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>
35 #include <linux/debugfs.h>
38 #include <asm/cputable.h>
39 #include <asm/cache.h>
40 #include <asm/cacheflush.h>
41 #include <asm/tlbflush.h>
42 #include <asm/uaccess.h>
44 #include <asm/kvm_ppc.h>
45 #include <asm/kvm_book3s.h>
46 #include <asm/mmu_context.h>
47 #include <asm/lppaca.h>
48 #include <asm/processor.h>
49 #include <asm/cputhreads.h>
51 #include <asm/hvcall.h>
52 #include <asm/switch_to.h>
54 #include <linux/gfp.h>
55 #include <linux/vmalloc.h>
56 #include <linux/highmem.h>
57 #include <linux/hugetlb.h>
58 #include <linux/module.h>
62 #define CREATE_TRACE_POINTS
65 /* #define EXIT_DEBUG */
66 /* #define EXIT_DEBUG_SIMPLE */
67 /* #define EXIT_DEBUG_INT */
69 /* Used to indicate that a guest page fault needs to be handled */
70 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
72 /* Used as a "null" value for timebase values */
73 #define TB_NIL (~(u64)0)
75 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
77 #if defined(CONFIG_PPC_64K_PAGES)
78 #define MPP_BUFFER_ORDER 0
79 #elif defined(CONFIG_PPC_4K_PAGES)
80 #define MPP_BUFFER_ORDER 3
84 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
85 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
87 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
91 wait_queue_head_t *wqp;
93 wqp = kvm_arch_vcpu_wq(vcpu);
94 if (waitqueue_active(wqp)) {
95 wake_up_interruptible(wqp);
96 ++vcpu->stat.halt_wakeup;
101 /* CPU points to the first thread of the core */
102 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
103 #ifdef CONFIG_PPC_ICP_NATIVE
104 int real_cpu = cpu + vcpu->arch.ptid;
105 if (paca[real_cpu].kvm_hstate.xics_phys)
106 xics_wake_cpu(real_cpu);
110 smp_send_reschedule(cpu);
116 * We use the vcpu_load/put functions to measure stolen time.
117 * Stolen time is counted as time when either the vcpu is able to
118 * run as part of a virtual core, but the task running the vcore
119 * is preempted or sleeping, or when the vcpu needs something done
120 * in the kernel by the task running the vcpu, but that task is
121 * preempted or sleeping. Those two things have to be counted
122 * separately, since one of the vcpu tasks will take on the job
123 * of running the core, and the other vcpu tasks in the vcore will
124 * sleep waiting for it to do that, but that sleep shouldn't count
127 * Hence we accumulate stolen time when the vcpu can run as part of
128 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
129 * needs its task to do other things in the kernel (for example,
130 * service a page fault) in busy_stolen. We don't accumulate
131 * stolen time for a vcore when it is inactive, or for a vcpu
132 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
133 * a misnomer; it means that the vcpu task is not executing in
134 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
135 * the kernel. We don't have any way of dividing up that time
136 * between time that the vcpu is genuinely stopped, time that
137 * the task is actively working on behalf of the vcpu, and time
138 * that the task is preempted, so we don't count any of it as
141 * Updates to busy_stolen are protected by arch.tbacct_lock;
142 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
143 * lock. The stolen times are measured in units of timebase ticks.
144 * (Note that the != TB_NIL checks below are purely defensive;
145 * they should never fail.)
148 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
150 struct kvmppc_vcore *vc = vcpu->arch.vcore;
154 * We can test vc->runner without taking the vcore lock,
155 * because only this task ever sets vc->runner to this
156 * vcpu, and once it is set to this vcpu, only this task
157 * ever sets it to NULL.
