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>
33 #include <linux/srcu.h>
36 #include <asm/cputable.h>
37 #include <asm/cacheflush.h>
38 #include <asm/tlbflush.h>
39 #include <asm/uaccess.h>
41 #include <asm/kvm_ppc.h>
42 #include <asm/kvm_book3s.h>
43 #include <asm/mmu_context.h>
44 #include <asm/lppaca.h>
45 #include <asm/processor.h>
46 #include <asm/cputhreads.h>
48 #include <asm/hvcall.h>
49 #include <asm/switch_to.h>
51 #include <linux/gfp.h>
52 #include <linux/vmalloc.h>
53 #include <linux/highmem.h>
54 #include <linux/hugetlb.h>
56 /* #define EXIT_DEBUG */
57 /* #define EXIT_DEBUG_SIMPLE */
58 /* #define EXIT_DEBUG_INT */
60 /* Used to indicate that a guest page fault needs to be handled */
61 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
63 /* Used as a "null" value for timebase values */
64 #define TB_NIL (~(u64)0)
66 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
67 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
69 void kvmppc_fast_vcpu_kick(struct kvm_vcpu *vcpu)
73 wait_queue_head_t *wqp;
75 wqp = kvm_arch_vcpu_wq(vcpu);
76 if (waitqueue_active(wqp)) {
77 wake_up_interruptible(wqp);
78 ++vcpu->stat.halt_wakeup;
83 /* CPU points to the first thread of the core */
84 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
85 int real_cpu = cpu + vcpu->arch.ptid;
86 if (paca[real_cpu].kvm_hstate.xics_phys)
87 xics_wake_cpu(real_cpu);
88 else if (cpu_online(cpu))
89 smp_send_reschedule(cpu);
95 * We use the vcpu_load/put functions to measure stolen time.
96 * Stolen time is counted as time when either the vcpu is able to
97 * run as part of a virtual core, but the task running the vcore
98 * is preempted or sleeping, or when the vcpu needs something done
99 * in the kernel by the task running the vcpu, but that task is
100 * preempted or sleeping. Those two things have to be counted
101 * separately, since one of the vcpu tasks will take on the job
102 * of running the core, and the other vcpu tasks in the vcore will
103 * sleep waiting for it to do that, but that sleep shouldn't count
106 * Hence we accumulate stolen time when the vcpu can run as part of
107 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
108 * needs its task to do other things in the kernel (for example,
109 * service a page fault) in busy_stolen. We don't accumulate
110 * stolen time for a vcore when it is inactive, or for a vcpu
111 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
112 * a misnomer; it means that the vcpu task is not executing in
113 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
114 * the kernel. We don't have any way of dividing up that time
115 * between time that the vcpu is genuinely stopped, time that
116 * the task is actively working on behalf of the vcpu, and time
117 * that the task is preempted, so we don't count any of it as
120 * Updates to busy_stolen are protected by arch.tbacct_lock;
121 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
122 * of the vcpu that has taken responsibility for running the vcore
123 * (i.e. vc->runner). The stolen times are measured in units of
124 * timebase ticks. (Note that the != TB_NIL checks below are
125 * purely defensive; they should never fail.)
128 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
130 struct kvmppc_vcore *vc = vcpu->arch.vcore;
132 spin_lock(&vcpu->arch.tbacct_lock);
133 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
134 vc->preempt_tb != TB_NIL) {
135 vc->stolen_tb += mftb() - vc->preempt_tb;
136 vc->preempt_tb = TB_NIL;
138 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
139 vcpu->arch.busy_preempt != TB_NIL) {
140 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
141 vcpu->arch.busy_preempt = TB_NIL;
143 spin_unlock(&vcpu->arch.tbacct_lock);
146 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
148 struct kvmppc_vcore *vc = vcpu->arch.vcore;
150 spin_lock(&vcpu->arch.tbacct_lock);
151 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
152 vc->preempt_tb = mftb();
153 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
154 vcpu->arch.busy_preempt = mftb();
155 spin_unlock(&vcpu->arch.tbacct_lock);
158 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
160 vcpu->arch.shregs.msr = msr;
161 kvmppc_end_cede(vcpu);
164 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
166 vcpu->arch.pvr = pvr;
169 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
173 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
174 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
175 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
176 for (r = 0; r < 16; ++r)
177 pr_err("r%2d = %.16lx r%d = %.16lx\n",
178 r, kvmppc_get_gpr(vcpu, r),
179 r+16, kvmppc_get_gpr(vcpu, r+16));
180 pr_err("ctr = %.16lx lr = %.16lx\n",
181 vcpu->arch.ctr, vcpu->arch.lr);
182 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
183 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
184 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
185 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
186 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
187 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
188 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
189 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
190 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
191 pr_err("fault dar = %.16lx dsisr = %.8x\n",
192 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
193 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
