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>
50 #include <linux/gfp.h>
51 #include <linux/vmalloc.h>
52 #include <linux/highmem.h>
53 #include <linux/hugetlb.h>
55 /* #define EXIT_DEBUG */
56 /* #define EXIT_DEBUG_SIMPLE */
57 /* #define EXIT_DEBUG_INT */
59 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
60 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
62 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
64 struct kvmppc_vcore *vc = vcpu->arch.vcore;
66 local_paca->kvm_hstate.kvm_vcpu = vcpu;
67 local_paca->kvm_hstate.kvm_vcore = vc;
68 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
69 vc->stolen_tb += mftb() - vc->preempt_tb;
72 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
74 struct kvmppc_vcore *vc = vcpu->arch.vcore;
76 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
77 vc->preempt_tb = mftb();
80 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
82 vcpu->arch.shregs.msr = msr;
83 kvmppc_end_cede(vcpu);
86 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
91 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
95 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
96 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
97 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
98 for (r = 0; r < 16; ++r)
99 pr_err("r%2d = %.16lx r%d = %.16lx\n",
100 r, kvmppc_get_gpr(vcpu, r),
101 r+16, kvmppc_get_gpr(vcpu, r+16));
102 pr_err("ctr = %.16lx lr = %.16lx\n",
103 vcpu->arch.ctr, vcpu->arch.lr);
104 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
105 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
106 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
107 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
108 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
109 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
110 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
111 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
112 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
113 pr_err("fault dar = %.16lx dsisr = %.8x\n",
114 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
115 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
116 for (r = 0; r < vcpu->arch.slb_max; ++r)
117 pr_err(" ESID = %.16llx VSID = %.16llx\n",
118 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
119 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
120 vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
121 vcpu->arch.last_inst);
124 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
127 struct kvm_vcpu *v, *ret = NULL;
129 mutex_lock(&kvm->lock);
130 kvm_for_each_vcpu(r, v, kvm) {
131 if (v->vcpu_id == id) {
136 mutex_unlock(&kvm->lock);
140 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
142 vpa->shared_proc = 1;
143 vpa->yield_count = 1;
146 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
147 unsigned long addr, unsigned long len)
149 /* check address is cacheline aligned */
150 if (addr & (L1_CACHE_BYTES - 1))
152 spin_lock(&vcpu->arch.vpa_update_lock);
153 if (v->next_gpa != addr || v->len != len) {
155 v->len = addr ? len : 0;
156 v->update_pending = 1;
158 spin_unlock(&vcpu->arch.vpa_update_lock);
162 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
171 static int vpa_is_registered(struct kvmppc_vpa *vpap)
173 if (vpap->update_pending)
174 return vpap->next_gpa != 0;
175 return vpap->pinned_addr != NULL;
178 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
180 unsigned long vcpuid, unsigned long vpa)
182 struct kvm *kvm = vcpu->kvm;
183 unsigned long len, nb;
185 struct kvm_vcpu *tvcpu;
188 struct kvmppc_vpa *vpap;
190 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
194 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
195 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
196 subfunc == H_VPA_REG_SLB) {
197 /* Registering new area - address must be cache-line aligned */
198 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
201 /* convert logical addr to kernel addr and read length */
202 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
205 if (subfunc == H_VPA_REG_VPA)
206 len = ((struct reg_vpa *)va)->length.hword;
208 len = ((struct reg_vpa *)va)->length.word;
209 kvmppc_unpin_guest_page(kvm, va);
212 if (len > nb || len < sizeof(struct reg_vpa))
221 spin_lock(&tvcpu->arch.vpa_update_lock);
224 case H_VPA_REG_VPA: /* register VPA */
225 if (len < sizeof(struct lppaca))
227 vpap = &tvcpu->arch.vpa;
231 case H_VPA_REG_DTL: /* register DTL */
232 if (len < sizeof(struct dtl_entry))
234 len -= len % sizeof(struct dtl_entry);
236 /* Check that they have previously registered a VPA */
238 if (!vpa_is_registered(&tvcpu->arch.vpa))
241 vpap = &tvcpu->arch.dtl;
245 case H_VPA_REG_SLB: /* register SLB shadow buffer */
246 /* Check that they have previously registered a VPA */
248 if (!vpa_is_registered(&tvcpu->arch.vpa))
251 vpap = &tvcpu->arch.slb_shadow;
255 case H_VPA_DEREG_VPA: /* deregister VPA */
256 /* Check they don't still have a DTL or SLB buf registered */
258 if (vpa_is_registered(&tvcpu->arch.dtl) ||
259 vpa_is_registered(&tvcpu->arch.slb_shadow))
262 vpap = &tvcpu->arch.vpa;
266 case H_VPA_DEREG_DTL: /* deregister DTL */
267 vpap = &tvcpu->arch.dtl;
271 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
272 vpap = &tvcpu->arch.slb_shadow;
278 vpap->next_gpa = vpa;
280 vpap->update_pending = 1;
283 spin_unlock(&tvcpu->arch.vpa_update_lock);
288 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
290 struct kvm *kvm = vcpu->kvm;
296 * We need to pin the page pointed to by vpap->next_gpa,
297 * but we can't call kvmppc_pin_guest_page under the lock
298 * as it does get_user_pages() and down_read(). So we
299 * have to drop the lock, pin the page, then get the lock
300 * again and check that a new area didn't get registered
304 gpa = vpap->next_gpa;
305 spin_unlock(&vcpu->arch.vpa_update_lock);
309 va = kvmppc_pin_guest_page(kvm, vpap->next_gpa, &nb);
310 spin_lock(&vcpu->arch.vpa_update_lock);
311 if (gpa == vpap->next_gpa)
313 /* sigh... unpin that one and try again */
315 kvmppc_unpin_guest_page(kvm, va);
318 vpap->update_pending = 0;
319 if (va && nb < vpap->len) {
321 * If it's now too short, it must be that userspace
322 * has changed the mappings underlying guest memory,
323 * so unregister the region.