159 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
160 spin_lock_irqsave(&vc->stoltb_lock, flags);
161 if (vc->preempt_tb != TB_NIL) {
162 vc->stolen_tb += mftb() - vc->preempt_tb;
163 vc->preempt_tb = TB_NIL;
165 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
167 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
168 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
169 vcpu->arch.busy_preempt != TB_NIL) {
170 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
171 vcpu->arch.busy_preempt = TB_NIL;
173 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
176 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
178 struct kvmppc_vcore *vc = vcpu->arch.vcore;
181 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
182 spin_lock_irqsave(&vc->stoltb_lock, flags);
183 vc->preempt_tb = mftb();
184 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
186 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
187 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
188 vcpu->arch.busy_preempt = mftb();
189 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
192 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
194 vcpu->arch.shregs.msr = msr;
195 kvmppc_end_cede(vcpu);
198 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
200 vcpu->arch.pvr = pvr;
203 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
205 unsigned long pcr = 0;
206 struct kvmppc_vcore *vc = vcpu->arch.vcore;
209 switch (arch_compat) {
212 * If an arch bit is set in PCR, all the defined
213 * higher-order arch bits also have to be set.
215 pcr = PCR_ARCH_206 | PCR_ARCH_205;
227 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
228 /* POWER7 can't emulate POWER8 */
229 if (!(pcr & PCR_ARCH_206))
231 pcr &= ~PCR_ARCH_206;
235 spin_lock(&vc->lock);
236 vc->arch_compat = arch_compat;
238 spin_unlock(&vc->lock);
243 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
247 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
248 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
249 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
250 for (r = 0; r < 16; ++r)
251 pr_err("r%2d = %.16lx r%d = %.16lx\n",
252 r, kvmppc_get_gpr(vcpu, r),
253 r+16, kvmppc_get_gpr(vcpu, r+16));
254 pr_err("ctr = %.16lx lr = %.16lx\n",
255 vcpu->arch.ctr, vcpu->arch.lr);
256 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
257 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
258 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
259 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
260 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
261 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
262 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
263 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
264 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
265 pr_err("fault dar = %.16lx dsisr = %.8x\n",
266 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
267 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
268 for (r = 0; r < vcpu->arch.slb_max; ++r)
269 pr_err(" ESID = %.16llx VSID = %.16llx\n",
270 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
271 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
272 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
273 vcpu->arch.last_inst);
276 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
279 struct kvm_vcpu *v, *ret = NULL;
281 mutex_lock(&kvm->lock);
282 kvm_for_each_vcpu(r, v, kvm) {
283 if (v->vcpu_id == id) {
288 mutex_unlock(&kvm->lock);
292 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
294 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
295 vpa->yield_count = cpu_to_be32(1);
298 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
299 unsigned long addr, unsigned long len)
301 /* check address is cacheline aligned */
302 if (addr & (L1_CACHE_BYTES - 1))
304 spin_lock(&vcpu->arch.vpa_update_lock);
305 if (v->next_gpa != addr || v->len != len) {
307 v->len = addr ? len : 0;
308 v->update_pending = 1;
310 spin_unlock(&vcpu->arch.vpa_update_lock);
314 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
323 static int vpa_is_registered(struct kvmppc_vpa *vpap)
325 if (vpap->update_pending)
326 return vpap->next_gpa != 0;
327 return vpap->pinned_addr != NULL;
330 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
332 unsigned long vcpuid, unsigned long vpa)
334 struct kvm *kvm = vcpu->kvm;
335 unsigned long len, nb;
337 struct kvm_vcpu *tvcpu;
340 struct kvmppc_vpa *vpap;
342 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
346 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
347 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
348 subfunc == H_VPA_REG_SLB) {
349 /* Registering new area - address must be cache-line aligned */
