu->arch.ceded) {
333 if (waitqueue_active(&vcpu->wq)) {
334 wake_up_interruptible(&vcpu->wq);
335 vcpu->stat.halt_wakeup++;
342 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
343 kvmppc_get_gpr(vcpu, 5),
344 kvmppc_get_gpr(vcpu, 6));
349 kvmppc_set_gpr(vcpu, 3, ret);
350 vcpu->arch.hcall_needed = 0;
354 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
355 struct task_struct *tsk)
359 vcpu->stat.sum_exits++;
361 run->exit_reason = KVM_EXIT_UNKNOWN;
362 run->ready_for_interrupt_injection = 1;
363 switch (vcpu->arch.trap) {
364 /* We're good on these - the host merely wanted to get our attention */
365 case BOOK3S_INTERRUPT_HV_DECREMENTER:
366 vcpu->stat.dec_exits++;
369 case BOOK3S_INTERRUPT_EXTERNAL:
370 vcpu->stat.ext_intr_exits++;
373 case BOOK3S_INTERRUPT_PERFMON:
376 case BOOK3S_INTERRUPT_PROGRAM:
380 * Normally program interrupts are delivered directly
381 * to the guest by the hardware, but we can get here
382 * as a result of a hypervisor emulation interrupt
383 * (e40) getting turned into a 700 by BML RTAS.
385 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
386 kvmppc_core_queue_program(vcpu, flags);
390 case BOOK3S_INTERRUPT_SYSCALL:
392 /* hcall - punt to userspace */
395 if (vcpu->arch.shregs.msr & MSR_PR) {
396 /* sc 1 from userspace - reflect to guest syscall */
397 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
401 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
402 for (i = 0; i < 9; ++i)
403 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
404 run->exit_reason = KVM_EXIT_PAPR_HCALL;
405 vcpu->arch.hcall_needed = 1;
410 * We get these next two if the guest accesses a page which it thinks
411 * it has mapped but which is not actually present, either because
412 * it is for an emulated I/O device or because the corresonding
413 * host page has been paged out. Any other HDSI/HISI interrupts
414 * have been handled already.
416 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
417 r = kvmppc_book3s_hv_page_fault(run, vcpu,
418 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
420 case BOOK3S_INTERRUPT_H_INST_STORAGE:
421 r = kvmppc_book3s_hv_page_fault(run, vcpu,
422 kvmppc_get_pc(vcpu), 0);
425 * This occurs if the guest executes an illegal instruction.
426 * We just generate a program interrupt to the guest, since
427 * we don't emulate any guest instructions at this stage.
429 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
430 kvmppc_core_queue_program(vcpu, 0x80000);
434 kvmppc_dump_regs(vcpu);
435 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
436 vcpu->arch.trap, kvmppc_get_pc(vcpu),
437 vcpu->arch.shregs.msr);
446 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
447 struct kvm_sregs *sregs)
451 sregs->pvr = vcpu->arch.pvr;
453 memset(sregs, 0, sizeof(struct kvm_sregs));
454 for (i = 0; i < vcpu->arch.slb_max; i++) {
455 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
456 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
462 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
463 struct kvm_sregs *sregs)
467 kvmppc_set_pvr(vcpu, sregs->pvr);
470 for (i = 0; i < vcpu->arch.slb_nr; i++) {
471 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
472 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
473 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
477 vcpu->arch.slb_max = j;
482 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
487 case KVM_REG_PPC_HIOR:
488 r = put_user(0, (u64 __user *)reg->addr);
497 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
502 case KVM_REG_PPC_HIOR:
505 /* Only allow this to be set to zero */
506 r = get_user(hior, (u64 __user *)reg->addr);
507 if (!r && (hior != 0))
518 int kvmppc_core_check_processor_compat(void)
520 if (cpu_has_feature(CPU_FTR_HVMODE))
525 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
527 struct kvm_vcpu *vcpu;
530 struct kvmppc_vcore *vcore;
532 core = id / threads_per_core;
533 if (core >= KVM_MAX_VCORES)
537 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
541 err = kvm_vcpu_init(vcpu, kvm, id);
545 vcpu->arch.shared = &vcpu->arch.shregs;
546 vcpu->arch.last_cpu = -1;
547 vcpu->arch.mmcr[0] = MMCR0_FC;
548 vcpu->arch.ctrl = CTRL_RUNLATCH;
549 /* default to host PVR, since we can't spoof it */
550 vcpu->arch.pvr = mfspr(SPRN_PVR);
551 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
552 spin_lock_init(&vcpu->arch.vpa_update_lock);
554 kvmppc_mmu_book3s_hv_init(vcpu);
557 * We consider the vcpu stopped until we see the first run ioctl for it.