325 kvmppc_unpin_guest_page(kvm, va);
328 if (vpap->pinned_addr)
329 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr);
330 vpap->pinned_addr = va;
332 vpap->pinned_end = va + vpap->len;
335 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
337 spin_lock(&vcpu->arch.vpa_update_lock);
338 if (vcpu->arch.vpa.update_pending) {
339 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
340 if (vcpu->arch.vpa.pinned_addr)
341 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
343 if (vcpu->arch.dtl.update_pending) {
344 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
345 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
346 vcpu->arch.dtl_index = 0;
348 if (vcpu->arch.slb_shadow.update_pending)
349 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
350 spin_unlock(&vcpu->arch.vpa_update_lock);
353 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
354 struct kvmppc_vcore *vc)
356 struct dtl_entry *dt;
358 unsigned long old_stolen;
360 dt = vcpu->arch.dtl_ptr;
361 vpa = vcpu->arch.vpa.pinned_addr;
362 old_stolen = vcpu->arch.stolen_logged;
363 vcpu->arch.stolen_logged = vc->stolen_tb;
366 memset(dt, 0, sizeof(struct dtl_entry));
367 dt->dispatch_reason = 7;
368 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
369 dt->timebase = mftb();
370 dt->enqueue_to_dispatch_time = vc->stolen_tb - old_stolen;
371 dt->srr0 = kvmppc_get_pc(vcpu);
372 dt->srr1 = vcpu->arch.shregs.msr;
374 if (dt == vcpu->arch.dtl.pinned_end)
375 dt = vcpu->arch.dtl.pinned_addr;
376 vcpu->arch.dtl_ptr = dt;
377 /* order writing *dt vs. writing vpa->dtl_idx */
379 vpa->dtl_idx = ++vcpu->arch.dtl_index;
382 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
384 unsigned long req = kvmppc_get_gpr(vcpu, 3);
385 unsigned long target, ret = H_SUCCESS;
386 struct kvm_vcpu *tvcpu;
391 idx = srcu_read_lock(&vcpu->kvm->srcu);
392 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
393 kvmppc_get_gpr(vcpu, 5),
394 kvmppc_get_gpr(vcpu, 6),
395 kvmppc_get_gpr(vcpu, 7));
396 srcu_read_unlock(&vcpu->kvm->srcu, idx);
401 target = kvmppc_get_gpr(vcpu, 4);
402 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
407 tvcpu->arch.prodded = 1;
409 if (vcpu->arch.ceded) {
410 if (waitqueue_active(&vcpu->wq)) {
411 wake_up_interruptible(&vcpu->wq);
412 vcpu->stat.halt_wakeup++;
419 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
420 kvmppc_get_gpr(vcpu, 5),
421 kvmppc_get_gpr(vcpu, 6));
426 kvmppc_set_gpr(vcpu, 3, ret);
427 vcpu->arch.hcall_needed = 0;
431 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
432 struct task_struct *tsk)
437 vcpu->stat.sum_exits++;
439 run->exit_reason = KVM_EXIT_UNKNOWN;
440 run->ready_for_interrupt_injection = 1;
441 switch (vcpu->arch.trap) {
442 /* We're good on these - the host merely wanted to get our attention */
443 case BOOK3S_INTERRUPT_HV_DECREMENTER:
444 vcpu->stat.dec_exits++;
447 case BOOK3S_INTERRUPT_EXTERNAL:
448 vcpu->stat.ext_intr_exits++;
451 case BOOK3S_INTERRUPT_PERFMON:
454 case BOOK3S_INTERRUPT_PROGRAM:
458 * Normally program interrupts are delivered directly
459 * to the guest by the hardware, but we can get here
460 * as a result of a hypervisor emulation interrupt
461 * (e40) getting turned into a 700 by BML RTAS.
463 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
464 kvmppc_core_queue_program(vcpu, flags);
468 case BOOK3S_INTERRUPT_SYSCALL:
470 /* hcall - punt to userspace */
473 if (vcpu->arch.shregs.msr & MSR_PR) {
474 /* sc 1 from userspace - reflect to guest syscall */
475 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
479 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
480 for (i = 0; i < 9; ++i)
481 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
482 run->exit_reason = KVM_EXIT_PAPR_HCALL;
483 vcpu->arch.hcall_needed = 1;
488 * We get these next two if the guest accesses a page which it thinks
489 * it has mapped but which is not actually present, either because
490 * it is for an emulated I/O device or because the corresonding
491 * host page has been paged out. Any other HDSI/HISI interrupts
492 * have been handled already.