194 for (r = 0; r < vcpu->arch.slb_max; ++r)
195 pr_err(" ESID = %.16llx VSID = %.16llx\n",
196 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
197 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
198 vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
199 vcpu->arch.last_inst);
202 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
205 struct kvm_vcpu *v, *ret = NULL;
207 mutex_lock(&kvm->lock);
208 kvm_for_each_vcpu(r, v, kvm) {
209 if (v->vcpu_id == id) {
214 mutex_unlock(&kvm->lock);
218 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
220 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
221 vpa->yield_count = 1;
224 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
225 unsigned long addr, unsigned long len)
227 /* check address is cacheline aligned */
228 if (addr & (L1_CACHE_BYTES - 1))
230 spin_lock(&vcpu->arch.vpa_update_lock);
231 if (v->next_gpa != addr || v->len != len) {
233 v->len = addr ? len : 0;
234 v->update_pending = 1;
236 spin_unlock(&vcpu->arch.vpa_update_lock);
240 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
249 static int vpa_is_registered(struct kvmppc_vpa *vpap)
251 if (vpap->update_pending)
252 return vpap->next_gpa != 0;
253 return vpap->pinned_addr != NULL;
256 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
258 unsigned long vcpuid, unsigned long vpa)
260 struct kvm *kvm = vcpu->kvm;
261 unsigned long len, nb;
263 struct kvm_vcpu *tvcpu;
266 struct kvmppc_vpa *vpap;
268 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
272 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
273 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
274 subfunc == H_VPA_REG_SLB) {
275 /* Registering new area - address must be cache-line aligned */
276 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
279 /* convert logical addr to kernel addr and read length */
280 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
283 if (subfunc == H_VPA_REG_VPA)
284 len = ((struct reg_vpa *)va)->length.hword;
286 len = ((struct reg_vpa *)va)->length.word;
287 kvmppc_unpin_guest_page(kvm, va, vpa, false);
290 if (len > nb || len < sizeof(struct reg_vpa))
299 spin_lock(&tvcpu->arch.vpa_update_lock);
302 case H_VPA_REG_VPA: /* register VPA */
303 if (len < sizeof(struct lppaca))
305 vpap = &tvcpu->arch.vpa;
309 case H_VPA_REG_DTL: /* register DTL */
310 if (len < sizeof(struct dtl_entry))
312 len -= len % sizeof(struct dtl_entry);
314 /* Check that they have previously registered a VPA */
316 if (!vpa_is_registered(&tvcpu->arch.vpa))
319 vpap = &tvcpu->arch.dtl;
323 case H_VPA_REG_SLB: /* register SLB shadow buffer */
324 /* Check that they have previously registered a VPA */
326 if (!vpa_is_registered(&tvcpu->arch.vpa))
329 vpap = &tvcpu->arch.slb_shadow;
333 case H_VPA_DEREG_VPA: /* deregister VPA */
334 /* Check they don't still have a DTL or SLB buf registered */
336 if (vpa_is_registered(&tvcpu->arch.dtl) ||
337 vpa_is_registered(&tvcpu->arch.slb_shadow))
340 vpap = &tvcpu->arch.vpa;
344 case H_VPA_DEREG_DTL: /* deregister DTL */
345 vpap = &tvcpu->arch.dtl;
349 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
350 vpap = &tvcpu->arch.slb_shadow;
356 vpap->next_gpa = vpa;
358 vpap->update_pending = 1;
361 spin_unlock(&tvcpu->arch.vpa_update_lock);
366 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
368 struct kvm *kvm = vcpu->kvm;
374 * We need to pin the page pointed to by vpap->next_gpa,
375 * but we can't call kvmppc_pin_guest_page under the lock
376 * as it does get_user_pages() and down_read(). So we
377 * have to drop the lock, pin the page, then get the lock
378 * again and check that a new area didn't get registered
382 gpa = vpap->next_gpa;
383 spin_unlock(&vcpu->arch.vpa_update_lock);
387 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
388 spin_lock(&vcpu->arch.vpa_update_lock);
389 if (gpa == vpap->next_gpa)
391 /* sigh... unpin that one and try again */
393 kvmppc_unpin_guest_page(kvm, va, gpa, false);
396 vpap->update_pending = 0;
397 if (va && nb < vpap->len) {
399 * If it's now too short, it must be that userspace
400 * has changed the mappings underlying guest memory,
401 * so unregister the region.
403 kvmppc_unpin_guest_page(kvm, va, gpa, false);
406 if (vpap->pinned_addr)
407 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
410 vpap->pinned_addr = va;
413 vpap->pinned_end = va + vpap->len;
416 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
418 if (!(vcpu->arch.vpa.update_pending ||
419 vcpu->arch.slb_shadow.update_pending ||
420 vcpu->arch.dtl.update_pending))
423 spin_lock(&vcpu->arch.vpa_update_lock);
424 if (vcpu->arch.vpa.update_pending) {
425 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
426 if (vcpu->arch.vpa.pinned_addr)
427 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
429 if (vcpu->arch.dtl.update_pending) {
430 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
431 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
432 vcpu->arch.dtl_index = 0;
434 if (vcpu->arch.slb_shadow.update_pending)
435 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
436 spin_unlock(&vcpu->arch.vpa_update_lock);
440 * Return the accumulated stolen time for the vcore up until `now'.
441 * The caller should hold the vcore lock.
443 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
448 * If we are the task running the vcore, then since we hold
449 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
450 * can't be updated, so we don't need the tbacct_lock.
