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>
34 #include <linux/miscdevice.h>
37 #include <asm/cputable.h>
38 #include <asm/cacheflush.h>
39 #include <asm/tlbflush.h>
40 #include <asm/uaccess.h>
42 #include <asm/kvm_ppc.h>
43 #include <asm/kvm_book3s.h>
44 #include <asm/mmu_context.h>
45 #include <asm/lppaca.h>
46 #include <asm/processor.h>
47 #include <asm/cputhreads.h>
49 #include <asm/hvcall.h>
50 #include <asm/switch_to.h>
52 #include <linux/gfp.h>
53 #include <linux/vmalloc.h>
54 #include <linux/highmem.h>
55 #include <linux/hugetlb.h>
56 #include <linux/module.h>
60 /* #define EXIT_DEBUG */
61 /* #define EXIT_DEBUG_SIMPLE */
62 /* #define EXIT_DEBUG_INT */
64 /* Used to indicate that a guest page fault needs to be handled */
65 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
67 /* Used as a "null" value for timebase values */
68 #define TB_NIL (~(u64)0)
70 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
71 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
73 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
77 wait_queue_head_t *wqp;
79 wqp = kvm_arch_vcpu_wq(vcpu);
80 if (waitqueue_active(wqp)) {
81 wake_up_interruptible(wqp);
82 ++vcpu->stat.halt_wakeup;
87 /* CPU points to the first thread of the core */
88 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
89 #ifdef CONFIG_PPC_ICP_NATIVE
90 int real_cpu = cpu + vcpu->arch.ptid;
91 if (paca[real_cpu].kvm_hstate.xics_phys)
92 xics_wake_cpu(real_cpu);
96 smp_send_reschedule(cpu);
102 * We use the vcpu_load/put functions to measure stolen time.
103 * Stolen time is counted as time when either the vcpu is able to
104 * run as part of a virtual core, but the task running the vcore
105 * is preempted or sleeping, or when the vcpu needs something done
106 * in the kernel by the task running the vcpu, but that task is
107 * preempted or sleeping. Those two things have to be counted
108 * separately, since one of the vcpu tasks will take on the job
109 * of running the core, and the other vcpu tasks in the vcore will
110 * sleep waiting for it to do that, but that sleep shouldn't count
113 * Hence we accumulate stolen time when the vcpu can run as part of
114 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
115 * needs its task to do other things in the kernel (for example,
116 * service a page fault) in busy_stolen. We don't accumulate
117 * stolen time for a vcore when it is inactive, or for a vcpu
118 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
119 * a misnomer; it means that the vcpu task is not executing in
120 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
121 * the kernel. We don't have any way of dividing up that time
122 * between time that the vcpu is genuinely stopped, time that
123 * the task is actively working on behalf of the vcpu, and time
124 * that the task is preempted, so we don't count any of it as
127 * Updates to busy_stolen are protected by arch.tbacct_lock;
128 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
129 * of the vcpu that has taken responsibility for running the vcore
130 * (i.e. vc->runner). The stolen times are measured in units of
131 * timebase ticks. (Note that the != TB_NIL checks below are
132 * purely defensive; they should never fail.)
135 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
137 struct kvmppc_vcore *vc = vcpu->arch.vcore;
140 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
141 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
142 vc->preempt_tb != TB_NIL) {
143 vc->stolen_tb += mftb() - vc->preempt_tb;
144 vc->preempt_tb = TB_NIL;
146 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
147 vcpu->arch.busy_preempt != TB_NIL) {
148 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
149 vcpu->arch.busy_preempt = TB_NIL;
151 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
154 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
156 struct kvmppc_vcore *vc = vcpu->arch.vcore;
159 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
160 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
161 vc->preempt_tb = mftb();
162 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
163 vcpu->arch.busy_preempt = mftb();
164 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
167 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
169 vcpu->arch.shregs.msr = msr;
170 kvmppc_end_cede(vcpu);
173 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
175 vcpu->arch.pvr = pvr;
178 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
180 unsigned long pcr = 0;
181 struct kvmppc_vcore *vc = vcpu->arch.vcore;
184 if (!cpu_has_feature(CPU_FTR_ARCH_206))
185 return -EINVAL; /* 970 has no compat mode support */
187 switch (arch_compat) {
190 * If an arch bit is set in PCR, all the defined
191 * higher-order arch bits also have to be set.
193 pcr = PCR_ARCH_206 | PCR_ARCH_205;
205 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
206 /* POWER7 can't emulate POWER8 */
207 if (!(pcr & PCR_ARCH_206))
209 pcr &= ~PCR_ARCH_206;
213 spin_lock(&vc->lock);
214 vc->arch_compat = arch_compat;
216 spin_unlock(&vc->lock);
221 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
225 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
226 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
227 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
228 for (r = 0; r < 16; ++r)
229 pr_err("r%2d = %.16lx r%d = %.16lx\n",
230 r, kvmppc_get_gpr(vcpu, r),
231 r+16, kvmppc_get_gpr(vcpu, r+16));
232 pr_err("ctr = %.16lx lr = %.16lx\n",
233 vcpu->arch.ctr, vcpu->arch.lr);
234 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
235 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
236 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
237 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
238 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
239 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
240 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
241 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
242 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
243 pr_err("fault dar = %.16lx dsisr = %.8x\n",
244 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
245 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
246 for (r = 0; r < vcpu->arch.slb_max; ++r)
247 pr_err(" ESID = %.16llx VSID = %.16llx\n",
248 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
249 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
250 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
251 vcpu->arch.last_inst);
254 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
257 struct kvm_vcpu *v, *ret = NULL;
259 mutex_lock(&kvm->lock);
260 kvm_for_each_vcpu(r, v, kvm) {
261 if (v->vcpu_id == id) {
266 mutex_unlock(&kvm->lock);
270 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
272 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
273 vpa->yield_count = 1;
276 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
277 unsigned long addr, unsigned long len)
279 /* check address is cacheline aligned */
280 if (addr & (L1_CACHE_BYTES - 1))
282 spin_lock(&vcpu->arch.vpa_update_lock);
283 if (v->next_gpa != addr || v->len != len) {
285 v->len = addr ? len : 0;
286 v->update_pending = 1;
288 spin_unlock(&vcpu->arch.vpa_update_lock);
292 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
301 static int vpa_is_registered(struct kvmppc_vpa *vpap)
303 if (vpap->update_pending)
304 return vpap->next_gpa != 0;
305 return vpap->pinned_addr != NULL;
308 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
310 unsigned long vcpuid, unsigned long vpa)
312 struct kvm *kvm = vcpu->kvm;
313 unsigned long len, nb;
315 struct kvm_vcpu *tvcpu;
318 struct kvmppc_vpa *vpap;
320 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
324 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
325 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
326 subfunc == H_VPA_REG_SLB) {
327 /* Registering new area - address must be cache-line aligned */
328 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
331 /* convert logical addr to kernel addr and read length */
332 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
335 if (subfunc == H_VPA_REG_VPA)
336 len = ((struct reg_vpa *)va)->length.hword;
338 len = ((struct reg_vpa *)va)->length.word;
339 kvmppc_unpin_guest_page(kvm, va, vpa, false);
342 if (len > nb || len < sizeof(struct reg_vpa))
351 spin_lock(&tvcpu->arch.vpa_update_lock);
354 case H_VPA_REG_VPA: /* register VPA */
355 if (len < sizeof(struct lppaca))
357 vpap = &tvcpu->arch.vpa;
361 case H_VPA_REG_DTL: /* register DTL */
362 if (len < sizeof(struct dtl_entry))
364 len -= len % sizeof(struct dtl_entry);
366 /* Check that they have previously registered a VPA */
368 if (!vpa_is_registered(&tvcpu->arch.vpa))
371 vpap = &tvcpu->arch.dtl;
375 case H_VPA_REG_SLB: /* register SLB shadow buffer */
376 /* Check that they have previously registered a VPA */
378 if (!vpa_is_registered(&tvcpu->arch.vpa))
381 vpap = &tvcpu->arch.slb_shadow;
385 case H_VPA_DEREG_VPA: /* deregister VPA */
386 /* Check they don't still have a DTL or SLB buf registered */
388 if (vpa_is_registered(&tvcpu->arch.dtl) ||
389 vpa_is_registered(&tvcpu->arch.slb_shadow))
392 vpap = &tvcpu->arch.vpa;
396 case H_VPA_DEREG_DTL: /* deregister DTL */
397 vpap = &tvcpu->arch.dtl;
401 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
402 vpap = &tvcpu->arch.slb_shadow;
408 vpap->next_gpa = vpa;
410 vpap->update_pending = 1;
413 spin_unlock(&tvcpu->arch.vpa_update_lock);
418 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
420 struct kvm *kvm = vcpu->kvm;
426 * We need to pin the page pointed to by vpap->next_gpa,
427 * but we can't call kvmppc_pin_guest_page under the lock
428 * as it does get_user_pages() and down_read(). So we
429 * have to drop the lock, pin the page, then get the lock
430 * again and check that a new area didn't get registered
434 gpa = vpap->next_gpa;
435 spin_unlock(&vcpu->arch.vpa_update_lock);
439 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
440 spin_lock(&vcpu->arch.vpa_update_lock);
441 if (gpa == vpap->next_gpa)
443 /* sigh... unpin that one and try again */
445 kvmppc_unpin_guest_page(kvm, va, gpa, false);
448 vpap->update_pending = 0;
449 if (va && nb < vpap->len) {
451 * If it's now too short, it must be that userspace
452 * has changed the mappings underlying guest memory,
453 * so unregister the region.