494 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
495 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
496 r = kvmppc_book3s_hv_page_fault(run, vcpu,
497 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
498 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
500 case BOOK3S_INTERRUPT_H_INST_STORAGE:
501 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
502 r = kvmppc_book3s_hv_page_fault(run, vcpu,
503 kvmppc_get_pc(vcpu), 0);
504 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
507 * This occurs if the guest executes an illegal instruction.
508 * We just generate a program interrupt to the guest, since
509 * we don't emulate any guest instructions at this stage.
511 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
512 kvmppc_core_queue_program(vcpu, 0x80000);
516 kvmppc_dump_regs(vcpu);
517 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
518 vcpu->arch.trap, kvmppc_get_pc(vcpu),
519 vcpu->arch.shregs.msr);
528 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
529 struct kvm_sregs *sregs)
533 sregs->pvr = vcpu->arch.pvr;
535 memset(sregs, 0, sizeof(struct kvm_sregs));
536 for (i = 0; i < vcpu->arch.slb_max; i++) {
537 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
538 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
544 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
545 struct kvm_sregs *sregs)
549 kvmppc_set_pvr(vcpu, sregs->pvr);
552 for (i = 0; i < vcpu->arch.slb_nr; i++) {
553 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
554 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
555 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
559 vcpu->arch.slb_max = j;
564 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
570 case KVM_REG_PPC_HIOR:
571 *val = get_reg_val(id, 0);
573 case KVM_REG_PPC_DABR:
574 *val = get_reg_val(id, vcpu->arch.dabr);
576 case KVM_REG_PPC_DSCR:
577 *val = get_reg_val(id, vcpu->arch.dscr);
579 case KVM_REG_PPC_PURR:
580 *val = get_reg_val(id, vcpu->arch.purr);
582 case KVM_REG_PPC_SPURR:
583 *val = get_reg_val(id, vcpu->arch.spurr);
585 case KVM_REG_PPC_AMR:
586 *val = get_reg_val(id, vcpu->arch.amr);
588 case KVM_REG_PPC_UAMOR:
589 *val = get_reg_val(id, vcpu->arch.uamor);
591 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
592 i = id - KVM_REG_PPC_MMCR0;
593 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
595 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
596 i = id - KVM_REG_PPC_PMC1;
597 *val = get_reg_val(id, vcpu->arch.pmc[i]);
600 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
601 if (cpu_has_feature(CPU_FTR_VSX)) {
602 /* VSX => FP reg i is stored in arch.vsr[2*i] */
603 long int i = id - KVM_REG_PPC_FPR0;
604 *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
606 /* let generic code handle it */
610 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
611 if (cpu_has_feature(CPU_FTR_VSX)) {
612 long int i = id - KVM_REG_PPC_VSR0;
613 val->vsxval[0] = vcpu->arch.vsr[2 * i];
614 val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
619 #endif /* CONFIG_VSX */
620 case KVM_REG_PPC_VPA_ADDR:
621 spin_lock(&vcpu->arch.vpa_update_lock);
622 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
623 spin_unlock(&vcpu->arch.vpa_update_lock);
625 case KVM_REG_PPC_VPA_SLB:
626 spin_lock(&vcpu->arch.vpa_update_lock);
627 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
628 val->vpaval.length = vcpu->arch.slb_shadow.len;
629 spin_unlock(&vcpu->arch.vpa_update_lock);
631 case KVM_REG_PPC_VPA_DTL:
632 spin_lock(&vcpu->arch.vpa_update_lock);
633 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
634 val->vpaval.length = vcpu->arch.dtl.len;
635 spin_unlock(&vcpu->arch.vpa_update_lock);
645 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
649 unsigned long addr, len;
652 case KVM_REG_PPC_HIOR:
653 /* Only allow this to be set to zero */
654 if (set_reg_val(id, *val))
657 case KVM_REG_PPC_DABR:
658 vcpu->arch.dabr = set_reg_val(id, *val);
660 case KVM_REG_PPC_DSCR:
661 vcpu->arch.dscr = set_reg_val(id, *val);
663 case KVM_REG_PPC_PURR:
664 vcpu->arch.purr = set_reg_val(id, *val);
666 case KVM_REG_PPC_SPURR:
667 vcpu->arch.spurr = set_reg_val(id, *val);
669 case KVM_REG_PPC_AMR:
670 vcpu->arch.amr = set_reg_val(id, *val);
672 case KVM_REG_PPC_UAMOR:
673 vcpu->arch.uamor = set_reg_val(id, *val);
675 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
676 i = id - KVM_REG_PPC_MMCR0;
677 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
679 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
680 i = id - KVM_REG_PPC_PMC1;
681 vcpu->arch.pmc[i] = set_reg_val(id, *val);
684 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
685 if (cpu_has_feature(CPU_FTR_VSX)) {
686 /* VSX => FP reg i is stored in arch.vsr[2*i] */