451 * If the vcore is inactive, it can't become active (since we
452 * hold the vcore lock), so the vcpu load/put functions won't
453 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
455 if (vc->vcore_state != VCORE_INACTIVE &&
456 vc->runner->arch.run_task != current) {
457 spin_lock(&vc->runner->arch.tbacct_lock);
459 if (vc->preempt_tb != TB_NIL)
460 p += now - vc->preempt_tb;
461 spin_unlock(&vc->runner->arch.tbacct_lock);
468 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
469 struct kvmppc_vcore *vc)
471 struct dtl_entry *dt;
473 unsigned long stolen;
474 unsigned long core_stolen;
477 dt = vcpu->arch.dtl_ptr;
478 vpa = vcpu->arch.vpa.pinned_addr;
480 core_stolen = vcore_stolen_time(vc, now);
481 stolen = core_stolen - vcpu->arch.stolen_logged;
482 vcpu->arch.stolen_logged = core_stolen;
483 spin_lock(&vcpu->arch.tbacct_lock);
484 stolen += vcpu->arch.busy_stolen;
485 vcpu->arch.busy_stolen = 0;
486 spin_unlock(&vcpu->arch.tbacct_lock);
489 memset(dt, 0, sizeof(struct dtl_entry));
490 dt->dispatch_reason = 7;
491 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
492 dt->timebase = now + vc->tb_offset;
493 dt->enqueue_to_dispatch_time = stolen;
494 dt->srr0 = kvmppc_get_pc(vcpu);
495 dt->srr1 = vcpu->arch.shregs.msr;
497 if (dt == vcpu->arch.dtl.pinned_end)
498 dt = vcpu->arch.dtl.pinned_addr;
499 vcpu->arch.dtl_ptr = dt;
500 /* order writing *dt vs. writing vpa->dtl_idx */
502 vpa->dtl_idx = ++vcpu->arch.dtl_index;
503 vcpu->arch.dtl.dirty = true;
506 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
508 unsigned long req = kvmppc_get_gpr(vcpu, 3);
509 unsigned long target, ret = H_SUCCESS;
510 struct kvm_vcpu *tvcpu;
515 idx = srcu_read_lock(&vcpu->kvm->srcu);
516 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
517 kvmppc_get_gpr(vcpu, 5),
518 kvmppc_get_gpr(vcpu, 6),
519 kvmppc_get_gpr(vcpu, 7));
520 srcu_read_unlock(&vcpu->kvm->srcu, idx);
525 target = kvmppc_get_gpr(vcpu, 4);
526 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
531 tvcpu->arch.prodded = 1;
533 if (vcpu->arch.ceded) {
534 if (waitqueue_active(&vcpu->wq)) {
535 wake_up_interruptible(&vcpu->wq);
536 vcpu->stat.halt_wakeup++;
541 target = kvmppc_get_gpr(vcpu, 4);
544 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
549 kvm_vcpu_yield_to(tvcpu);
552 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
553 kvmppc_get_gpr(vcpu, 5),
554 kvmppc_get_gpr(vcpu, 6));
557 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
560 rc = kvmppc_rtas_hcall(vcpu);
567 /* Send the error out to userspace via KVM_RUN */
576 if (kvmppc_xics_enabled(vcpu)) {
577 ret = kvmppc_xics_hcall(vcpu, req);
583 kvmppc_set_gpr(vcpu, 3, ret);
584 vcpu->arch.hcall_needed = 0;
588 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
589 struct task_struct *tsk)
593 vcpu->stat.sum_exits++;
595 run->exit_reason = KVM_EXIT_UNKNOWN;
596 run->ready_for_interrupt_injection = 1;
597 switch (vcpu->arch.trap) {
598 /* We're good on these - the host merely wanted to get our attention */
599 case BOOK3S_INTERRUPT_HV_DECREMENTER:
600 vcpu->stat.dec_exits++;
603 case BOOK3S_INTERRUPT_EXTERNAL:
604 vcpu->stat.ext_intr_exits++;
607 case BOOK3S_INTERRUPT_PERFMON:
610 case BOOK3S_INTERRUPT_MACHINE_CHECK:
612 * Deliver a machine check interrupt to the guest.
613 * We have to do this, even if the host has handled the
614 * machine check, because machine checks use SRR0/1 and
615 * the interrupt might have trashed guest state in them.
617 kvmppc_book3s_queue_irqprio(vcpu,
618 BOOK3S_INTERRUPT_MACHINE_CHECK);
621 case BOOK3S_INTERRUPT_PROGRAM:
625 * Normally program interrupts are delivered directly
626 * to the guest by the hardware, but we can get here
627 * as a result of a hypervisor emulation interrupt
628 * (e40) getting turned into a 700 by BML RTAS.
630 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
631 kvmppc_core_queue_program(vcpu, flags);
635 case BOOK3S_INTERRUPT_SYSCALL:
637 /* hcall - punt to userspace */
640 if (vcpu->arch.shregs.msr & MSR_PR) {
641 /* sc 1 from userspace - reflect to guest syscall */
642 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
646 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
647 for (i = 0; i < 9; ++i)
648 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
649 run->exit_reason = KVM_EXIT_PAPR_HCALL;
650 vcpu->arch.hcall_needed = 1;
655 * We get these next two if the guest accesses a page which it thinks
656 * it has mapped but which is not actually present, either because
657 * it is for an emulated I/O device or because the corresonding
658 * host page has been paged out. Any other HDSI/HISI interrupts
659 * have been handled already.
661 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
662 r = RESUME_PAGE_FAULT;
664 case BOOK3S_INTERRUPT_H_INST_STORAGE:
665 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
666 vcpu->arch.fault_dsisr = 0;
667 r = RESUME_PAGE_FAULT;
670 * This occurs if the guest executes an illegal instruction.
671 * We just generate a program interrupt to the guest, since
672 * we don't emulate any guest instructions at this stage.