559 vcpu->arch.state = KVMPPC_VCPU_STOPPED;
561 init_waitqueue_head(&vcpu->arch.cpu_run);
563 mutex_lock(&kvm->lock);
564 vcore = kvm->arch.vcores[core];
566 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
568 INIT_LIST_HEAD(&vcore->runnable_threads);
569 spin_lock_init(&vcore->lock);
570 init_waitqueue_head(&vcore->wq);
572 kvm->arch.vcores[core] = vcore;
574 mutex_unlock(&kvm->lock);
579 spin_lock(&vcore->lock);
580 ++vcore->num_threads;
581 spin_unlock(&vcore->lock);
582 vcpu->arch.vcore = vcore;
584 vcpu->arch.cpu_type = KVM_CPU_3S_64;
585 kvmppc_sanity_check(vcpu);
590 kmem_cache_free(kvm_vcpu_cache, vcpu);
595 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
597 spin_lock(&vcpu->arch.vpa_update_lock);
598 if (vcpu->arch.dtl.pinned_addr)
599 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl.pinned_addr);
600 if (vcpu->arch.slb_shadow.pinned_addr)
601 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow.pinned_addr);
602 if (vcpu->arch.vpa.pinned_addr)
603 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa.pinned_addr);
604 spin_unlock(&vcpu->arch.vpa_update_lock);
605 kvm_vcpu_uninit(vcpu);
606 kmem_cache_free(kvm_vcpu_cache, vcpu);
609 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
611 unsigned long dec_nsec, now;
614 if (now > vcpu->arch.dec_expires) {
615 /* decrementer has already gone negative */
616 kvmppc_core_queue_dec(vcpu);
617 kvmppc_core_prepare_to_enter(vcpu);
620 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
622 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
624 vcpu->arch.timer_running = 1;
627 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
629 vcpu->arch.ceded = 0;
630 if (vcpu->arch.timer_running) {
631 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
632 vcpu->arch.timer_running = 0;
636 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
637 extern void xics_wake_cpu(int cpu);
639 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
640 struct kvm_vcpu *vcpu)
644 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
646 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
649 /* decrement the physical thread id of each following vcpu */
651 list_for_each_entry_continue(v, &vc->runnable_threads, arch.run_list)
653 list_del(&vcpu->arch.run_list);
656 static int kvmppc_grab_hwthread(int cpu)
658 struct paca_struct *tpaca;
663 /* Ensure the thread won't go into the kernel if it wakes */
664 tpaca->kvm_hstate.hwthread_req = 1;
667 * If the thread is already executing in the kernel (e.g. handling
668 * a stray interrupt), wait for it to get back to nap mode.
669 * The smp_mb() is to ensure that our setting of hwthread_req
670 * is visible before we look at hwthread_state, so if this
671 * races with the code at system_reset_pSeries and the thread
672 * misses our setting of hwthread_req, we are sure to see its
673 * setting of hwthread_state, and vice versa.