687 long int i = id - KVM_REG_PPC_FPR0;
688 vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
690 /* let generic code handle it */
694 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
695 if (cpu_has_feature(CPU_FTR_VSX)) {
696 long int i = id - KVM_REG_PPC_VSR0;
697 vcpu->arch.vsr[2 * i] = val->vsxval[0];
698 vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
703 #endif /* CONFIG_VSX */
704 case KVM_REG_PPC_VPA_ADDR:
705 addr = set_reg_val(id, *val);
707 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
708 vcpu->arch.dtl.next_gpa))
710 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
712 case KVM_REG_PPC_VPA_SLB:
713 addr = val->vpaval.addr;
714 len = val->vpaval.length;
716 if (addr && !vcpu->arch.vpa.next_gpa)
718 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
720 case KVM_REG_PPC_VPA_DTL:
721 addr = val->vpaval.addr;
722 len = val->vpaval.length;
724 if (len < sizeof(struct dtl_entry))
726 if (addr && !vcpu->arch.vpa.next_gpa)
728 len -= len % sizeof(struct dtl_entry);
729 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
739 int kvmppc_core_check_processor_compat(void)
741 if (cpu_has_feature(CPU_FTR_HVMODE))
746 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
748 struct kvm_vcpu *vcpu;
751 struct kvmppc_vcore *vcore;
753 core = id / threads_per_core;
754 if (core >= KVM_MAX_VCORES)
758 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
762 err = kvm_vcpu_init(vcpu, kvm, id);
766 vcpu->arch.shared = &vcpu->arch.shregs;
767 vcpu->arch.last_cpu = -1;
768 vcpu->arch.mmcr[0] = MMCR0_FC;
769 vcpu->arch.ctrl = CTRL_RUNLATCH;
770 /* default to host PVR, since we can't spoof it */
771 vcpu->arch.pvr = mfspr(SPRN_PVR);
772 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
773 spin_lock_init(&vcpu->arch.vpa_update_lock);
775 kvmppc_mmu_book3s_hv_init(vcpu);
778 * We consider the vcpu stopped until we see the first run ioctl for it.
780 vcpu->arch.state = KVMPPC_VCPU_STOPPED;
782 init_waitqueue_head(&vcpu->arch.cpu_run);
784 mutex_lock(&kvm->lock);
785 vcore = kvm->arch.vcores[core];
787 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
789 INIT_LIST_HEAD(&vcore->runnable_threads);
790 spin_lock_init(&vcore->lock);
791 init_waitqueue_head(&vcore->wq);
792 vcore->preempt_tb = mftb();
794 kvm->arch.vcores[core] = vcore;
796 mutex_unlock(&kvm->lock);
801 spin_lock(&vcore->lock);
802 ++vcore->num_threads;
803 spin_unlock(&vcore->lock);
804 vcpu->arch.vcore = vcore;
805 vcpu->arch.stolen_logged = vcore->stolen_tb;
807 vcpu->arch.cpu_type = KVM_CPU_3S_64;
808 kvmppc_sanity_check(vcpu);
813 kmem_cache_free(kvm_vcpu_cache, vcpu);
818 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
820 spin_lock(&vcpu->arch.vpa_update_lock);
821 if (vcpu->arch.dtl.pinned_addr)
822 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl.pinned_addr);
823 if (vcpu->arch.slb_shadow.pinned_addr)
824 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow.pinned_addr);
825 if (vcpu->arch.vpa.pinned_addr)
826 kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa.pinned_addr);
827 spin_unlock(&vcpu->arch.vpa_update_lock);
828 kvm_vcpu_uninit(vcpu);
829 kmem_cache_free(kvm_vcpu_cache, vcpu);
832 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
834 unsigned long dec_nsec, now;
837 if (now > vcpu->arch.dec_expires) {
838 /* decrementer has already gone negative */
839 kvmppc_core_queue_dec(vcpu);
840 kvmppc_core_prepare_to_enter(vcpu);
843 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
845 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
847 vcpu->arch.timer_running = 1;
850 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
852 vcpu->arch.ceded = 0;
853 if (vcpu->arch.timer_running) {
854 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
855 vcpu->arch.timer_running = 0;
859 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
860 extern void xics_wake_cpu(int cpu);
862 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
863 struct kvm_vcpu *vcpu)
865 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
867 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
870 list_del(&vcpu->arch.run_list);
873 static int kvmppc_grab_hwthread(int cpu)
875 struct paca_struct *tpaca;
880 /* Ensure the thread won't go into the kernel if it wakes */
881 tpaca->kvm_hstate.hwthread_req = 1;
884 * If the thread is already executing in the kernel (e.g. handling
885 * a stray interrupt), wait for it to get back to nap mode.
886 * The smp_mb() is to ensure that our setting of hwthread_req
887 * is visible before we look at hwthread_state, so if this
888 * races with the code at system_reset_pSeries and the thread
889 * misses our setting of hwthread_req, we are sure to see its
890 * setting of hwthread_state, and vice versa.