674 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
675 kvmppc_core_queue_program(vcpu, 0x80000);
679 kvmppc_dump_regs(vcpu);
680 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
681 vcpu->arch.trap, kvmppc_get_pc(vcpu),
682 vcpu->arch.shregs.msr);
691 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
692 struct kvm_sregs *sregs)
696 memset(sregs, 0, sizeof(struct kvm_sregs));
697 sregs->pvr = vcpu->arch.pvr;
698 for (i = 0; i < vcpu->arch.slb_max; i++) {
699 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
700 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
706 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
707 struct kvm_sregs *sregs)
711 kvmppc_set_pvr(vcpu, sregs->pvr);
714 for (i = 0; i < vcpu->arch.slb_nr; i++) {
715 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
716 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
717 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
721 vcpu->arch.slb_max = j;
726 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
732 case KVM_REG_PPC_HIOR:
733 *val = get_reg_val(id, 0);
735 case KVM_REG_PPC_DABR:
736 *val = get_reg_val(id, vcpu->arch.dabr);
738 case KVM_REG_PPC_DSCR:
739 *val = get_reg_val(id, vcpu->arch.dscr);
741 case KVM_REG_PPC_PURR:
742 *val = get_reg_val(id, vcpu->arch.purr);
744 case KVM_REG_PPC_SPURR:
745 *val = get_reg_val(id, vcpu->arch.spurr);
747 case KVM_REG_PPC_AMR:
748 *val = get_reg_val(id, vcpu->arch.amr);
750 case KVM_REG_PPC_UAMOR:
751 *val = get_reg_val(id, vcpu->arch.uamor);
753 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
754 i = id - KVM_REG_PPC_MMCR0;
755 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
757 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
758 i = id - KVM_REG_PPC_PMC1;
759 *val = get_reg_val(id, vcpu->arch.pmc[i]);
761 case KVM_REG_PPC_SIAR:
762 *val = get_reg_val(id, vcpu->arch.siar);
764 case KVM_REG_PPC_SDAR:
765 *val = get_reg_val(id, vcpu->arch.sdar);
768 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
769 if (cpu_has_feature(CPU_FTR_VSX)) {
770 /* VSX => FP reg i is stored in arch.vsr[2*i] */
771 long int i = id - KVM_REG_PPC_FPR0;
772 *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
774 /* let generic code handle it */
778 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
779 if (cpu_has_feature(CPU_FTR_VSX)) {
780 long int i = id - KVM_REG_PPC_VSR0;
781 val->vsxval[0] = vcpu->arch.vsr[2 * i];
782 val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
787 #endif /* CONFIG_VSX */
788 case KVM_REG_PPC_VPA_ADDR:
789 spin_lock(&vcpu->arch.vpa_update_lock);
790 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
791 spin_unlock(&vcpu->arch.vpa_update_lock);
793 case KVM_REG_PPC_VPA_SLB:
794 spin_lock(&vcpu->arch.vpa_update_lock);
795 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
796 val->vpaval.length = vcpu->arch.slb_shadow.len;
797 spin_unlock(&vcpu->arch.vpa_update_lock);
799 case KVM_REG_PPC_VPA_DTL:
800 spin_lock(&vcpu->arch.vpa_update_lock);
801 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
802 val->vpaval.length = vcpu->arch.dtl.len;
803 spin_unlock(&vcpu->arch.vpa_update_lock);
805 case KVM_REG_PPC_TB_OFFSET:
806 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
816 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
820 unsigned long addr, len;
823 case KVM_REG_PPC_HIOR:
824 /* Only allow this to be set to zero */
825 if (set_reg_val(id, *val))
828 case KVM_REG_PPC_DABR:
829 vcpu->arch.dabr = set_reg_val(id, *val);
831 case KVM_REG_PPC_DSCR:
832 vcpu->arch.dscr = set_reg_val(id, *val);
834 case KVM_REG_PPC_PURR:
835 vcpu->arch.purr = set_reg_val(id, *val);
837 case KVM_REG_PPC_SPURR:
838 vcpu->arch.spurr = set_reg_val(id, *val);
840 case KVM_REG_PPC_AMR:
841 vcpu->arch.amr = set_reg_val(id, *val);
843 case KVM_REG_PPC_UAMOR:
844 vcpu->arch.uamor = set_reg_val(id, *val);
846 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
847 i = id - KVM_REG_PPC_MMCR0;
848 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
850 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
851 i = id - KVM_REG_PPC_PMC1;
852 vcpu->arch.pmc[i] = set_reg_val(id, *val);
854 case KVM_REG_PPC_SIAR:
855 vcpu->arch.siar = set_reg_val(id, *val);
857 case KVM_REG_PPC_SDAR:
858 vcpu->arch.sdar = set_reg_val(id, *val);
861 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
862 if (cpu_has_feature(CPU_FTR_VSX)) {
863 /* VSX => FP reg i is stored in arch.vsr[2*i] */
864 long int i = id - KVM_REG_PPC_FPR0;
865 vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
867 /* let generic code handle it */
871 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
872 if (cpu_has_feature(CPU_FTR_VSX)) {
873 long int i = id - KVM_REG_PPC_VSR0;
874 vcpu->arch.vsr[2 * i] = val->vsxval[0];
875 vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
880 #endif /* CONFIG_VSX */
881 case KVM_REG_PPC_VPA_ADDR:
882 addr = set_reg_val(id, *val);
884 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
885 vcpu->arch.dtl.next_gpa))
887 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
889 case KVM_REG_PPC_VPA_SLB:
890 addr = val->vpaval.addr;
891 len = val->vpaval.length;
893 if (addr && !vcpu->arch.vpa.next_gpa)
895 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
897 case KVM_REG_PPC_VPA_DTL:
898 addr = val->vpaval.addr;
899 len = val->vpaval.length;
901 if (addr && (len < sizeof(struct dtl_entry) ||
902 !vcpu->arch.vpa.next_gpa))
904 len -= len % sizeof(struct dtl_entry);
905 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
907 case KVM_REG_PPC_TB_OFFSET:
908 /* round up to multiple of 2^24 */
909 vcpu->arch.vcore->tb_offset =
910 ALIGN(set_reg_val(id, *val), 1UL << 24);
920 int kvmppc_core_check_processor_compat(void)
922 if (cpu_has_feature(CPU_FTR_HVMODE))
927 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
929 struct kvm_vcpu *vcpu;
932 struct kvmppc_vcore *vcore;
934 core = id / threads_per_core;
935 if (core >= KVM_MAX_VCORES)
939 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
943 err = kvm_vcpu_init(vcpu, kvm, id);
947 vcpu->arch.shared = &vcpu->arch.shregs;
948 vcpu->arch.mmcr[0] = MMCR0_FC;
949 vcpu->arch.ctrl = CTRL_RUNLATCH;
950 /* default to host PVR, since we can't spoof it */
951 vcpu->arch.pvr = mfspr(SPRN_PVR);
952 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
953 spin_lock_init(&vcpu->arch.vpa_update_lock);
954 spin_lock_init(&vcpu->arch.tbacct_lock);
955 vcpu->arch.busy_preempt = TB_NIL;