676 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
677 if (--timeout <= 0) {
678 pr_err("KVM: couldn't grab cpu %d\n", cpu);
686 static void kvmppc_release_hwthread(int cpu)
688 struct paca_struct *tpaca;
691 tpaca->kvm_hstate.hwthread_req = 0;
692 tpaca->kvm_hstate.kvm_vcpu = NULL;
695 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
698 struct paca_struct *tpaca;
699 struct kvmppc_vcore *vc = vcpu->arch.vcore;
701 if (vcpu->arch.timer_running) {
702 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
703 vcpu->arch.timer_running = 0;
705 cpu = vc->pcpu + vcpu->arch.ptid;
707 tpaca->kvm_hstate.kvm_vcpu = vcpu;
708 tpaca->kvm_hstate.kvm_vcore = vc;
709 tpaca->kvm_hstate.napping = 0;
710 vcpu->cpu = vc->pcpu;
712 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
713 if (vcpu->arch.ptid) {
714 kvmppc_grab_hwthread(cpu);
721 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
727 while (vc->nap_count < vc->n_woken) {
728 if (++i >= 1000000) {
729 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
730 vc->nap_count, vc->n_woken);
739 * Check that we are on thread 0 and that any other threads in
740 * this core are off-line.
742 static int on_primary_thread(void)
744 int cpu = smp_processor_id();
745 int thr = cpu_thread_in_core(cpu);
749 while (++thr < threads_per_core)
750 if (cpu_online(cpu + thr))
756 * Run a set of guest threads on a physical core.
757 * Called with vc->lock held.
759 static int kvmppc_run_core(struct kvmppc_vcore *vc)
761 struct kvm_vcpu *vcpu, *vcpu0, *vnext;
766 /* don't start if any threads have a signal pending */
767 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
768 if (signal_pending(vcpu->arch.run_task))
772 * Make sure we are running on thread 0, and that
773 * secondary threads are offline.
774 * XXX we should also block attempts to bring any
775 * secondary threads online.
777 if (threads_per_core > 1 && !on_primary_thread()) {
778 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
779 vcpu->arch.ret = -EBUSY;
784 * Assign physical thread IDs, first to non-ceded vcpus
785 * and then to ceded ones.
789 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
790 if (!vcpu->arch.ceded) {
793 vcpu->arch.ptid = ptid++;
797 return 0; /* nothing to run */
798 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
799 if (vcpu->arch.ceded)
800 vcpu->arch.ptid = ptid++;
804 vc->entry_exit_count = 0;
805 vc->vcore_state = VCORE_RUNNING;
807 vc->pcpu = smp_processor_id();
808 vc->napping_threads = 0;
809 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
810 kvmppc_start_thread(vcpu);
811 if (vcpu->arch.vpa.update_pending ||
812 vcpu->arch.slb_shadow.update_pending ||
813 vcpu->arch.dtl.update_pending)
814 kvmppc_update_vpas(vcpu);
816 /* Grab any remaining hw threads so they can't go into the kernel */
817 for (i = ptid; i < threads_per_core; ++i)
818 kvmppc_grab_hwthread(vc->pcpu + i);
821 spin_unlock(&vc->lock);
824 __kvmppc_vcore_entry(NULL, vcpu0);
825 for (i = 0; i < threads_per_core; ++i)
826 kvmppc_release_hwthread(vc->pcpu + i);
828 spin_lock(&vc->lock);
829 /* disable sending of IPIs on virtual external irqs */
830 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
832 /* wait for secondary threads to finish writing their state to memory */
833 if (vc->nap_count < vc->n_woken)
834 kvmppc_wait_for_nap(vc);
835 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
836 vc->vcore_state = VCORE_EXITING;
837 spin_unlock(&vc->lock);
839 /* make sure updates to secondary vcpu structs are visible now */
847 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
848 /* cancel pending dec exception if dec is positive */
849 if (now < vcpu->arch.dec_expires &&
850 kvmppc_core_pending_dec(vcpu))
851 kvmppc_core_dequeue_dec(vcpu);
855 ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
856 vcpu->arch.run_task);
858 vcpu->arch.ret = ret;
861 if (vcpu->arch.ceded) {
862 if (ret != RESUME_GUEST)
863 kvmppc_end_cede(vcpu);
865 kvmppc_set_timer(vcpu);
869 spin_lock(&vc->lock);
871 vc->vcore_state = VCORE_INACTIVE;
872 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
874 if (vcpu->arch.ret != RESUME_GUEST) {
875 kvmppc_remove_runnable(vc, vcpu);
876 wake_up(&vcpu->arch.cpu_run);
884 * Wait for some other vcpu thread to execute us, and
885 * wake us up when we need to handle something in the host.