893 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
894 if (--timeout <= 0) {
895 pr_err("KVM: couldn't grab cpu %d\n", cpu);
903 static void kvmppc_release_hwthread(int cpu)
905 struct paca_struct *tpaca;
908 tpaca->kvm_hstate.hwthread_req = 0;
909 tpaca->kvm_hstate.kvm_vcpu = NULL;
912 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
915 struct paca_struct *tpaca;
916 struct kvmppc_vcore *vc = vcpu->arch.vcore;
918 if (vcpu->arch.timer_running) {
919 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
920 vcpu->arch.timer_running = 0;
922 cpu = vc->pcpu + vcpu->arch.ptid;
924 tpaca->kvm_hstate.kvm_vcpu = vcpu;
925 tpaca->kvm_hstate.kvm_vcore = vc;
926 tpaca->kvm_hstate.napping = 0;
927 vcpu->cpu = vc->pcpu;
929 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
930 if (vcpu->arch.ptid) {
931 kvmppc_grab_hwthread(cpu);
938 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
944 while (vc->nap_count < vc->n_woken) {
945 if (++i >= 1000000) {
946 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
947 vc->nap_count, vc->n_woken);
956 * Check that we are on thread 0 and that any other threads in
957 * this core are off-line.
959 static int on_primary_thread(void)
961 int cpu = smp_processor_id();
962 int thr = cpu_thread_in_core(cpu);
966 while (++thr < threads_per_core)
967 if (cpu_online(cpu + thr))
973 * Run a set of guest threads on a physical core.
974 * Called with vc->lock held.
976 static int kvmppc_run_core(struct kvmppc_vcore *vc)
978 struct kvm_vcpu *vcpu, *vcpu0, *vnext;
981 int ptid, i, need_vpa_update;
984 /* don't start if any threads have a signal pending */
986 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
987 if (signal_pending(vcpu->arch.run_task))
989 need_vpa_update |= vcpu->arch.vpa.update_pending |
990 vcpu->arch.slb_shadow.update_pending |
991 vcpu->arch.dtl.update_pending;
995 * Initialize *vc, in particular vc->vcore_state, so we can
996 * drop the vcore lock if necessary.
1000 vc->entry_exit_count = 0;
1001 vc->vcore_state = VCORE_RUNNING;
1003 vc->napping_threads = 0;
1006 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1007 * which can't be called with any spinlocks held.
1009 if (need_vpa_update) {
1010 spin_unlock(&vc->lock);
1011 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1012 kvmppc_update_vpas(vcpu);
1013 spin_lock(&vc->lock);
1017 * Make sure we are running on thread 0, and that
1018 * secondary threads are offline.
1019 * XXX we should also block attempts to bring any
1020 * secondary threads online.
1022 if (threads_per_core > 1 && !on_primary_thread()) {
1023 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1024 vcpu->arch.ret = -EBUSY;
1029 * Assign physical thread IDs, first to non-ceded vcpus
1030 * and then to ceded ones.
1034 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1035 if (!vcpu->arch.ceded) {
1038 vcpu->arch.ptid = ptid++;
1042 return 0; /* nothing to run */
1043 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1044 if (vcpu->arch.ceded)
1045 vcpu->arch.ptid = ptid++;
1047 vc->stolen_tb += mftb() - vc->preempt_tb;
1048 vc->pcpu = smp_processor_id();
1049 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1050 kvmppc_start_thread(vcpu);
1051 kvmppc_create_dtl_entry(vcpu, vc);
1053 /* Grab any remaining hw threads so they can't go into the kernel */
1054 for (i = ptid; i < threads_per_core; ++i)
1055 kvmppc_grab_hwthread(vc->pcpu + i);
1058 spin_unlock(&vc->lock);
1062 srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1064 __kvmppc_vcore_entry(NULL, vcpu0);
1065 for (i = 0; i < threads_per_core; ++i)
1066 kvmppc_release_hwthread(vc->pcpu + i);
1068 spin_lock(&vc->lock);
1069 /* disable sending of IPIs on virtual external irqs */
1070 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1072 /* wait for secondary threads to finish writing their state to memory */
1073 if (vc->nap_count < vc->n_woken)
1074 kvmppc_wait_for_nap(vc);
1075 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1076 vc->vcore_state = VCORE_EXITING;
1077 spin_unlock(&vc->lock);
1079 srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1081 /* make sure updates to secondary vcpu structs are visible now */
1089 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1090 /* cancel pending dec exception if dec is positive */
1091 if (now < vcpu->arch.dec_expires &&
1092 kvmppc_core_pending_dec(vcpu))
1093 kvmppc_core_dequeue_dec(vcpu);
1096 if (vcpu->arch.trap)
1097 ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1098 vcpu->arch.run_task);
1100 vcpu->arch.ret = ret;
1101 vcpu->arch.trap = 0;
1103 if (vcpu->arch.ceded) {
1104 if (ret != RESUME_GUEST)
1105 kvmppc_end_cede(vcpu);
1107 kvmppc_set_timer(vcpu);
1111 spin_lock(&vc->lock);
1113 vc->vcore_state = VCORE_INACTIVE;
1114 vc->preempt_tb = mftb();
1115 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1117 if (vcpu->arch.ret != RESUME_GUEST) {
1118 kvmppc_remove_runnable(vc, vcpu);
1119 wake_up(&vcpu->arch.cpu_run);
1127 * Wait for some other vcpu thread to execute us, and
1128 * wake us up when we need to handle something in the host.