957 kvmppc_mmu_book3s_hv_init(vcpu);
959 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
961 init_waitqueue_head(&vcpu->arch.cpu_run);
963 mutex_lock(&kvm->lock);
964 vcore = kvm->arch.vcores[core];
966 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
968 INIT_LIST_HEAD(&vcore->runnable_threads);
969 spin_lock_init(&vcore->lock);
970 init_waitqueue_head(&vcore->wq);
971 vcore->preempt_tb = TB_NIL;
973 kvm->arch.vcores[core] = vcore;
974 kvm->arch.online_vcores++;
976 mutex_unlock(&kvm->lock);
981 spin_lock(&vcore->lock);
982 ++vcore->num_threads;
983 spin_unlock(&vcore->lock);
984 vcpu->arch.vcore = vcore;
986 vcpu->arch.cpu_type = KVM_CPU_3S_64;
987 kvmppc_sanity_check(vcpu);
992 kmem_cache_free(kvm_vcpu_cache, vcpu);
997 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
999 if (vpa->pinned_addr)
1000 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1004 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1006 spin_lock(&vcpu->arch.vpa_update_lock);
1007 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1008 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1009 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1010 spin_unlock(&vcpu->arch.vpa_update_lock);
1011 kvm_vcpu_uninit(vcpu);
1012 kmem_cache_free(kvm_vcpu_cache, vcpu);
1015 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1017 unsigned long dec_nsec, now;
1020 if (now > vcpu->arch.dec_expires) {
1021 /* decrementer has already gone negative */
1022 kvmppc_core_queue_dec(vcpu);
1023 kvmppc_core_prepare_to_enter(vcpu);
1026 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1028 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1030 vcpu->arch.timer_running = 1;
1033 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1035 vcpu->arch.ceded = 0;
1036 if (vcpu->arch.timer_running) {
1037 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1038 vcpu->arch.timer_running = 0;
1042 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1044 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1045 struct kvm_vcpu *vcpu)
1049 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1051 spin_lock(&vcpu->arch.tbacct_lock);
1053 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1054 vcpu->arch.stolen_logged;
1055 vcpu->arch.busy_preempt = now;
1056 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1057 spin_unlock(&vcpu->arch.tbacct_lock);
1059 list_del(&vcpu->arch.run_list);
1062 static int kvmppc_grab_hwthread(int cpu)
1064 struct paca_struct *tpaca;
1065 long timeout = 1000;
1069 /* Ensure the thread won't go into the kernel if it wakes */
1070 tpaca->kvm_hstate.hwthread_req = 1;
1071 tpaca->kvm_hstate.kvm_vcpu = NULL;
1074 * If the thread is already executing in the kernel (e.g. handling
1075 * a stray interrupt), wait for it to get back to nap mode.
1076 * The smp_mb() is to ensure that our setting of hwthread_req
1077 * is visible before we look at hwthread_state, so if this
1078 * races with the code at system_reset_pSeries and the thread
1079 * misses our setting of hwthread_req, we are sure to see its
1080 * setting of hwthread_state, and vice versa.
1083 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1084 if (--timeout <= 0) {
1085 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1093 static void kvmppc_release_hwthread(int cpu)
1095 struct paca_struct *tpaca;
1098 tpaca->kvm_hstate.hwthread_req = 0;
1099 tpaca->kvm_hstate.kvm_vcpu = NULL;
1102 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1105 struct paca_struct *tpaca;
1106 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1108 if (vcpu->arch.timer_running) {
1109 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1110 vcpu->arch.timer_running = 0;
1112 cpu = vc->pcpu + vcpu->arch.ptid;
1114 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1115 tpaca->kvm_hstate.kvm_vcore = vc;
1116 tpaca->kvm_hstate.napping = 0;
1117 vcpu->cpu = vc->pcpu;
1119 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1120 if (vcpu->arch.ptid) {
1127 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1133 while (vc->nap_count < vc->n_woken) {
1134 if (++i >= 1000000) {
1135 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1136 vc->nap_count, vc->n_woken);
1145 * Check that we are on thread 0 and that any other threads in
1146 * this core are off-line. Then grab the threads so they can't
1149 static int on_primary_thread(void)
1151 int cpu = smp_processor_id();
1152 int thr = cpu_thread_in_core(cpu);
1156 while (++thr < threads_per_core)
1157 if (cpu_online(cpu + thr))
1160 /* Grab all hw threads so they can't go into the kernel */
1161 for (thr = 1; thr < threads_per_core; ++thr) {
1162 if (kvmppc_grab_hwthread(cpu + thr)) {
1163 /* Couldn't grab one; let the others go */
1165 kvmppc_release_hwthread(cpu + thr);
1166 } while (--thr > 0);
1174 * Run a set of guest threads on a physical core.
1175 * Called with vc->lock held.
1177 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1179 struct kvm_vcpu *vcpu, *vcpu0, *vnext;
1182 int ptid, i, need_vpa_update;
1184 struct kvm_vcpu *vcpus_to_update[threads_per_core];
1186 /* don't start if any threads have a signal pending */
1187 need_vpa_update = 0;
1188 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1189 if (signal_pending(vcpu->arch.run_task))
1191 if (vcpu->arch.vpa.update_pending ||
1192 vcpu->arch.slb_shadow.update_pending ||
1193 vcpu->arch.dtl.update_pending)
1194 vcpus_to_update[need_vpa_update++] = vcpu;
1198 * Initialize *vc, in particular vc->vcore_state, so we can
1199 * drop the vcore lock if necessary.
1203 vc->entry_exit_count = 0;
1204 vc->vcore_state = VCORE_STARTING;
1206 vc->napping_threads = 0;
1209 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1210 * which can't be called with any spinlocks held.
1212 if (need_vpa_update) {
1213 spin_unlock(&vc->lock);
1214 for (i = 0; i < need_vpa_update; ++i)
1215 kvmppc_update_vpas(vcpus_to_update[i]);
1216 spin_lock(&vc->lock);
1220 * Assign physical thread IDs, first to non-ceded vcpus
1221 * and then to ceded ones.
1225 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1226 if (!vcpu->arch.ceded) {
1229 vcpu->arch.ptid = ptid++;
1233 goto out; /* nothing to run; should never happen */
1234 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1235 if (vcpu->arch.ceded)
1236 vcpu->arch.ptid = ptid++;
1239 * Make sure we are running on thread 0, and that
1240 * secondary threads are offline.