455 kvmppc_unpin_guest_page(kvm, va, gpa, false);
458 if (vpap->pinned_addr)
459 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
462 vpap->pinned_addr = va;
465 vpap->pinned_end = va + vpap->len;
468 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
470 if (!(vcpu->arch.vpa.update_pending ||
471 vcpu->arch.slb_shadow.update_pending ||
472 vcpu->arch.dtl.update_pending))
475 spin_lock(&vcpu->arch.vpa_update_lock);
476 if (vcpu->arch.vpa.update_pending) {
477 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
478 if (vcpu->arch.vpa.pinned_addr)
479 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
481 if (vcpu->arch.dtl.update_pending) {
482 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
483 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
484 vcpu->arch.dtl_index = 0;
486 if (vcpu->arch.slb_shadow.update_pending)
487 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
488 spin_unlock(&vcpu->arch.vpa_update_lock);
492 * Return the accumulated stolen time for the vcore up until `now'.
493 * The caller should hold the vcore lock.
495 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
500 * If we are the task running the vcore, then since we hold
501 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
502 * can't be updated, so we don't need the tbacct_lock.
503 * If the vcore is inactive, it can't become active (since we
504 * hold the vcore lock), so the vcpu load/put functions won't
505 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
507 if (vc->vcore_state != VCORE_INACTIVE &&
508 vc->runner->arch.run_task != current) {
509 spin_lock_irq(&vc->runner->arch.tbacct_lock);
511 if (vc->preempt_tb != TB_NIL)
512 p += now - vc->preempt_tb;
513 spin_unlock_irq(&vc->runner->arch.tbacct_lock);
520 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
521 struct kvmppc_vcore *vc)
523 struct dtl_entry *dt;
525 unsigned long stolen;
526 unsigned long core_stolen;
529 dt = vcpu->arch.dtl_ptr;
530 vpa = vcpu->arch.vpa.pinned_addr;
532 core_stolen = vcore_stolen_time(vc, now);
533 stolen = core_stolen - vcpu->arch.stolen_logged;
534 vcpu->arch.stolen_logged = core_stolen;
535 spin_lock_irq(&vcpu->arch.tbacct_lock);
536 stolen += vcpu->arch.busy_stolen;
537 vcpu->arch.busy_stolen = 0;
538 spin_unlock_irq(&vcpu->arch.tbacct_lock);
541 memset(dt, 0, sizeof(struct dtl_entry));
542 dt->dispatch_reason = 7;
543 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
544 dt->timebase = now + vc->tb_offset;
545 dt->enqueue_to_dispatch_time = stolen;
546 dt->srr0 = kvmppc_get_pc(vcpu);
547 dt->srr1 = vcpu->arch.shregs.msr;
549 if (dt == vcpu->arch.dtl.pinned_end)
550 dt = vcpu->arch.dtl.pinned_addr;
551 vcpu->arch.dtl_ptr = dt;
552 /* order writing *dt vs. writing vpa->dtl_idx */
554 vpa->dtl_idx = ++vcpu->arch.dtl_index;
555 vcpu->arch.dtl.dirty = true;
558 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
560 unsigned long req = kvmppc_get_gpr(vcpu, 3);
561 unsigned long target, ret = H_SUCCESS;
562 struct kvm_vcpu *tvcpu;
567 idx = srcu_read_lock(&vcpu->kvm->srcu);
568 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
569 kvmppc_get_gpr(vcpu, 5),
570 kvmppc_get_gpr(vcpu, 6),
571 kvmppc_get_gpr(vcpu, 7));
572 srcu_read_unlock(&vcpu->kvm->srcu, idx);
577 target = kvmppc_get_gpr(vcpu, 4);
578 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
583 tvcpu->arch.prodded = 1;
585 if (vcpu->arch.ceded) {
586 if (waitqueue_active(&vcpu->wq)) {
587 wake_up_interruptible(&vcpu->wq);
588 vcpu->stat.halt_wakeup++;
593 target = kvmppc_get_gpr(vcpu, 4);
596 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
601 kvm_vcpu_yield_to(tvcpu);
604 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
605 kvmppc_get_gpr(vcpu, 5),
606 kvmppc_get_gpr(vcpu, 6));
609 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
612 idx = srcu_read_lock(&vcpu->kvm->srcu);
613 rc = kvmppc_rtas_hcall(vcpu);
614 srcu_read_unlock(&vcpu->kvm->srcu, idx);
621 /* Send the error out to userspace via KVM_RUN */
630 if (kvmppc_xics_enabled(vcpu)) {
631 ret = kvmppc_xics_hcall(vcpu, req);
637 kvmppc_set_gpr(vcpu, 3, ret);
638 vcpu->arch.hcall_needed = 0;
642 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
643 struct task_struct *tsk)
647 vcpu->stat.sum_exits++;
649 run->exit_reason = KVM_EXIT_UNKNOWN;
650 run->ready_for_interrupt_injection = 1;
651 switch (vcpu->arch.trap) {
652 /* We're good on these - the host merely wanted to get our attention */
653 case BOOK3S_INTERRUPT_HV_DECREMENTER:
654 vcpu->stat.dec_exits++;
657 case BOOK3S_INTERRUPT_EXTERNAL:
658 case BOOK3S_INTERRUPT_H_DOORBELL:
659 vcpu->stat.ext_intr_exits++;
662 case BOOK3S_INTERRUPT_PERFMON:
665 case BOOK3S_INTERRUPT_MACHINE_CHECK:
667 * Deliver a machine check interrupt to the guest.