1130 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1134 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1135 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1137 finish_wait(&vcpu->arch.cpu_run, &wait);
1141 * All the vcpus in this vcore are idle, so wait for a decrementer
1142 * or external interrupt to one of the vcpus. vc->lock is held.
1144 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1150 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1151 vc->vcore_state = VCORE_SLEEPING;
1152 spin_unlock(&vc->lock);
1153 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1154 if (!v->arch.ceded || v->arch.pending_exceptions) {
1161 finish_wait(&vc->wq, &wait);
1162 spin_lock(&vc->lock);
1163 vc->vcore_state = VCORE_INACTIVE;
1166 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1170 struct kvmppc_vcore *vc;
1171 struct kvm_vcpu *v, *vn;
1173 kvm_run->exit_reason = 0;
1174 vcpu->arch.ret = RESUME_GUEST;
1175 vcpu->arch.trap = 0;
1178 * Synchronize with other threads in this virtual core
1180 vc = vcpu->arch.vcore;
1181 spin_lock(&vc->lock);
1182 vcpu->arch.ceded = 0;
1183 vcpu->arch.run_task = current;
1184 vcpu->arch.kvm_run = kvm_run;
1185 prev_state = vcpu->arch.state;
1186 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1187 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1191 * This happens the first time this is called for a vcpu.
1192 * If the vcore is already running, we may be able to start
1193 * this thread straight away and have it join in.
1195 if (prev_state == KVMPPC_VCPU_STOPPED) {
1196 if (vc->vcore_state == VCORE_RUNNING &&
1197 VCORE_EXIT_COUNT(vc) == 0) {
1198 vcpu->arch.ptid = vc->n_runnable - 1;
1199 kvmppc_start_thread(vcpu);
1202 } else if (prev_state == KVMPPC_VCPU_BUSY_IN_HOST)
1205 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1206 !signal_pending(current)) {
1207 if (vc->n_busy || vc->vcore_state != VCORE_INACTIVE) {
1208 spin_unlock(&vc->lock);
1209 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1210 spin_lock(&vc->lock);
1215 list_for_each_entry(v, &vc->runnable_threads, arch.run_list)
1216 n_ceded += v->arch.ceded;
1217 if (n_ceded == vc->n_runnable)
1218 kvmppc_vcore_blocked(vc);
1220 kvmppc_run_core(vc);
1222 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1224 kvmppc_core_prepare_to_enter(v);
1225 if (signal_pending(v->arch.run_task)) {
1226 kvmppc_remove_runnable(vc, v);
1227 v->stat.signal_exits++;
1228 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1229 v->arch.ret = -EINTR;
1230 wake_up(&v->arch.cpu_run);
1236 if (signal_pending(current)) {
1237 if (vc->vcore_state == VCORE_RUNNING ||
1238 vc->vcore_state == VCORE_EXITING) {
1239 spin_unlock(&vc->lock);
1240 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1241 spin_lock(&vc->lock);
1243 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1244 kvmppc_remove_runnable(vc, vcpu);
1245 vcpu->stat.signal_exits++;
1246 kvm_run->exit_reason = KVM_EXIT_INTR;
1247 vcpu->arch.ret = -EINTR;
1251 spin_unlock(&vc->lock);
1252 return vcpu->arch.ret;
1255 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1259 if (!vcpu->arch.sane) {
1260 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1264 kvmppc_core_prepare_to_enter(vcpu);
1266 /* No need to go into the guest when all we'll do is come back out */
1267 if (signal_pending(current)) {
1268 run->exit_reason = KVM_EXIT_INTR;
1272 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1273 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1276 /* On the first time here, set up HTAB and VRMA or RMA */
1277 if (!vcpu->kvm->arch.rma_setup_done) {
1278 r = kvmppc_hv_setup_htab_rma(vcpu);
1283 flush_fp_to_thread(current);
1284 flush_altivec_to_thread(current);
1285 flush_vsx_to_thread(current);
1286 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1287 vcpu->arch.pgdir = current->mm->pgd;
1290 r = kvmppc_run_vcpu(run, vcpu);
1292 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1293 !(vcpu->arch.shregs.msr & MSR_PR)) {
1294 r = kvmppc_pseries_do_hcall(vcpu);
1295 kvmppc_core_prepare_to_enter(vcpu);
1297 } while (r == RESUME_GUEST);
1300 atomic_dec(&vcpu->kvm->arch.vcpus_running);
1305 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1306 Assumes POWER7 or PPC970. */
1307 static inline int lpcr_rmls(unsigned long rma_size)
1310 case 32ul << 20: /* 32 MB */
1311 if (cpu_has_feature(CPU_FTR_ARCH_206))
1312 return 8; /* only supported on POWER7 */
1314 case 64ul << 20: /* 64 MB */
1316 case 128ul << 20: /* 128 MB */