887 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
891 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
892 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
894 finish_wait(&vcpu->arch.cpu_run, &wait);
898 * All the vcpus in this vcore are idle, so wait for a decrementer
899 * or external interrupt to one of the vcpus. vc->lock is held.
901 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
907 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
908 vc->vcore_state = VCORE_SLEEPING;
909 spin_unlock(&vc->lock);
910 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
911 if (!v->arch.ceded || v->arch.pending_exceptions) {
918 finish_wait(&vc->wq, &wait);
919 spin_lock(&vc->lock);
920 vc->vcore_state = VCORE_INACTIVE;
923 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
927 struct kvmppc_vcore *vc;
928 struct kvm_vcpu *v, *vn;
930 kvm_run->exit_reason = 0;
931 vcpu->arch.ret = RESUME_GUEST;
935 * Synchronize with other threads in this virtual core
937 vc = vcpu->arch.vcore;
938 spin_lock(&vc->lock);
939 vcpu->arch.ceded = 0;
940 vcpu->arch.run_task = current;
941 vcpu->arch.kvm_run = kvm_run;
942 prev_state = vcpu->arch.state;
943 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
944 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
948 * This happens the first time this is called for a vcpu.
949 * If the vcore is already running, we may be able to start
950 * this thread straight away and have it join in.
952 if (prev_state == KVMPPC_VCPU_STOPPED) {
953 if (vc->vcore_state == VCORE_RUNNING &&
954 VCORE_EXIT_COUNT(vc) == 0) {
955 vcpu->arch.ptid = vc->n_runnable - 1;
956 kvmppc_start_thread(vcpu);
959 } else if (prev_state == KVMPPC_VCPU_BUSY_IN_HOST)
962 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
963 !signal_pending(current)) {
964 if (vc->n_busy || vc->vcore_state != VCORE_INACTIVE) {
965 spin_unlock(&vc->lock);
966 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
967 spin_lock(&vc->lock);
971 list_for_each_entry(v, &vc->runnable_threads, arch.run_list)
972 n_ceded += v->arch.ceded;
973 if (n_ceded == vc->n_runnable)
974 kvmppc_vcore_blocked(vc);
978 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
980 kvmppc_core_prepare_to_enter(v);
981 if (signal_pending(v->arch.run_task)) {
982 kvmppc_remove_runnable(vc, v);
983 v->stat.signal_exits++;
984 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
985 v->arch.ret = -EINTR;
986 wake_up(&v->arch.cpu_run);
991 if (signal_pending(current)) {
992 if (vc->vcore_state == VCORE_RUNNING ||
993 vc->vcore_state == VCORE_EXITING) {
994 spin_unlock(&vc->lock);
995 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
996 spin_lock(&vc->lock);
998 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
999 kvmppc_remove_runnable(vc, vcpu);
1000 vcpu->stat.signal_exits++;
1001 kvm_run->exit_reason = KVM_EXIT_INTR;
1002 vcpu->arch.ret = -EINTR;
1006 spin_unlock(&vc->lock);
1007 return vcpu->arch.ret;
1010 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1014 if (!vcpu->arch.sane) {
1015 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1019 kvmppc_core_prepare_to_enter(vcpu);
1021 /* No need to go into the guest when all we'll do is come back out */
1022 if (signal_pending(current)) {
1023 run->exit_reason = KVM_EXIT_INTR;
1027 /* On the first time here, set up VRMA or RMA */
1028 if (!vcpu->kvm->arch.rma_setup_done) {
1029 r = kvmppc_hv_setup_rma(vcpu);
1034 flush_fp_to_thread(current);
1035 flush_altivec_to_thread(current);
1036 flush_vsx_to_thread(current);
1037 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1038 vcpu->arch.pgdir = current->mm->pgd;
1041 r = kvmppc_run_vcpu(run, vcpu);
1043 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1044 !(vcpu->arch.shregs.msr & MSR_PR)) {