1242 if (threads_per_core > 1 && !on_primary_thread()) {
1243 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1244 vcpu->arch.ret = -EBUSY;
1248 vc->pcpu = smp_processor_id();
1249 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1250 kvmppc_start_thread(vcpu);
1251 kvmppc_create_dtl_entry(vcpu, vc);
1254 vc->vcore_state = VCORE_RUNNING;
1256 spin_unlock(&vc->lock);
1260 srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1262 __kvmppc_vcore_entry(NULL, vcpu0);
1264 spin_lock(&vc->lock);
1265 /* disable sending of IPIs on virtual external irqs */
1266 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1268 /* wait for secondary threads to finish writing their state to memory */
1269 if (vc->nap_count < vc->n_woken)
1270 kvmppc_wait_for_nap(vc);
1271 for (i = 0; i < threads_per_core; ++i)
1272 kvmppc_release_hwthread(vc->pcpu + i);
1273 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1274 vc->vcore_state = VCORE_EXITING;
1275 spin_unlock(&vc->lock);
1277 srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1279 /* make sure updates to secondary vcpu structs are visible now */
1286 spin_lock(&vc->lock);
1288 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1289 /* cancel pending dec exception if dec is positive */
1290 if (now < vcpu->arch.dec_expires &&
1291 kvmppc_core_pending_dec(vcpu))
1292 kvmppc_core_dequeue_dec(vcpu);
1295 if (vcpu->arch.trap)
1296 ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1297 vcpu->arch.run_task);
1299 vcpu->arch.ret = ret;
1300 vcpu->arch.trap = 0;
1302 if (vcpu->arch.ceded) {
1303 if (ret != RESUME_GUEST)
1304 kvmppc_end_cede(vcpu);
1306 kvmppc_set_timer(vcpu);
1311 vc->vcore_state = VCORE_INACTIVE;
1312 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1314 if (vcpu->arch.ret != RESUME_GUEST) {
1315 kvmppc_remove_runnable(vc, vcpu);
1316 wake_up(&vcpu->arch.cpu_run);
1322 * Wait for some other vcpu thread to execute us, and
1323 * wake us up when we need to handle something in the host.
1325 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1329 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1330 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1332 finish_wait(&vcpu->arch.cpu_run, &wait);
1336 * All the vcpus in this vcore are idle, so wait for a decrementer
1337 * or external interrupt to one of the vcpus. vc->lock is held.
1339 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1343 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1344 vc->vcore_state = VCORE_SLEEPING;
1345 spin_unlock(&vc->lock);
1347 finish_wait(&vc->wq, &wait);
1348 spin_lock(&vc->lock);
1349 vc->vcore_state = VCORE_INACTIVE;
1352 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1355 struct kvmppc_vcore *vc;
1356 struct kvm_vcpu *v, *vn;
1358 kvm_run->exit_reason = 0;
1359 vcpu->arch.ret = RESUME_GUEST;
1360 vcpu->arch.trap = 0;
1361 kvmppc_update_vpas(vcpu);
1364 * Synchronize with other threads in this virtual core
1366 vc = vcpu->arch.vcore;
1367 spin_lock(&vc->lock);
1368 vcpu->arch.ceded = 0;
1369 vcpu->arch.run_task = current;
1370 vcpu->arch.kvm_run = kvm_run;
1371 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1372 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1373 vcpu->arch.busy_preempt = TB_NIL;
1374 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1378 * This happens the first time this is called for a vcpu.
1379 * If the vcore is already running, we may be able to start
1380 * this thread straight away and have it join in.
1382 if (!signal_pending(current)) {
1383 if (vc->vcore_state == VCORE_RUNNING &&
1384 VCORE_EXIT_COUNT(vc) == 0) {
1385 vcpu->arch.ptid = vc->n_runnable - 1;
1386 kvmppc_create_dtl_entry(vcpu, vc);
1387 kvmppc_start_thread(vcpu);
1388 } else if (vc->vcore_state == VCORE_SLEEPING) {
1394 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1395 !signal_pending(current)) {
1396 if (vc->vcore_state != VCORE_INACTIVE) {
1397 spin_unlock(&vc->lock);
1398 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1399 spin_lock(&vc->lock);
1402 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1404 kvmppc_core_prepare_to_enter(v);
1405 if (signal_pending(v->arch.run_task)) {
1406 kvmppc_remove_runnable(vc, v);
1407 v->stat.signal_exits++;
1408 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1409 v->arch.ret = -EINTR;
1410 wake_up(&v->arch.cpu_run);
1413 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1417 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1418 if (!v->arch.pending_exceptions)
1419 n_ceded += v->arch.ceded;
1423 if (n_ceded == vc->n_runnable)
1424 kvmppc_vcore_blocked(vc);
1426 kvmppc_run_core(vc);
1430 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1431 (vc->vcore_state == VCORE_RUNNING ||
1432 vc->vcore_state == VCORE_EXITING)) {
1433 spin_unlock(&vc->lock);
1434 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1435 spin_lock(&vc->lock);
1438 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1439 kvmppc_remove_runnable(vc, vcpu);
1440 vcpu->stat.signal_exits++;
1441 kvm_run->exit_reason = KVM_EXIT_INTR;
1442 vcpu->arch.ret = -EINTR;
1445 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1446 /* Wake up some vcpu to run the core */
1447 v = list_first_entry(&vc->runnable_threads,
1448 struct kvm_vcpu, arch.run_list);
1449 wake_up(&v->arch.cpu_run);
1452 spin_unlock(&vc->lock);
1453 return vcpu->arch.ret;
1456 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1461 if (!vcpu->arch.sane) {
1462 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1466 kvmppc_core_prepare_to_enter(vcpu);
1468 /* No need to go into the guest when all we'll do is come back out */
1469 if (signal_pending(current)) {
1470 run->exit_reason = KVM_EXIT_INTR;
1474 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1475 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1478 /* On the first time here, set up HTAB and VRMA or RMA */
1479 if (!vcpu->kvm->arch.rma_setup_done) {
1480 r = kvmppc_hv_setup_htab_rma(vcpu);