1318 case 256ul << 20: /* 256 MB */
1320 case 1ul << 30: /* 1 GB */
1322 case 16ul << 30: /* 16 GB */
1324 case 256ul << 30: /* 256 GB */
1331 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1333 struct kvmppc_linear_info *ri = vma->vm_file->private_data;
1336 if (vmf->pgoff >= ri->npages)
1337 return VM_FAULT_SIGBUS;
1339 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1345 static const struct vm_operations_struct kvm_rma_vm_ops = {
1346 .fault = kvm_rma_fault,
1349 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1351 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1352 vma->vm_ops = &kvm_rma_vm_ops;
1356 static int kvm_rma_release(struct inode *inode, struct file *filp)
1358 struct kvmppc_linear_info *ri = filp->private_data;
1360 kvm_release_rma(ri);
1364 static struct file_operations kvm_rma_fops = {
1365 .mmap = kvm_rma_mmap,
1366 .release = kvm_rma_release,
1369 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1371 struct kvmppc_linear_info *ri;
1374 ri = kvm_alloc_rma();
1378 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1380 kvm_release_rma(ri);
1382 ret->rma_size = ri->npages << PAGE_SHIFT;
1386 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1389 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1393 (*sps)->page_shift = def->shift;
1394 (*sps)->slb_enc = def->sllp;
1395 (*sps)->enc[0].page_shift = def->shift;
1396 (*sps)->enc[0].pte_enc = def->penc;
1400 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1402 struct kvm_ppc_one_seg_page_size *sps;
1404 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1405 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1406 info->flags |= KVM_PPC_1T_SEGMENTS;
1407 info->slb_size = mmu_slb_size;
1409 /* We only support these sizes for now, and no muti-size segments */
1410 sps = &info->sps[0];
1411 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1412 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1413 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1419 * Get (and clear) the dirty memory log for a memory slot.
1421 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1423 struct kvm_memory_slot *memslot;
1427 mutex_lock(&kvm->slots_lock);
1430 if (log->slot >= KVM_MEMORY_SLOTS)
1433 memslot = id_to_memslot(kvm->memslots, log->slot);
1435 if (!memslot->dirty_bitmap)
1438 n = kvm_dirty_bitmap_bytes(memslot);
1439 memset(memslot->dirty_bitmap, 0, n);
1441 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1446 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1451 mutex_unlock(&kvm->slots_lock);
1455 static unsigned long slb_pgsize_encoding(unsigned long psize)
1457 unsigned long senc = 0;
1459 if (psize > 0x1000) {
1461 if (psize == 0x10000)
1462 senc |= SLB_VSID_LP_01;
1467 static void unpin_slot(struct kvm_memory_slot *memslot)
1469 unsigned long *physp;
1470 unsigned long j, npages, pfn;
1473 physp = memslot->arch.slot_phys;
1474 npages = memslot->npages;
1477 for (j = 0; j < npages; j++) {
1478 if (!(physp[j] & KVMPPC_GOT_PAGE))
1480 pfn = physp[j] >> PAGE_SHIFT;
1481 page = pfn_to_page(pfn);
1487 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1488 struct kvm_memory_slot *dont)
1490 if (!dont || free->arch.rmap != dont->arch.rmap) {
1491 vfree(free->arch.rmap);
1492 free->arch.rmap = NULL;
1494 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1496 vfree(free->arch.slot_phys);
1497 free->arch.slot_phys = NULL;
1501 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1502 unsigned long npages)
1504 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1505 if (!slot->arch.rmap)
1507 slot->arch.slot_phys = NULL;
1512 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1513 struct kvm_memory_slot *memslot,
1514 struct kvm_userspace_memory_region *mem)
1516 unsigned long *phys;
1518 /* Allocate a slot_phys array if needed */
1519 phys = memslot->arch.slot_phys;
1520 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1521 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1524 memslot->arch.slot_phys = phys;
1530 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1531 struct kvm_userspace_memory_region *mem,
1532 struct kvm_memory_slot old)
1534 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1535 struct kvm_memory_slot *memslot;
1537 if (npages && old.npages) {
1539 * If modifying a memslot, reset all the rmap dirty bits.
1540 * If this is a new memslot, we don't need to do anything
1541 * since the rmap array starts out as all zeroes,
1542 * i.e. no pages are dirty.