1045 r = kvmppc_pseries_do_hcall(vcpu);
1046 kvmppc_core_prepare_to_enter(vcpu);
1048 } while (r == RESUME_GUEST);
1052 static long kvmppc_stt_npages(unsigned long window_size)
1054 return ALIGN((window_size >> SPAPR_TCE_SHIFT)
1055 * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
1058 static void release_spapr_tce_table(struct kvmppc_spapr_tce_table *stt)
1060 struct kvm *kvm = stt->kvm;
1063 mutex_lock(&kvm->lock);
1064 list_del(&stt->list);
1065 for (i = 0; i < kvmppc_stt_npages(stt->window_size); i++)
1066 __free_page(stt->pages[i]);
1068 mutex_unlock(&kvm->lock);
1073 static int kvm_spapr_tce_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1075 struct kvmppc_spapr_tce_table *stt = vma->vm_file->private_data;
1078 if (vmf->pgoff >= kvmppc_stt_npages(stt->window_size))
1079 return VM_FAULT_SIGBUS;
1081 page = stt->pages[vmf->pgoff];
1087 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
1088 .fault = kvm_spapr_tce_fault,
1091 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
1093 vma->vm_ops = &kvm_spapr_tce_vm_ops;
1097 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
1099 struct kvmppc_spapr_tce_table *stt = filp->private_data;
1101 release_spapr_tce_table(stt);
1105 static struct file_operations kvm_spapr_tce_fops = {
1106 .mmap = kvm_spapr_tce_mmap,
1107 .release = kvm_spapr_tce_release,
1110 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
1111 struct kvm_create_spapr_tce *args)
1113 struct kvmppc_spapr_tce_table *stt = NULL;
1118 /* Check this LIOBN hasn't been previously allocated */
1119 list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) {
1120 if (stt->liobn == args->liobn)
1124 npages = kvmppc_stt_npages(args->window_size);
1126 stt = kzalloc(sizeof(*stt) + npages* sizeof(struct page *),
1131 stt->liobn = args->liobn;
1132 stt->window_size = args->window_size;
1135 for (i = 0; i < npages; i++) {
1136 stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
1143 mutex_lock(&kvm->lock);
1144 list_add(&stt->list, &kvm->arch.spapr_tce_tables);
1146 mutex_unlock(&kvm->lock);
1148 return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
1153 for (i = 0; i < npages; i++)
1155 __free_page(stt->pages[i]);
1162 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1163 Assumes POWER7 or PPC970. */
1164 static inline int lpcr_rmls(unsigned long rma_size)
1167 case 32ul << 20: /* 32 MB */
1168 if (cpu_has_feature(CPU_FTR_ARCH_206))
1169 return 8; /* only supported on POWER7 */
1171 case 64ul << 20: /* 64 MB */
1173 case 128ul << 20: /* 128 MB */
1175 case 256ul << 20: /* 256 MB */
1177 case 1ul << 30: /* 1 GB */
1179 case 16ul << 30: /* 16 GB */
1181 case 256ul << 30: /* 256 GB */
1188 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1190 struct kvmppc_linear_info *ri = vma->vm_file->private_data;
1193 if (vmf->pgoff >= ri->npages)
1194 return VM_FAULT_SIGBUS;
1196 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1202 static const struct vm_operations_struct kvm_rma_vm_ops = {
1203 .fault = kvm_rma_fault,
1206 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1208 vma->vm_flags |= VM_RESERVED;
1209 vma->vm_ops = &kvm_rma_vm_ops;
1213 static int kvm_rma_release(struct inode *inode, struct file *filp)
1215 struct kvmppc_linear_info *ri = filp->private_data;
1217 kvm_release_rma(ri);
1221 static struct file_operations kvm_rma_fops = {
1222 .mmap = kvm_rma_mmap,
1223 .release = kvm_rma_release,
1226 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1228 struct kvmppc_linear_info *ri;
1231 ri = kvm_alloc_rma();
1235 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1237 kvm_release_rma(ri);
1239 ret->rma_size = ri->npages << PAGE_SHIFT;
1244 * Get (and clear) the dirty memory log for a memory slot.