668 * We have to do this, even if the host has handled the
669 * machine check, because machine checks use SRR0/1 and
670 * the interrupt might have trashed guest state in them.
672 kvmppc_book3s_queue_irqprio(vcpu,
673 BOOK3S_INTERRUPT_MACHINE_CHECK);
676 case BOOK3S_INTERRUPT_PROGRAM:
680 * Normally program interrupts are delivered directly
681 * to the guest by the hardware, but we can get here
682 * as a result of a hypervisor emulation interrupt
683 * (e40) getting turned into a 700 by BML RTAS.
685 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
686 kvmppc_core_queue_program(vcpu, flags);
690 case BOOK3S_INTERRUPT_SYSCALL:
692 /* hcall - punt to userspace */
695 /* hypercall with MSR_PR has already been handled in rmode,
696 * and never reaches here.
699 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
700 for (i = 0; i < 9; ++i)
701 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
702 run->exit_reason = KVM_EXIT_PAPR_HCALL;
703 vcpu->arch.hcall_needed = 1;
708 * We get these next two if the guest accesses a page which it thinks
709 * it has mapped but which is not actually present, either because
710 * it is for an emulated I/O device or because the corresonding
711 * host page has been paged out. Any other HDSI/HISI interrupts
712 * have been handled already.
714 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
715 r = RESUME_PAGE_FAULT;
717 case BOOK3S_INTERRUPT_H_INST_STORAGE:
718 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
719 vcpu->arch.fault_dsisr = 0;
720 r = RESUME_PAGE_FAULT;
723 * This occurs if the guest executes an illegal instruction.
724 * We just generate a program interrupt to the guest, since
725 * we don't emulate any guest instructions at this stage.
727 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
728 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
732 * This occurs if the guest (kernel or userspace), does something that
733 * is prohibited by HFSCR. We just generate a program interrupt to
736 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
737 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
741 kvmppc_dump_regs(vcpu);
742 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
743 vcpu->arch.trap, kvmppc_get_pc(vcpu),
744 vcpu->arch.shregs.msr);
745 run->hw.hardware_exit_reason = vcpu->arch.trap;
753 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
754 struct kvm_sregs *sregs)
758 memset(sregs, 0, sizeof(struct kvm_sregs));
759 sregs->pvr = vcpu->arch.pvr;
760 for (i = 0; i < vcpu->arch.slb_max; i++) {
761 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
762 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
768 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
769 struct kvm_sregs *sregs)
773 kvmppc_set_pvr_hv(vcpu, sregs->pvr);
776 for (i = 0; i < vcpu->arch.slb_nr; i++) {
777 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
778 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
779 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
783 vcpu->arch.slb_max = j;
788 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
790 struct kvmppc_vcore *vc = vcpu->arch.vcore;
793 spin_lock(&vc->lock);
795 * If ILE (interrupt little-endian) has changed, update the
796 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
798 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
799 struct kvm *kvm = vcpu->kvm;
800 struct kvm_vcpu *vcpu;
803 mutex_lock(&kvm->lock);
804 kvm_for_each_vcpu(i, vcpu, kvm) {
805 if (vcpu->arch.vcore != vc)
807 if (new_lpcr & LPCR_ILE)
808 vcpu->arch.intr_msr |= MSR_LE;
810 vcpu->arch.intr_msr &= ~MSR_LE;
812 mutex_unlock(&kvm->lock);
816 * Userspace can only modify DPFD (default prefetch depth),
817 * ILE (interrupt little-endian) and TC (translation control).
818 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
820 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
821 if (cpu_has_feature(CPU_FTR_ARCH_207S))
823 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
824 spin_unlock(&vc->lock);
827 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
828 union kvmppc_one_reg *val)
834 case KVM_REG_PPC_HIOR:
835 *val = get_reg_val(id, 0);
837 case KVM_REG_PPC_DABR:
838 *val = get_reg_val(id, vcpu->arch.dabr);
840 case KVM_REG_PPC_DABRX:
841 *val = get_reg_val(id, vcpu->arch.dabrx);
843 case KVM_REG_PPC_DSCR:
844 *val = get_reg_val(id, vcpu->arch.dscr);
846 case KVM_REG_PPC_PURR:
847 *val = get_reg_val(id, vcpu->arch.purr);
849 case KVM_REG_PPC_SPURR:
850 *val = get_reg_val(id, vcpu->arch.spurr);
852 case KVM_REG_PPC_AMR:
853 *val = get_reg_val(id, vcpu->arch.amr);
855 case KVM_REG_PPC_UAMOR:
856 *val = get_reg_val(id, vcpu->arch.uamor);
858 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
859 i = id - KVM_REG_PPC_MMCR0;
860 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
862 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
863 i = id - KVM_REG_PPC_PMC1;
864 *val = get_reg_val(id, vcpu->arch.pmc[i]);
866 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
867 i = id - KVM_REG_PPC_SPMC1;
868 *val = get_reg_val(id, vcpu->arch.spmc[i]);
870 case KVM_REG_PPC_SIAR:
871 *val = get_reg_val(id, vcpu->arch.siar);
873 case KVM_REG_PPC_SDAR:
874 *val = get_reg_val(id, vcpu->arch.sdar);
876 case KVM_REG_PPC_SIER:
877 *val = get_reg_val(id, vcpu->arch.sier);
879 case KVM_REG_PPC_IAMR:
880 *val = get_reg_val(id, vcpu->arch.iamr);
882 case KVM_REG_PPC_PSPB:
883 *val = get_reg_val(id, vcpu->arch.pspb);
885 case KVM_REG_PPC_DPDES:
886 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
888 case KVM_REG_PPC_DAWR:
889 *val = get_reg_val(id, vcpu->arch.dawr);
891 case KVM_REG_PPC_DAWRX:
892 *val = get_reg_val(id, vcpu->arch.dawrx);
894 case KVM_REG_PPC_CIABR:
895 *val = get_reg_val(id, vcpu->arch.ciabr);
898 *val = get_reg_val(id, vcpu->arch.ic);
900 case KVM_REG_PPC_VTB:
901 *val = get_reg_val(id, vcpu->arch.vtb);
903 case KVM_REG_PPC_CSIGR:
904 *val = get_reg_val(id, vcpu->arch.csigr);
906 case KVM_REG_PPC_TACR:
907 *val = get_reg_val(id, vcpu->arch.tacr);
909 case KVM_REG_PPC_TCSCR:
910 *val = get_reg_val(id, vcpu->arch.tcscr);
912 case KVM_REG_PPC_PID:
913 *val = get_reg_val(id, vcpu->arch.pid);
915 case KVM_REG_PPC_ACOP:
916 *val = get_reg_val(id, vcpu->arch.acop);
918 case KVM_REG_PPC_WORT:
919 *val = get_reg_val(id, vcpu->arch.wort);
921 case KVM_REG_PPC_VPA_ADDR:
922 spin_lock(&vcpu->arch.vpa_update_lock);
923 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
924 spin_unlock(&vcpu->arch.vpa_update_lock);
926 case KVM_REG_PPC_VPA_SLB:
927 spin_lock(&vcpu->arch.vpa_update_lock);
928 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
929 val->vpaval.length = vcpu->arch.slb_shadow.len;
930 spin_unlock(&vcpu->arch.vpa_update_lock);
932 case KVM_REG_PPC_VPA_DTL:
933 spin_lock(&vcpu->arch.vpa_update_lock);
934 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
935 val->vpaval.length = vcpu->arch.dtl.len;
936 spin_unlock(&vcpu->arch.vpa_update_lock);
938 case KVM_REG_PPC_TB_OFFSET:
939 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
941 case KVM_REG_PPC_LPCR:
942 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
944 case KVM_REG_PPC_PPR:
945 *val = get_reg_val(id, vcpu->arch.ppr);
947 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
948 case KVM_REG_PPC_TFHAR:
949 *val = get_reg_val(id, vcpu->arch.tfhar);
951 case KVM_REG_PPC_TFIAR:
952 *val = get_reg_val(id, vcpu->arch.tfiar);
954 case KVM_REG_PPC_TEXASR:
955 *val = get_reg_val(id, vcpu->arch.texasr);
957 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
958 i = id - KVM_REG_PPC_TM_GPR0;
959 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
961 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
964 i = id - KVM_REG_PPC_TM_VSR0;
966 for (j = 0; j < TS_FPRWIDTH; j++)
967 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
969 if (cpu_has_feature(CPU_FTR_ALTIVEC))
970 val->vval = vcpu->arch.vr_tm.vr[i-32];
976 case KVM_REG_PPC_TM_CR:
977 *val = get_reg_val(id, vcpu->arch.cr_tm);
979 case KVM_REG_PPC_TM_LR:
980 *val = get_reg_val(id, vcpu->arch.lr_tm);
982 case KVM_REG_PPC_TM_CTR:
983 *val = get_reg_val(id, vcpu->arch.ctr_tm);
985 case KVM_REG_PPC_TM_FPSCR:
986 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
988 case KVM_REG_PPC_TM_AMR:
989 *val = get_reg_val(id, vcpu->arch.amr_tm);
991 case KVM_REG_PPC_TM_PPR:
992 *val = get_reg_val(id, vcpu->arch.ppr_tm);
994 case KVM_REG_PPC_TM_VRSAVE:
995 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
997 case KVM_REG_PPC_TM_VSCR:
998 if (cpu_has_feature(CPU_FTR_ALTIVEC))
999 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1003 case KVM_REG_PPC_TM_DSCR:
1004 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1006 case KVM_REG_PPC_TM_TAR:
1007 *val = get_reg_val(id, vcpu->arch.tar_tm);
1010 case KVM_REG_PPC_ARCH_COMPAT:
1011 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1021 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1022 union kvmppc_one_reg *val)
1026 unsigned long addr, len;
1029 case KVM_REG_PPC_HIOR:
1030 /* Only allow this to be set to zero */
1031 if (set_reg_val(id, *val))
1034 case KVM_REG_PPC_DABR:
1035 vcpu->arch.dabr = set_reg_val(id, *val);
1037 case KVM_REG_PPC_DABRX:
1038 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1040 case KVM_REG_PPC_DSCR:
1041 vcpu->arch.dscr = set_reg_val(id, *val);
1043 case KVM_REG_PPC_PURR:
1044 vcpu->arch.purr = set_reg_val(id, *val);
1046 case KVM_REG_PPC_SPURR:
1047 vcpu->arch.spurr = set_reg_val(id, *val);
1049 case KVM_REG_PPC_AMR:
1050 vcpu->arch.amr = set_reg_val(id, *val);
1052 case KVM_REG_PPC_UAMOR:
1053 vcpu->arch.uamor = set_reg_val(id, *val);
1055 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1056 i = id - KVM_REG_PPC_MMCR0;
1057 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1059 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1060 i = id - KVM_REG_PPC_PMC1;
1061 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1063 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1064 i = id - KVM_REG_PPC_SPMC1;
1065 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1067 case KVM_REG_PPC_SIAR:
1068 vcpu->arch.siar = set_reg_val(id, *val);
1070 case KVM_REG_PPC_SDAR:
1071 vcpu->arch.sdar = set_reg_val(id, *val);
1073 case KVM_REG_PPC_SIER:
1074 vcpu->arch.sier = set_reg_val(id, *val);
1076 case KVM_REG_PPC_IAMR:
1077 vcpu->arch.iamr = set_reg_val(id, *val);
1079 case KVM_REG_PPC_PSPB:
1080 vcpu->arch.pspb = set_reg_val(id, *val);
1082 case KVM_REG_PPC_DPDES:
1083 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1085 case KVM_REG_PPC_DAWR:
1086 vcpu->arch.dawr = set_reg_val(id, *val);
1088 case KVM_REG_PPC_DAWRX:
1089 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1091 case KVM_REG_PPC_CIABR:
1092 vcpu->arch.ciabr = set_reg_val(id, *val);
1093 /* Don't allow setting breakpoints in hypervisor code */
1094 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1095 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1097 case KVM_REG_PPC_IC:
1098 vcpu->arch.ic = set_reg_val(id, *val);
1100 case KVM_REG_PPC_VTB:
1101 vcpu->arch.vtb = set_reg_val(id, *val);
1103 case KVM_REG_PPC_CSIGR:
1104 vcpu->arch.csigr = set_reg_val(id, *val);
1106 case KVM_REG_PPC_TACR:
1107 vcpu->arch.tacr = set_reg_val(id, *val);
1109 case KVM_REG_PPC_TCSCR:
1110 vcpu->arch.tcscr = set_reg_val(id, *val);
1112 case KVM_REG_PPC_PID:
1113 vcpu->arch.pid = set_reg_val(id, *val);
1115 case KVM_REG_PPC_ACOP:
1116 vcpu->arch.acop = set_reg_val(id, *val);
1118 case KVM_REG_PPC_WORT:
1119 vcpu->arch.wort = set_reg_val(id, *val);
1121 case KVM_REG_PPC_VPA_ADDR:
1122 addr = set_reg_val(id, *val);
1124 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1125 vcpu->arch.dtl.next_gpa))
1127 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1129 case KVM_REG_PPC_VPA_SLB:
1130 addr = val->vpaval.addr;
1131 len = val->vpaval.length;
1133 if (addr && !vcpu->arch.vpa.next_gpa)
1135 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1137 case KVM_REG_PPC_VPA_DTL:
1138 addr = val->vpaval.addr;
1139 len = val->vpaval.length;
1141 if (addr && (len < sizeof(struct dtl_entry) ||
1142 !vcpu->arch.vpa.next_gpa))
1144 len -= len % sizeof(struct dtl_entry);
1145 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1147 case KVM_REG_PPC_TB_OFFSET:
1148 /* round up to multiple of 2^24 */
1149 vcpu->arch.vcore->tb_offset =
1150 ALIGN(set_reg_val(id, *val), 1UL << 24);
1152 case KVM_REG_PPC_LPCR:
1153 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1155 case KVM_REG_PPC_PPR:
1156 vcpu->arch.ppr = set_reg_val(id, *val);
1158 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1159 case KVM_REG_PPC_TFHAR:
1160 vcpu->arch.tfhar = set_reg_val(id, *val);
1162 case KVM_REG_PPC_TFIAR:
1163 vcpu->arch.tfiar = set_reg_val(id, *val);
1165 case KVM_REG_PPC_TEXASR:
1166 vcpu->arch.texasr = set_reg_val(id, *val);
1168 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1169 i = id - KVM_REG_PPC_TM_GPR0;
1170 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1172 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1175 i = id - KVM_REG_PPC_TM_VSR0;
1177 for (j = 0; j < TS_FPRWIDTH; j++)
1178 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1180 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1181 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1186 case KVM_REG_PPC_TM_CR:
1187 vcpu->arch.cr_tm = set_reg_val(id, *val);
1189 case KVM_REG_PPC_TM_LR:
1190 vcpu->arch.lr_tm = set_reg_val(id, *val);
1192 case KVM_REG_PPC_TM_CTR:
1193 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1195 case KVM_REG_PPC_TM_FPSCR:
1196 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1198 case KVM_REG_PPC_TM_AMR:
1199 vcpu->arch.amr_tm = set_reg_val(id, *val);
1201 case KVM_REG_PPC_TM_PPR:
1202 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1204 case KVM_REG_PPC_TM_VRSAVE:
1205 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1207 case KVM_REG_PPC_TM_VSCR:
1208 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1209 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1213 case KVM_REG_PPC_TM_DSCR:
1214 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1216 case KVM_REG_PPC_TM_TAR:
1217 vcpu->arch.tar_tm = set_reg_val(id, *val);
1220 case KVM_REG_PPC_ARCH_COMPAT:
1221 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1231 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1234 struct kvm_vcpu *vcpu;
1237 struct kvmppc_vcore *vcore;
1239 core = id / threads_per_subcore;
1240 if (core >= KVM_MAX_VCORES)
1244 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1248 err = kvm_vcpu_init(vcpu, kvm, id);
1252 vcpu->arch.shared = &vcpu->arch.shregs;
1253 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1255 * The shared struct is never shared on HV,
1256 * so we can always use host endianness
1258 #ifdef __BIG_ENDIAN__
1259 vcpu->arch.shared_big_endian = true;
1261 vcpu->arch.shared_big_endian = false;
1264 vcpu->arch.mmcr[0] = MMCR0_FC;
1265 vcpu->arch.ctrl = CTRL_RUNLATCH;
1266 /* default to host PVR, since we can't spoof it */
1267 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1268 spin_lock_init(&vcpu->arch.vpa_update_lock);
1269 spin_lock_init(&vcpu->arch.tbacct_lock);
1270 vcpu->arch.busy_preempt = TB_NIL;
1271 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1273 kvmppc_mmu_book3s_hv_init(vcpu);
1275 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1277 init_waitqueue_head(&vcpu->arch.cpu_run);
1279 mutex_lock(&kvm->lock);
1280 vcore = kvm->arch.vcores[core];
1282 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1284 INIT_LIST_HEAD(&vcore->runnable_threads);
1285 spin_lock_init(&vcore->lock);
1286 init_waitqueue_head(&vcore->wq);
1287 vcore->preempt_tb = TB_NIL;
1288 vcore->lpcr = kvm->arch.lpcr;
1289 vcore->first_vcpuid = core * threads_per_subcore;
1292 kvm->arch.vcores[core] = vcore;
1293 kvm->arch.online_vcores++;
1295 mutex_unlock(&kvm->lock);
1300 spin_lock(&vcore->lock);
1301 ++vcore->num_threads;
1302 spin_unlock(&vcore->lock);
1303 vcpu->arch.vcore = vcore;
1304 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1306 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1307 kvmppc_sanity_check(vcpu);
1312 kmem_cache_free(kvm_vcpu_cache, vcpu);
1314 return ERR_PTR(err);
1317 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1319 if (vpa->pinned_addr)
1320 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1324 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1326 spin_lock(&vcpu->arch.vpa_update_lock);
1327 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1328 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1329 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1330 spin_unlock(&vcpu->arch.vpa_update_lock);
1331 kvm_vcpu_uninit(vcpu);
1332 kmem_cache_free(kvm_vcpu_cache, vcpu);
1335 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1337 /* Indicate we want to get back into the guest */
1341 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1343 unsigned long dec_nsec, now;
1346 if (now > vcpu->arch.dec_expires) {
1347 /* decrementer has already gone negative */
1348 kvmppc_core_queue_dec(vcpu);
1349 kvmppc_core_prepare_to_enter(vcpu);
1352 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1354 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1356 vcpu->arch.timer_running = 1;
1359 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1361 vcpu->arch.ceded = 0;
1362 if (vcpu->arch.timer_running) {
1363 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1364 vcpu->arch.timer_running = 0;
1368 extern void __kvmppc_vcore_entry(void);
1370 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1371 struct kvm_vcpu *vcpu)
1375 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1377 spin_lock_irq(&vcpu->arch.tbacct_lock);
1379 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1380 vcpu->arch.stolen_logged;
1381 vcpu->arch.busy_preempt = now;
1382 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1383 spin_unlock_irq(&vcpu->arch.tbacct_lock);
1385 list_del(&vcpu->arch.run_list);
1388 static int kvmppc_grab_hwthread(int cpu)
1390 struct paca_struct *tpaca;
1391 long timeout = 1000;
1395 /* Ensure the thread won't go into the kernel if it wakes */
1396 tpaca->kvm_hstate.hwthread_req = 1;
1397 tpaca->kvm_hstate.kvm_vcpu = NULL;
1400 * If the thread is already executing in the kernel (e.g. handling
1401 * a stray interrupt), wait for it to get back to nap mode.
1402 * The smp_mb() is to ensure that our setting of hwthread_req
1403 * is visible before we look at hwthread_state, so if this
1404 * races with the code at system_reset_pSeries and the thread
1405 * misses our setting of hwthread_req, we are sure to see its
1406 * setting of hwthread_state, and vice versa.
1409 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1410 if (--timeout <= 0) {
1411 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1419 static void kvmppc_release_hwthread(int cpu)
1421 struct paca_struct *tpaca;
1424 tpaca->kvm_hstate.hwthread_req = 0;
1425 tpaca->kvm_hstate.kvm_vcpu = NULL;
1428 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1431 struct paca_struct *tpaca;
1432 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1434 if (vcpu->arch.timer_running) {
1435 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1436 vcpu->arch.timer_running = 0;
1438 cpu = vc->pcpu + vcpu->arch.ptid;
1440 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1441 tpaca->kvm_hstate.kvm_vcore = vc;
1442 tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1443 vcpu->cpu = vc->pcpu;
1445 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1446 if (cpu != smp_processor_id()) {
1448 if (vcpu->arch.ptid)
1454 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1460 while (vc->nap_count < vc->n_woken) {
1461 if (++i >= 1000000) {
1462 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1463 vc->nap_count, vc->n_woken);
1472 * Check that we are on thread 0 and that any other threads in
1473 * this core are off-line. Then grab the threads so they can't
1476 static int on_primary_thread(void)
1478 int cpu = smp_processor_id();
1481 /* Are we on a primary subcore? */
1482 if (cpu_thread_in_subcore(cpu))
1486 while (++thr < threads_per_subcore)
1487 if (cpu_online(cpu + thr))
1490 /* Grab all hw threads so they can't go into the kernel */
1491 for (thr = 1; thr < threads_per_subcore; ++thr) {
1492 if (kvmppc_grab_hwthread(cpu + thr)) {
1493 /* Couldn't grab one; let the others go */
1495 kvmppc_release_hwthread(cpu + thr);
1496 } while (--thr > 0);
1504 * Run a set of guest threads on a physical core.
1505 * Called with vc->lock held.
1507 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1509 struct kvm_vcpu *vcpu, *vnext;
1512 int i, need_vpa_update;
1514 struct kvm_vcpu *vcpus_to_update[threads_per_core];
1516 /* don't start if any threads have a signal pending */
1517 need_vpa_update = 0;
1518 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1519 if (signal_pending(vcpu->arch.run_task))
1521 if (vcpu->arch.vpa.update_pending ||
1522 vcpu->arch.slb_shadow.update_pending ||
1523 vcpu->arch.dtl.update_pending)
1524 vcpus_to_update[need_vpa_update++] = vcpu;
1528 * Initialize *vc, in particular vc->vcore_state, so we can
1529 * drop the vcore lock if necessary.