1485 flush_fp_to_thread(current);
1486 flush_altivec_to_thread(current);
1487 flush_vsx_to_thread(current);
1488 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1489 vcpu->arch.pgdir = current->mm->pgd;
1490 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1493 r = kvmppc_run_vcpu(run, vcpu);
1495 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1496 !(vcpu->arch.shregs.msr & MSR_PR)) {
1497 r = kvmppc_pseries_do_hcall(vcpu);
1498 kvmppc_core_prepare_to_enter(vcpu);
1499 } else if (r == RESUME_PAGE_FAULT) {
1500 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1501 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1502 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1503 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1505 } while (r == RESUME_GUEST);
1508 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1509 atomic_dec(&vcpu->kvm->arch.vcpus_running);
1514 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1515 Assumes POWER7 or PPC970. */
1516 static inline int lpcr_rmls(unsigned long rma_size)
1519 case 32ul << 20: /* 32 MB */
1520 if (cpu_has_feature(CPU_FTR_ARCH_206))
1521 return 8; /* only supported on POWER7 */
1523 case 64ul << 20: /* 64 MB */
1525 case 128ul << 20: /* 128 MB */
1527 case 256ul << 20: /* 256 MB */
1529 case 1ul << 30: /* 1 GB */
1531 case 16ul << 30: /* 16 GB */
1533 case 256ul << 30: /* 256 GB */
1540 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1543 struct kvm_rma_info *ri = vma->vm_file->private_data;
1545 if (vmf->pgoff >= kvm_rma_pages)
1546 return VM_FAULT_SIGBUS;
1548 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1554 static const struct vm_operations_struct kvm_rma_vm_ops = {
1555 .fault = kvm_rma_fault,
1558 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1560 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1561 vma->vm_ops = &kvm_rma_vm_ops;
1565 static int kvm_rma_release(struct inode *inode, struct file *filp)
1567 struct kvm_rma_info *ri = filp->private_data;
1569 kvm_release_rma(ri);
1573 static const struct file_operations kvm_rma_fops = {
1574 .mmap = kvm_rma_mmap,
1575 .release = kvm_rma_release,
1578 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1581 struct kvm_rma_info *ri;
1583 * Only do this on PPC970 in HV mode
1585 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1586 !cpu_has_feature(CPU_FTR_ARCH_201))
1592 ri = kvm_alloc_rma();
1596 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1598 kvm_release_rma(ri);
1600 ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1604 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1607 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1611 (*sps)->page_shift = def->shift;
1612 (*sps)->slb_enc = def->sllp;
1613 (*sps)->enc[0].page_shift = def->shift;
1615 * Only return base page encoding. We don't want to return
1616 * all the supporting pte_enc, because our H_ENTER doesn't
1617 * support MPSS yet. Once they do, we can start passing all
1618 * support pte_enc here
1620 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1624 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1626 struct kvm_ppc_one_seg_page_size *sps;
1628 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1629 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1630 info->flags |= KVM_PPC_1T_SEGMENTS;
1631 info->slb_size = mmu_slb_size;
1633 /* We only support these sizes for now, and no muti-size segments */
1634 sps = &info->sps[0];
1635 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1636 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1637 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1643 * Get (and clear) the dirty memory log for a memory slot.
1645 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1647 struct kvm_memory_slot *memslot;
1651 mutex_lock(&kvm->slots_lock);
1654 if (log->slot >= KVM_USER_MEM_SLOTS)
1657 memslot = id_to_memslot(kvm->memslots, log->slot);
1659 if (!memslot->dirty_bitmap)
1662 n = kvm_dirty_bitmap_bytes(memslot);
1663 memset(memslot->dirty_bitmap, 0, n);
1665 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1670 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1675 mutex_unlock(&kvm->slots_lock);
1679 static void unpin_slot(struct kvm_memory_slot *memslot)
1681 unsigned long *physp;
1682 unsigned long j, npages, pfn;
1685 physp = memslot->arch.slot_phys;
1686 npages = memslot->npages;
1689 for (j = 0; j < npages; j++) {
1690 if (!(physp[j] & KVMPPC_GOT_PAGE))
1692 pfn = physp[j] >> PAGE_SHIFT;
1693 page = pfn_to_page(pfn);
1699 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1700 struct kvm_memory_slot *dont)
1702 if (!dont || free->arch.rmap != dont->arch.rmap) {
1703 vfree(free->arch.rmap);
1704 free->arch.rmap = NULL;
1706 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1708 vfree(free->arch.slot_phys);
1709 free->arch.slot_phys = NULL;
1713 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1714 unsigned long npages)
1716 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1717 if (!slot->arch.rmap)
1719 slot->arch.slot_phys = NULL;
1724 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1725 struct kvm_memory_slot *memslot,
1726 struct kvm_userspace_memory_region *mem)
1728 unsigned long *phys;
1730 /* Allocate a slot_phys array if needed */
1731 phys = memslot->arch.slot_phys;
1732 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1733 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1736 memslot->arch.slot_phys = phys;
1742 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1743 struct kvm_userspace_memory_region *mem,
1744 const struct kvm_memory_slot *old)
1746 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1747 struct kvm_memory_slot *memslot;
1749 if (npages && old->npages) {
1751 * If modifying a memslot, reset all the rmap dirty bits.
1752 * If this is a new memslot, we don't need to do anything
1753 * since the rmap array starts out as all zeroes,
1754 * i.e. no pages are dirty.