1246 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1248 struct kvm_memory_slot *memslot;
1252 mutex_lock(&kvm->slots_lock);
1255 if (log->slot >= KVM_MEMORY_SLOTS)
1258 memslot = id_to_memslot(kvm->memslots, log->slot);
1260 if (!memslot->dirty_bitmap)
1263 n = kvm_dirty_bitmap_bytes(memslot);
1264 memset(memslot->dirty_bitmap, 0, n);
1266 r = kvmppc_hv_get_dirty_log(kvm, memslot);
1271 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1276 mutex_unlock(&kvm->slots_lock);
1280 static unsigned long slb_pgsize_encoding(unsigned long psize)
1282 unsigned long senc = 0;
1284 if (psize > 0x1000) {
1286 if (psize == 0x10000)
1287 senc |= SLB_VSID_LP_01;
1292 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1293 struct kvm_userspace_memory_region *mem)
1295 unsigned long npages;
1296 unsigned long *phys;
1298 /* Allocate a slot_phys array */
1299 phys = kvm->arch.slot_phys[mem->slot];
1300 if (!kvm->arch.using_mmu_notifiers && !phys) {
1301 npages = mem->memory_size >> PAGE_SHIFT;
1302 phys = vzalloc(npages * sizeof(unsigned long));
1305 kvm->arch.slot_phys[mem->slot] = phys;
1306 kvm->arch.slot_npages[mem->slot] = npages;
1312 static void unpin_slot(struct kvm *kvm, int slot_id)
1314 unsigned long *physp;
1315 unsigned long j, npages, pfn;
1318 physp = kvm->arch.slot_phys[slot_id];
1319 npages = kvm->arch.slot_npages[slot_id];
1321 spin_lock(&kvm->arch.slot_phys_lock);
1322 for (j = 0; j < npages; j++) {
1323 if (!(physp[j] & KVMPPC_GOT_PAGE))
1325 pfn = physp[j] >> PAGE_SHIFT;
1326 page = pfn_to_page(pfn);
1328 page = compound_head(page);
1332 kvm->arch.slot_phys[slot_id] = NULL;
1333 spin_unlock(&kvm->arch.slot_phys_lock);
1338 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1339 struct kvm_userspace_memory_region *mem)
1343 static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu)
1346 struct kvm *kvm = vcpu->kvm;
1347 struct kvmppc_linear_info *ri = NULL;
1349 struct kvm_memory_slot *memslot;
1350 struct vm_area_struct *vma;
1351 unsigned long lpcr, senc;
1352 unsigned long psize, porder;
1353 unsigned long rma_size;
1355 unsigned long *physp;
1356 unsigned long i, npages;
1358 mutex_lock(&kvm->lock);
1359 if (kvm->arch.rma_setup_done)
1360 goto out; /* another vcpu beat us to it */
1362 /* Look up the memslot for guest physical address 0 */
1363 memslot = gfn_to_memslot(kvm, 0);
1365 /* We must have some memory at 0 by now */
1367 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1370 /* Look up the VMA for the start of this memory slot */
1371 hva = memslot->userspace_addr;
1372 down_read(¤t->mm->mmap_sem);
1373 vma = find_vma(current->mm, hva);
1374 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1377 psize = vma_kernel_pagesize(vma);
1378 porder = __ilog2(psize);
1380 /* Is this one of our preallocated RMAs? */
1381 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1382 hva == vma->vm_start)
1383 ri = vma->vm_file->private_data;
1385 up_read(¤t->mm->mmap_sem);
1388 /* On POWER7, use VRMA; on PPC970, give up */
1390 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1391 pr_err("KVM: CPU requires an RMO\n");
1395 /* We can handle 4k, 64k or 16M pages in the VRMA */
1397 if (!(psize == 0x1000 || psize == 0x10000 ||
1398 psize == 0x1000000))
1401 /* Update VRMASD field in the LPCR */
1402 senc = slb_pgsize_encoding(psize);
1403 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1404 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1405 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1406 lpcr |= senc << (LPCR_VRMASD_SH - 4);
1407 kvm->arch.lpcr = lpcr;
1409 /* Create HPTEs in the hash page table for the VRMA */
1410 kvmppc_map_vrma(vcpu, memslot, porder);
1413 /* Set up to use an RMO region */