1533 vc->entry_exit_count = 0;
1534 vc->vcore_state = VCORE_STARTING;
1536 vc->napping_threads = 0;
1539 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1540 * which can't be called with any spinlocks held.
1542 if (need_vpa_update) {
1543 spin_unlock(&vc->lock);
1544 for (i = 0; i < need_vpa_update; ++i)
1545 kvmppc_update_vpas(vcpus_to_update[i]);
1546 spin_lock(&vc->lock);
1550 * Make sure we are running on primary threads, and that secondary
1551 * threads are offline. Also check if the number of threads in this
1552 * guest are greater than the current system threads per guest.
1554 if ((threads_per_core > 1) &&
1555 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1556 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1557 vcpu->arch.ret = -EBUSY;
1562 vc->pcpu = smp_processor_id();
1563 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1564 kvmppc_start_thread(vcpu);
1565 kvmppc_create_dtl_entry(vcpu, vc);
1568 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1569 get_paca()->kvm_hstate.kvm_vcore = vc;
1570 get_paca()->kvm_hstate.ptid = 0;
1572 vc->vcore_state = VCORE_RUNNING;
1574 spin_unlock(&vc->lock);
1578 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1580 __kvmppc_vcore_entry();
1582 spin_lock(&vc->lock);
1583 /* disable sending of IPIs on virtual external irqs */
1584 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1586 /* wait for secondary threads to finish writing their state to memory */
1587 if (vc->nap_count < vc->n_woken)
1588 kvmppc_wait_for_nap(vc);
1589 for (i = 0; i < threads_per_subcore; ++i)
1590 kvmppc_release_hwthread(vc->pcpu + i);
1591 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1592 vc->vcore_state = VCORE_EXITING;
1593 spin_unlock(&vc->lock);
1595 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1597 /* make sure updates to secondary vcpu structs are visible now */
1604 spin_lock(&vc->lock);
1606 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1607 /* cancel pending dec exception if dec is positive */
1608 if (now < vcpu->arch.dec_expires &&
1609 kvmppc_core_pending_dec(vcpu))
1610 kvmppc_core_dequeue_dec(vcpu);
1613 if (vcpu->arch.trap)
1614 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1615 vcpu->arch.run_task);
1617 vcpu->arch.ret = ret;
1618 vcpu->arch.trap = 0;
1620 if (vcpu->arch.ceded) {
1621 if (!is_kvmppc_resume_guest(ret))
1622 kvmppc_end_cede(vcpu);
1624 kvmppc_set_timer(vcpu);
1629 vc->vcore_state = VCORE_INACTIVE;
1630 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1632 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1633 kvmppc_remove_runnable(vc, vcpu);
1634 wake_up(&vcpu->arch.cpu_run);
1640 * Wait for some other vcpu thread to execute us, and
1641 * wake us up when we need to handle something in the host.
1643 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1647 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1648 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1650 finish_wait(&vcpu->arch.cpu_run, &wait);
1654 * All the vcpus in this vcore are idle, so wait for a decrementer
1655 * or external interrupt to one of the vcpus. vc->lock is held.
1657 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1661 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1662 vc->vcore_state = VCORE_SLEEPING;
1663 spin_unlock(&vc->lock);
1665 finish_wait(&vc->wq, &wait);
1666 spin_lock(&vc->lock);
1667 vc->vcore_state = VCORE_INACTIVE;
1670 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1673 struct kvmppc_vcore *vc;
1674 struct kvm_vcpu *v, *vn;
1676 kvm_run->exit_reason = 0;
1677 vcpu->arch.ret = RESUME_GUEST;
1678 vcpu->arch.trap = 0;
1679 kvmppc_update_vpas(vcpu);
1682 * Synchronize with other threads in this virtual core
1684 vc = vcpu->arch.vcore;
1685 spin_lock(&vc->lock);
1686 vcpu->arch.ceded = 0;
1687 vcpu->arch.run_task = current;
1688 vcpu->arch.kvm_run = kvm_run;
1689 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1690 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1691 vcpu->arch.busy_preempt = TB_NIL;
1692 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1696 * This happens the first time this is called for a vcpu.
1697 * If the vcore is already running, we may be able to start
1698 * this thread straight away and have it join in.
1700 if (!signal_pending(current)) {
1701 if (vc->vcore_state == VCORE_RUNNING &&
1702 VCORE_EXIT_COUNT(vc) == 0) {
1703 kvmppc_create_dtl_entry(vcpu, vc);
1704 kvmppc_start_thread(vcpu);
1705 } else if (vc->vcore_state == VCORE_SLEEPING) {
1711 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1712 !signal_pending(current)) {
1713 if (vc->vcore_state != VCORE_INACTIVE) {
1714 spin_unlock(&vc->lock);
1715 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1716 spin_lock(&vc->lock);
1719 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1721 kvmppc_core_prepare_to_enter(v);
1722 if (signal_pending(v->arch.run_task)) {
1723 kvmppc_remove_runnable(vc, v);
1724 v->stat.signal_exits++;
1725 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1726 v->arch.ret = -EINTR;
1727 wake_up(&v->arch.cpu_run);
1730 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1734 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1735 if (!v->arch.pending_exceptions)
1736 n_ceded += v->arch.ceded;
1740 if (n_ceded == vc->n_runnable)
1741 kvmppc_vcore_blocked(vc);
1743 kvmppc_run_core(vc);
1747 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1748 (vc->vcore_state == VCORE_RUNNING ||
1749 vc->vcore_state == VCORE_EXITING)) {
1750 spin_unlock(&vc->lock);
1751 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1752 spin_lock(&vc->lock);
1755 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1756 kvmppc_remove_runnable(vc, vcpu);
1757 vcpu->stat.signal_exits++;
1758 kvm_run->exit_reason = KVM_EXIT_INTR;
1759 vcpu->arch.ret = -EINTR;
1762 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1763 /* Wake up some vcpu to run the core */
1764 v = list_first_entry(&vc->runnable_threads,
1765 struct kvm_vcpu, arch.run_list);
1766 wake_up(&v->arch.cpu_run);
1769 spin_unlock(&vc->lock);
1770 return vcpu->arch.ret;
1773 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1778 if (!vcpu->arch.sane) {
1779 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1783 kvmppc_core_prepare_to_enter(vcpu);
1785 /* No need to go into the guest when all we'll do is come back out */
1786 if (signal_pending(current)) {
1787 run->exit_reason = KVM_EXIT_INTR;
1791 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1792 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1795 /* On the first time here, set up HTAB and VRMA or RMA */