1756 memslot = id_to_memslot(kvm->memslots, mem->slot);
1757 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1761 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1764 struct kvm *kvm = vcpu->kvm;
1765 struct kvm_rma_info *ri = NULL;
1767 struct kvm_memory_slot *memslot;
1768 struct vm_area_struct *vma;
1769 unsigned long lpcr, senc;
1770 unsigned long psize, porder;
1771 unsigned long rma_size;
1773 unsigned long *physp;
1774 unsigned long i, npages;
1777 mutex_lock(&kvm->lock);
1778 if (kvm->arch.rma_setup_done)
1779 goto out; /* another vcpu beat us to it */
1781 /* Allocate hashed page table (if not done already) and reset it */
1782 if (!kvm->arch.hpt_virt) {
1783 err = kvmppc_alloc_hpt(kvm, NULL);
1785 pr_err("KVM: Couldn't alloc HPT\n");
1790 /* Look up the memslot for guest physical address 0 */
1791 srcu_idx = srcu_read_lock(&kvm->srcu);
1792 memslot = gfn_to_memslot(kvm, 0);
1794 /* We must have some memory at 0 by now */
1796 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1799 /* Look up the VMA for the start of this memory slot */
1800 hva = memslot->userspace_addr;
1801 down_read(¤t->mm->mmap_sem);
1802 vma = find_vma(current->mm, hva);
1803 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1806 psize = vma_kernel_pagesize(vma);
1807 porder = __ilog2(psize);
1809 /* Is this one of our preallocated RMAs? */
1810 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1811 hva == vma->vm_start)
1812 ri = vma->vm_file->private_data;
1814 up_read(¤t->mm->mmap_sem);
1817 /* On POWER7, use VRMA; on PPC970, give up */
1819 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1820 pr_err("KVM: CPU requires an RMO\n");
1824 /* We can handle 4k, 64k or 16M pages in the VRMA */
1826 if (!(psize == 0x1000 || psize == 0x10000 ||
1827 psize == 0x1000000))
1830 /* Update VRMASD field in the LPCR */
1831 senc = slb_pgsize_encoding(psize);
1832 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1833 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1834 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1835 lpcr |= senc << (LPCR_VRMASD_SH - 4);
1836 kvm->arch.lpcr = lpcr;
1838 /* Create HPTEs in the hash page table for the VRMA */
1839 kvmppc_map_vrma(vcpu, memslot, porder);
1842 /* Set up to use an RMO region */
1843 rma_size = kvm_rma_pages;
1844 if (rma_size > memslot->npages)
1845 rma_size = memslot->npages;
1846 rma_size <<= PAGE_SHIFT;
1847 rmls = lpcr_rmls(rma_size);
1849 if ((long)rmls < 0) {
1850 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1853 atomic_inc(&ri->use_count);
1856 /* Update LPCR and RMOR */
1857 lpcr = kvm->arch.lpcr;
1858 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1859 /* PPC970; insert RMLS value (split field) in HID4 */
1860 lpcr &= ~((1ul << HID4_RMLS0_SH) |
1861 (3ul << HID4_RMLS2_SH));
1862 lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1863 ((rmls & 3) << HID4_RMLS2_SH);
1864 /* RMOR is also in HID4 */
1865 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1869 lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1870 lpcr |= rmls << LPCR_RMLS_SH;
1871 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
1873 kvm->arch.lpcr = lpcr;
1874 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1875 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1877 /* Initialize phys addrs of pages in RMO */
1878 npages = kvm_rma_pages;
1879 porder = __ilog2(npages);
1880 physp = memslot->arch.slot_phys;
1882 if (npages > memslot->npages)
1883 npages = memslot->npages;
1884 spin_lock(&kvm->arch.slot_phys_lock);
1885 for (i = 0; i < npages; ++i)
1886 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1888 spin_unlock(&kvm->arch.slot_phys_lock);
1892 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1894 kvm->arch.rma_setup_done = 1;
1897 srcu_read_unlock(&kvm->srcu, srcu_idx);
1899 mutex_unlock(&kvm->lock);
1903 up_read(¤t->mm->mmap_sem);
1907 int kvmppc_core_init_vm(struct kvm *kvm)
1909 unsigned long lpcr, lpid;
1911 /* Allocate the guest's logical partition ID */
1913 lpid = kvmppc_alloc_lpid();
1916 kvm->arch.lpid = lpid;
1919 * Since we don't flush the TLB when tearing down a VM,
1920 * and this lpid might have previously been used,
1921 * make sure we flush on each core before running the new VM.
1923 cpumask_setall(&kvm->arch.need_tlb_flush);
1925 INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1926 INIT_LIST_HEAD(&kvm->arch.rtas_tokens);
1928 kvm->arch.rma = NULL;
1930 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1932 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1933 /* PPC970; HID4 is effectively the LPCR */
1934 kvm->arch.host_lpid = 0;
1935 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1936 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1937 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1938 ((lpid & 0xf) << HID4_LPID5_SH);
1940 /* POWER7; init LPCR for virtual RMA mode */
1941 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1942 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1943 lpcr &= LPCR_PECE | LPCR_LPES;
1944 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1945 LPCR_VPM0 | LPCR_VPM1;
1946 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1947 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1949 kvm->arch.lpcr = lpcr;
1951 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1952 spin_lock_init(&kvm->arch.slot_phys_lock);
1955 * Don't allow secondary CPU threads to come online
1956 * while any KVM VMs exist.
1958 inhibit_secondary_onlining();
1963 void kvmppc_core_destroy_vm(struct kvm *kvm)
1965 uninhibit_secondary_onlining();
1967 if (kvm->arch.rma) {
1968 kvm_release_rma(kvm->arch.rma);
1969 kvm->arch.rma = NULL;
1972 kvmppc_rtas_tokens_free(kvm);
1974 kvmppc_free_hpt(kvm);
1975 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1978 /* These are stubs for now */
1979 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1983 /* We don't need to emulate any privileged instructions or dcbz */
1984 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1985 unsigned int inst, int *advance)
1987 return EMULATE_FAIL;
1990 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
1992 return EMULATE_FAIL;
1995 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
1997 return EMULATE_FAIL;
2000 static int kvmppc_book3s_hv_init(void)
2004 r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2009 r = kvmppc_mmu_hv_init();
2014 static void kvmppc_book3s_hv_exit(void)
2019 module_init(kvmppc_book3s_hv_init);
2020 module_exit(kvmppc_book3s_hv_exit);