1414 rma_size = ri->npages;
1415 if (rma_size > memslot->npages)
1416 rma_size = memslot->npages;
1417 rma_size <<= PAGE_SHIFT;
1418 rmls = lpcr_rmls(rma_size);
1421 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1424 atomic_inc(&ri->use_count);
1427 /* Update LPCR and RMOR */
1428 lpcr = kvm->arch.lpcr;
1429 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1430 /* PPC970; insert RMLS value (split field) in HID4 */
1431 lpcr &= ~((1ul << HID4_RMLS0_SH) |
1432 (3ul << HID4_RMLS2_SH));
1433 lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1434 ((rmls & 3) << HID4_RMLS2_SH);
1435 /* RMOR is also in HID4 */
1436 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1440 lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1441 lpcr |= rmls << LPCR_RMLS_SH;
1442 kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1444 kvm->arch.lpcr = lpcr;
1445 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1446 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1448 /* Initialize phys addrs of pages in RMO */
1449 npages = ri->npages;
1450 porder = __ilog2(npages);
1451 physp = kvm->arch.slot_phys[memslot->id];
1452 spin_lock(&kvm->arch.slot_phys_lock);
1453 for (i = 0; i < npages; ++i)
1454 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) + porder;
1455 spin_unlock(&kvm->arch.slot_phys_lock);
1458 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1460 kvm->arch.rma_setup_done = 1;
1463 mutex_unlock(&kvm->lock);
1467 up_read(¤t->mm->mmap_sem);
1471 int kvmppc_core_init_vm(struct kvm *kvm)
1476 /* Allocate hashed page table */
1477 r = kvmppc_alloc_hpt(kvm);
1481 INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1483 kvm->arch.rma = NULL;
1485 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1487 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1488 /* PPC970; HID4 is effectively the LPCR */
1489 unsigned long lpid = kvm->arch.lpid;
1490 kvm->arch.host_lpid = 0;
1491 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1492 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1493 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1494 ((lpid & 0xf) << HID4_LPID5_SH);
1496 /* POWER7; init LPCR for virtual RMA mode */
1497 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1498 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1499 lpcr &= LPCR_PECE | LPCR_LPES;
1500 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1501 LPCR_VPM0 | LPCR_VPM1;
1502 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1503 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1505 kvm->arch.lpcr = lpcr;
1507 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1508 spin_lock_init(&kvm->arch.slot_phys_lock);
1512 void kvmppc_core_destroy_vm(struct kvm *kvm)
1516 if (!kvm->arch.using_mmu_notifiers)
1517 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
1520 if (kvm->arch.rma) {
1521 kvm_release_rma(kvm->arch.rma);
1522 kvm->arch.rma = NULL;
1525 kvmppc_free_hpt(kvm);
1526 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1529 /* These are stubs for now */
1530 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1534 /* We don't need to emulate any privileged instructions or dcbz */
1535 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1536 unsigned int inst, int *advance)
1538 return EMULATE_FAIL;
1541 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, int rs)
1543 return EMULATE_FAIL;
1546 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, int rt)
1548 return EMULATE_FAIL;
1551 static int kvmppc_book3s_hv_init(void)
1555 r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1560 r = kvmppc_mmu_hv_init();
1565 static void kvmppc_book3s_hv_exit(void)
1570 module_init(kvmppc_book3s_hv_init);
1571 module_exit(kvmppc_book3s_hv_exit);