1796 if (!vcpu->kvm->arch.rma_setup_done) {
1797 r = kvmppc_hv_setup_htab_rma(vcpu);
1802 flush_fp_to_thread(current);
1803 flush_altivec_to_thread(current);
1804 flush_vsx_to_thread(current);
1805 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1806 vcpu->arch.pgdir = current->mm->pgd;
1807 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1810 r = kvmppc_run_vcpu(run, vcpu);
1812 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1813 !(vcpu->arch.shregs.msr & MSR_PR)) {
1814 r = kvmppc_pseries_do_hcall(vcpu);
1815 kvmppc_core_prepare_to_enter(vcpu);
1816 } else if (r == RESUME_PAGE_FAULT) {
1817 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1818 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1819 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1820 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1822 } while (is_kvmppc_resume_guest(r));
1825 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1826 atomic_dec(&vcpu->kvm->arch.vcpus_running);
1831 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1832 Assumes POWER7 or PPC970. */
1833 static inline int lpcr_rmls(unsigned long rma_size)
1836 case 32ul << 20: /* 32 MB */
1837 if (cpu_has_feature(CPU_FTR_ARCH_206))
1838 return 8; /* only supported on POWER7 */
1840 case 64ul << 20: /* 64 MB */
1842 case 128ul << 20: /* 128 MB */
1844 case 256ul << 20: /* 256 MB */
1846 case 1ul << 30: /* 1 GB */
1848 case 16ul << 30: /* 16 GB */
1850 case 256ul << 30: /* 256 GB */
1857 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1860 struct kvm_rma_info *ri = vma->vm_file->private_data;
1862 if (vmf->pgoff >= kvm_rma_pages)
1863 return VM_FAULT_SIGBUS;
1865 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1871 static const struct vm_operations_struct kvm_rma_vm_ops = {
1872 .fault = kvm_rma_fault,
1875 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1877 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1878 vma->vm_ops = &kvm_rma_vm_ops;
1882 static int kvm_rma_release(struct inode *inode, struct file *filp)
1884 struct kvm_rma_info *ri = filp->private_data;
1886 kvm_release_rma(ri);
1890 static const struct file_operations kvm_rma_fops = {
1891 .mmap = kvm_rma_mmap,
1892 .release = kvm_rma_release,
1895 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1896 struct kvm_allocate_rma *ret)
1899 struct kvm_rma_info *ri;
1901 * Only do this on PPC970 in HV mode
1903 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1904 !cpu_has_feature(CPU_FTR_ARCH_201))
1910 ri = kvm_alloc_rma();
1914 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1916 kvm_release_rma(ri);
1918 ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1922 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1925 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1929 (*sps)->page_shift = def->shift;
1930 (*sps)->slb_enc = def->sllp;
1931 (*sps)->enc[0].page_shift = def->shift;
1933 * Only return base page encoding. We don't want to return
1934 * all the supporting pte_enc, because our H_ENTER doesn't
1935 * support MPSS yet. Once they do, we can start passing all
1936 * support pte_enc here
1938 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1940 * Add 16MB MPSS support if host supports it
1942 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
1943 (*sps)->enc[1].page_shift = 24;
1944 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
1949 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1950 struct kvm_ppc_smmu_info *info)
1952 struct kvm_ppc_one_seg_page_size *sps;
1954 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1955 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1956 info->flags |= KVM_PPC_1T_SEGMENTS;
1957 info->slb_size = mmu_slb_size;
1959 /* We only support these sizes for now, and no muti-size segments */
1960 sps = &info->sps[0];
1961 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1962 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1963 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1969 * Get (and clear) the dirty memory log for a memory slot.
1971 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1972 struct kvm_dirty_log *log)
1974 struct kvm_memory_slot *memslot;
1978 mutex_lock(&kvm->slots_lock);
1981 if (log->slot >= KVM_USER_MEM_SLOTS)
1984 memslot = id_to_memslot(kvm->memslots, log->slot);
1986 if (!memslot->dirty_bitmap)
1989 n = kvm_dirty_bitmap_bytes(memslot);
1990 memset(memslot->dirty_bitmap, 0, n);
1992 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1997 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2002 mutex_unlock(&kvm->slots_lock);
2006 static void unpin_slot(struct kvm_memory_slot *memslot)
2008 unsigned long *physp;
2009 unsigned long j, npages, pfn;
2012 physp = memslot->arch.slot_phys;
2013 npages = memslot->npages;
2016 for (j = 0; j < npages; j++) {
2017 if (!(physp[j] & KVMPPC_GOT_PAGE))
2019 pfn = physp[j] >> PAGE_SHIFT;
2020 page = pfn_to_page(pfn);
2026 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2027 struct kvm_memory_slot *dont)
2029 if (!dont || free->arch.rmap != dont->arch.rmap) {
2030 vfree(free->arch.rmap);
2031 free->arch.rmap = NULL;
2033 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
2035 vfree(free->arch.slot_phys);
2036 free->arch.slot_phys = NULL;
2040 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2041 unsigned long npages)
2043 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2044 if (!slot->arch.rmap)
2046 slot->arch.slot_phys = NULL;
2051 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2052 struct kvm_memory_slot *memslot,
2053 struct kvm_userspace_memory_region *mem)
2055 unsigned long *phys;
2057 /* Allocate a slot_phys array if needed */
2058 phys = memslot->arch.slot_phys;
2059 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
2060 phys = vzalloc(memslot->npages * sizeof(unsigned long));
2063 memslot->arch.slot_phys = phys;
2069 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2070 struct kvm_userspace_memory_region *mem,
2071 const struct kvm_memory_slot *old)
2073 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2074 struct kvm_memory_slot *memslot;
2076 if (npages && old->npages) {
2078 * If modifying a memslot, reset all the rmap dirty bits.
2079 * If this is a new memslot, we don't need to do anything
2080 * since the rmap array starts out as all zeroes,
2081 * i.e. no pages are dirty.
2083 memslot = id_to_memslot(kvm->memslots, mem->slot);
2084 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2089 * Update LPCR values in kvm->arch and in vcores.
2090 * Caller must hold kvm->lock.