1544 memslot = id_to_memslot(kvm->memslots, mem->slot);
1545 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1549 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1552 struct kvm *kvm = vcpu->kvm;
1553 struct kvmppc_linear_info *ri = NULL;
1555 struct kvm_memory_slot *memslot;
1556 struct vm_area_struct *vma;
1557 unsigned long lpcr, senc;
1558 unsigned long psize, porder;
1559 unsigned long rma_size;
1561 unsigned long *physp;
1562 unsigned long i, npages;
1565 mutex_lock(&kvm->lock);
1566 if (kvm->arch.rma_setup_done)
1567 goto out; /* another vcpu beat us to it */
1569 /* Allocate hashed page table (if not done already) and reset it */
1570 if (!kvm->arch.hpt_virt) {
1571 err = kvmppc_alloc_hpt(kvm, NULL);
1573 pr_err("KVM: Couldn't alloc HPT\n");
1578 /* Look up the memslot for guest physical address 0 */
1579 srcu_idx = srcu_read_lock(&kvm->srcu);
1580 memslot = gfn_to_memslot(kvm, 0);
1582 /* We must have some memory at 0 by now */
1584 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1587 /* Look up the VMA for the start of this memory slot */
1588 hva = memslot->userspace_addr;
1589 down_read(¤t->mm->mmap_sem);
1590 vma = find_vma(current->mm, hva);
1591 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1594 psize = vma_kernel_pagesize(vma);
1595 porder = __ilog2(psize);
1597 /* Is this one of our preallocated RMAs? */
1598 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1599 hva == vma->vm_start)
1600 ri = vma->vm_file->private_data;
1602 up_read(¤t->mm->mmap_sem);
1605 /* On POWER7, use VRMA; on PPC970, give up */
1607 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1608 pr_err("KVM: CPU requires an RMO\n");
1612 /* We can handle 4k, 64k or 16M pages in the VRMA */
1614 if (!(psize == 0x1000 || psize == 0x10000 ||
1615 psize == 0x1000000))
1618 /* Update VRMASD field in the LPCR */
1619 senc = slb_pgsize_encoding(psize);
1620 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1621 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1622 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1623 lpcr |= senc << (LPCR_VRMASD_SH - 4);
1624 kvm->arch.lpcr = lpcr;
1626 /* Create HPTEs in the hash page table for the VRMA */
1627 kvmppc_map_vrma(vcpu, memslot, porder);
1630 /* Set up to use an RMO region */
1631 rma_size = ri->npages;
1632 if (rma_size > memslot->npages)
1633 rma_size = memslot->npages;
1634 rma_size <<= PAGE_SHIFT;
1635 rmls = lpcr_rmls(rma_size);
1638 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1641 atomic_inc(&ri->use_count);
1644 /* Update LPCR and RMOR */
1645 lpcr = kvm->arch.lpcr;
1646 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1647 /* PPC970; insert RMLS value (split field) in HID4 */
1648 lpcr &= ~((1ul << HID4_RMLS0_SH) |
1649 (3ul << HID4_RMLS2_SH));
1650 lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1651 ((rmls & 3) << HID4_RMLS2_SH);
1652 /* RMOR is also in HID4 */
1653 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1657 lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1658 lpcr |= rmls << LPCR_RMLS_SH;
1659 kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1661 kvm->arch.lpcr = lpcr;
1662 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1663 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1665 /* Initialize phys addrs of pages in RMO */
1666 npages = ri->npages;
1667 porder = __ilog2(npages);
1668 physp = memslot->arch.slot_phys;
1670 if (npages > memslot->npages)
1671 npages = memslot->npages;
1672 spin_lock(&kvm->arch.slot_phys_lock);
1673 for (i = 0; i < npages; ++i)
1674 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1676 spin_unlock(&kvm->arch.slot_phys_lock);
1680 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1682 kvm->arch.rma_setup_done = 1;
1685 srcu_read_unlock(&kvm->srcu, srcu_idx);
1687 mutex_unlock(&kvm->lock);
1691 up_read(¤t->mm->mmap_sem);
1695 int kvmppc_core_init_vm(struct kvm *kvm)
1697 unsigned long lpcr, lpid;
1699 /* Allocate the guest's logical partition ID */
1701 lpid = kvmppc_alloc_lpid();
1704 kvm->arch.lpid = lpid;
1706 INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1708 kvm->arch.rma = NULL;
1710 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1712 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1713 /* PPC970; HID4 is effectively the LPCR */
1714 kvm->arch.host_lpid = 0;
1715 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1716 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1717 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1718 ((lpid & 0xf) << HID4_LPID5_SH);
1720 /* POWER7; init LPCR for virtual RMA mode */
1721 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1722 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1723 lpcr &= LPCR_PECE | LPCR_LPES;
1724 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1725 LPCR_VPM0 | LPCR_VPM1;
1726 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1727 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1729 kvm->arch.lpcr = lpcr;
1731 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1732 spin_lock_init(&kvm->arch.slot_phys_lock);
1736 void kvmppc_core_destroy_vm(struct kvm *kvm)
1738 if (kvm->arch.rma) {
1739 kvm_release_rma(kvm->arch.rma);
1740 kvm->arch.rma = NULL;
1743 kvmppc_free_hpt(kvm);
1744 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1747 /* These are stubs for now */
1748 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1752 /* We don't need to emulate any privileged instructions or dcbz */
1753 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1754 unsigned int inst, int *advance)
1756 return EMULATE_FAIL;
1759 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
1761 return EMULATE_FAIL;
1764 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
1766 return EMULATE_FAIL;
1769 static int kvmppc_book3s_hv_init(void)
1773 r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1778 r = kvmppc_mmu_hv_init();
1783 static void kvmppc_book3s_hv_exit(void)
1788 module_init(kvmppc_book3s_hv_init);
1789 module_exit(kvmppc_book3s_hv_exit);