2092 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2097 if ((kvm->arch.lpcr & mask) == lpcr)
2100 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2102 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2103 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2106 spin_lock(&vc->lock);
2107 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2108 spin_unlock(&vc->lock);
2109 if (++cores_done >= kvm->arch.online_vcores)
2114 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2119 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2122 struct kvm *kvm = vcpu->kvm;
2123 struct kvm_rma_info *ri = NULL;
2125 struct kvm_memory_slot *memslot;
2126 struct vm_area_struct *vma;
2127 unsigned long lpcr = 0, senc;
2128 unsigned long lpcr_mask = 0;
2129 unsigned long psize, porder;
2130 unsigned long rma_size;
2132 unsigned long *physp;
2133 unsigned long i, npages;
2136 mutex_lock(&kvm->lock);
2137 if (kvm->arch.rma_setup_done)
2138 goto out; /* another vcpu beat us to it */
2140 /* Allocate hashed page table (if not done already) and reset it */
2141 if (!kvm->arch.hpt_virt) {
2142 err = kvmppc_alloc_hpt(kvm, NULL);
2144 pr_err("KVM: Couldn't alloc HPT\n");
2149 /* Look up the memslot for guest physical address 0 */
2150 srcu_idx = srcu_read_lock(&kvm->srcu);
2151 memslot = gfn_to_memslot(kvm, 0);
2153 /* We must have some memory at 0 by now */
2155 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2158 /* Look up the VMA for the start of this memory slot */
2159 hva = memslot->userspace_addr;
2160 down_read(¤t->mm->mmap_sem);
2161 vma = find_vma(current->mm, hva);
2162 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2165 psize = vma_kernel_pagesize(vma);
2166 porder = __ilog2(psize);
2168 /* Is this one of our preallocated RMAs? */
2169 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2170 hva == vma->vm_start)
2171 ri = vma->vm_file->private_data;
2173 up_read(¤t->mm->mmap_sem);
2176 /* On POWER7, use VRMA; on PPC970, give up */
2178 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2179 pr_err("KVM: CPU requires an RMO\n");
2183 /* We can handle 4k, 64k or 16M pages in the VRMA */
2185 if (!(psize == 0x1000 || psize == 0x10000 ||
2186 psize == 0x1000000))
2189 /* Update VRMASD field in the LPCR */
2190 senc = slb_pgsize_encoding(psize);
2191 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2192 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2193 lpcr_mask = LPCR_VRMASD;
2194 /* the -4 is to account for senc values starting at 0x10 */
2195 lpcr = senc << (LPCR_VRMASD_SH - 4);
2197 /* Create HPTEs in the hash page table for the VRMA */
2198 kvmppc_map_vrma(vcpu, memslot, porder);
2201 /* Set up to use an RMO region */
2202 rma_size = kvm_rma_pages;
2203 if (rma_size > memslot->npages)
2204 rma_size = memslot->npages;
2205 rma_size <<= PAGE_SHIFT;
2206 rmls = lpcr_rmls(rma_size);
2208 if ((long)rmls < 0) {
2209 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2212 atomic_inc(&ri->use_count);
2215 /* Update LPCR and RMOR */
2216 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2217 /* PPC970; insert RMLS value (split field) in HID4 */
2218 lpcr_mask = (1ul << HID4_RMLS0_SH) |
2219 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2220 lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
2221 ((rmls & 3) << HID4_RMLS2_SH);
2222 /* RMOR is also in HID4 */
2223 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2227 lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2228 lpcr = rmls << LPCR_RMLS_SH;
2229 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
2231 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2232 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
2234 /* Initialize phys addrs of pages in RMO */
2235 npages = kvm_rma_pages;
2236 porder = __ilog2(npages);
2237 physp = memslot->arch.slot_phys;
2239 if (npages > memslot->npages)
2240 npages = memslot->npages;
2241 spin_lock(&kvm->arch.slot_phys_lock);
2242 for (i = 0; i < npages; ++i)
2243 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2245 spin_unlock(&kvm->arch.slot_phys_lock);
2249 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2251 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2253 kvm->arch.rma_setup_done = 1;
2256 srcu_read_unlock(&kvm->srcu, srcu_idx);
2258 mutex_unlock(&kvm->lock);
2262 up_read(¤t->mm->mmap_sem);
2266 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2268 unsigned long lpcr, lpid;
2270 /* Allocate the guest's logical partition ID */
2272 lpid = kvmppc_alloc_lpid();
2275 kvm->arch.lpid = lpid;
2278 * Since we don't flush the TLB when tearing down a VM,
2279 * and this lpid might have previously been used,
2280 * make sure we flush on each core before running the new VM.
2282 cpumask_setall(&kvm->arch.need_tlb_flush);
2284 kvm->arch.rma = NULL;
2286 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2288 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2289 /* PPC970; HID4 is effectively the LPCR */
2290 kvm->arch.host_lpid = 0;
2291 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2292 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2293 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2294 ((lpid & 0xf) << HID4_LPID5_SH);
2296 /* POWER7; init LPCR for virtual RMA mode */
2297 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2298 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2299 lpcr &= LPCR_PECE | LPCR_LPES;
2300 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2301 LPCR_VPM0 | LPCR_VPM1;
2302 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2303 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2304 /* On POWER8 turn on online bit to enable PURR/SPURR */
2305 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2308 kvm->arch.lpcr = lpcr;
2310 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2311 spin_lock_init(&kvm->arch.slot_phys_lock);
2314 * Track that we now have a HV mode VM active. This blocks secondary
2315 * CPU threads from coming online.
2317 kvm_hv_vm_activated();
2322 static void kvmppc_free_vcores(struct kvm *kvm)
2326 for (i = 0; i < KVM_MAX_VCORES; ++i)
2327 kfree(kvm->arch.vcores[i]);
2328 kvm->arch.online_vcores = 0;
2331 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2333 kvm_hv_vm_deactivated();
2335 kvmppc_free_vcores(kvm);
2336 if (kvm->arch.rma) {
2337 kvm_release_rma(kvm->arch.rma);
2338 kvm->arch.rma = NULL;
2341 kvmppc_free_hpt(kvm);
2344 /* We don't need to emulate any privileged instructions or dcbz */
2345 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2346 unsigned int inst, int *advance)
2348 return EMULATE_FAIL;
2351 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2354 return EMULATE_FAIL;
2357 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2360 return EMULATE_FAIL;
2363 static int kvmppc_core_check_processor_compat_hv(void)
2365 if (!cpu_has_feature(CPU_FTR_HVMODE))
2370 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2371 unsigned int ioctl, unsigned long arg)
2373 struct kvm *kvm __maybe_unused = filp->private_data;
2374 void __user *argp = (void __user *)arg;
2379 case KVM_ALLOCATE_RMA: {
2380 struct kvm_allocate_rma rma;
2381 struct kvm *kvm = filp->private_data;
2383 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2384 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2389 case KVM_PPC_ALLOCATE_HTAB: {
2393 if (get_user(htab_order, (u32 __user *)argp))
2395 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2399 if (put_user(htab_order, (u32 __user *)argp))
2405 case KVM_PPC_GET_HTAB_FD: {
2406 struct kvm_get_htab_fd ghf;
2409 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2411 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2422 static struct kvmppc_ops kvm_ops_hv = {
2423 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2424 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2425 .get_one_reg = kvmppc_get_one_reg_hv,
2426 .set_one_reg = kvmppc_set_one_reg_hv,
2427 .vcpu_load = kvmppc_core_vcpu_load_hv,
2428 .vcpu_put = kvmppc_core_vcpu_put_hv,
2429 .set_msr = kvmppc_set_msr_hv,
2430 .vcpu_run = kvmppc_vcpu_run_hv,
2431 .vcpu_create = kvmppc_core_vcpu_create_hv,
2432 .vcpu_free = kvmppc_core_vcpu_free_hv,
2433 .check_requests = kvmppc_core_check_requests_hv,
2434 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
2435 .flush_memslot = kvmppc_core_flush_memslot_hv,
2436 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2437 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
2438 .unmap_hva = kvm_unmap_hva_hv,
2439 .unmap_hva_range = kvm_unmap_hva_range_hv,
2440 .age_hva = kvm_age_hva_hv,
2441 .test_age_hva = kvm_test_age_hva_hv,
2442 .set_spte_hva = kvm_set_spte_hva_hv,
2443 .mmu_destroy = kvmppc_mmu_destroy_hv,
2444 .free_memslot = kvmppc_core_free_memslot_hv,
2445 .create_memslot = kvmppc_core_create_memslot_hv,
2446 .init_vm = kvmppc_core_init_vm_hv,
2447 .destroy_vm = kvmppc_core_destroy_vm_hv,
2448 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2449 .emulate_op = kvmppc_core_emulate_op_hv,
2450 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2451 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2452 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2453 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
2456 static int kvmppc_book3s_init_hv(void)
2460 * FIXME!! Do we need to check on all cpus ?
2462 r = kvmppc_core_check_processor_compat_hv();
2466 kvm_ops_hv.owner = THIS_MODULE;
2467 kvmppc_hv_ops = &kvm_ops_hv;
2469 r = kvmppc_mmu_hv_init();
2473 static void kvmppc_book3s_exit_hv(void)
2475 kvmppc_hv_ops = NULL;
2478 module_init(kvmppc_book3s_init_hv);
2479 module_exit(kvmppc_book3s_exit_hv);
2480 MODULE_LICENSE("GPL");
2481 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2482 MODULE_ALIAS("devname:kvm");