350 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
353 /* convert logical addr to kernel addr and read length */
354 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
357 if (subfunc == H_VPA_REG_VPA)
358 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
360 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
361 kvmppc_unpin_guest_page(kvm, va, vpa, false);
364 if (len > nb || len < sizeof(struct reg_vpa))
373 spin_lock(&tvcpu->arch.vpa_update_lock);
376 case H_VPA_REG_VPA: /* register VPA */
377 if (len < sizeof(struct lppaca))
379 vpap = &tvcpu->arch.vpa;
383 case H_VPA_REG_DTL: /* register DTL */
384 if (len < sizeof(struct dtl_entry))
386 len -= len % sizeof(struct dtl_entry);
388 /* Check that they have previously registered a VPA */
390 if (!vpa_is_registered(&tvcpu->arch.vpa))
393 vpap = &tvcpu->arch.dtl;
397 case H_VPA_REG_SLB: /* register SLB shadow buffer */
398 /* Check that they have previously registered a VPA */
400 if (!vpa_is_registered(&tvcpu->arch.vpa))
403 vpap = &tvcpu->arch.slb_shadow;
407 case H_VPA_DEREG_VPA: /* deregister VPA */
408 /* Check they don't still have a DTL or SLB buf registered */
410 if (vpa_is_registered(&tvcpu->arch.dtl) ||
411 vpa_is_registered(&tvcpu->arch.slb_shadow))
414 vpap = &tvcpu->arch.vpa;
418 case H_VPA_DEREG_DTL: /* deregister DTL */
419 vpap = &tvcpu->arch.dtl;
423 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
424 vpap = &tvcpu->arch.slb_shadow;
430 vpap->next_gpa = vpa;
432 vpap->update_pending = 1;
435 spin_unlock(&tvcpu->arch.vpa_update_lock);
440 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
442 struct kvm *kvm = vcpu->kvm;
448 * We need to pin the page pointed to by vpap->next_gpa,
449 * but we can't call kvmppc_pin_guest_page under the lock
450 * as it does get_user_pages() and down_read(). So we
451 * have to drop the lock, pin the page, then get the lock
452 * again and check that a new area didn't get registered
456 gpa = vpap->next_gpa;
457 spin_unlock(&vcpu->arch.vpa_update_lock);
461 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
462 spin_lock(&vcpu->arch.vpa_update_lock);
463 if (gpa == vpap->next_gpa)
465 /* sigh... unpin that one and try again */
467 kvmppc_unpin_guest_page(kvm, va, gpa, false);
470 vpap->update_pending = 0;
471 if (va && nb < vpap->len) {
473 * If it's now too short, it must be that userspace
474 * has changed the mappings underlying guest memory,
475 * so unregister the region.
477 kvmppc_unpin_guest_page(kvm, va, gpa, false);
480 if (vpap->pinned_addr)
481 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
484 vpap->pinned_addr = va;
487 vpap->pinned_end = va + vpap->len;
490 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
492 if (!(vcpu->arch.vpa.update_pending ||
493 vcpu->arch.slb_shadow.update_pending ||
494 vcpu->arch.dtl.update_pending))
497 spin_lock(&vcpu->arch.vpa_update_lock);
498 if (vcpu->arch.vpa.update_pending) {
499 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
500 if (vcpu->arch.vpa.pinned_addr)
501 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
503 if (vcpu->arch.dtl.update_pending) {
504 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
505 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
506 vcpu->arch.dtl_index = 0;
508 if (vcpu->arch.slb_shadow.update_pending)
509 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
510 spin_unlock(&vcpu->arch.vpa_update_lock);
514 * Return the accumulated stolen time for the vcore up until `now'.
515 * The caller should hold the vcore lock.
517 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
522 spin_lock_irqsave(&vc->stoltb_lock, flags);
524 if (vc->vcore_state != VCORE_INACTIVE &&
525 vc->preempt_tb != TB_NIL)
526 p += now - vc->preempt_tb;
527 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
531 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
532 struct kvmppc_vcore *vc)
534 struct dtl_entry *dt;
536 unsigned long stolen;
537 unsigned long core_stolen;
540 dt = vcpu->arch.dtl_ptr;
541 vpa = vcpu->arch.vpa.pinned_addr;
543 core_stolen = vcore_stolen_time(vc, now);
544 stolen = core_stolen - vcpu->arch.stolen_logged;
545 vcpu->arch.stolen_logged = core_stolen;
546 spin_lock_irq(&vcpu->arch.tbacct_lock);
547 stolen += vcpu->arch.busy_stolen;
548 vcpu->arch.busy_stolen = 0;
549 spin_unlock_irq(&vcpu->arch.tbacct_lock);
552 memset(dt, 0, sizeof(struct dtl_entry));
553 dt->dispatch_reason = 7;
554 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
555 dt->timebase = cpu_to_be64(now + vc->tb_offset);
556 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
557 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
558 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
560 if (dt == vcpu->arch.dtl.pinned_end)
561 dt = vcpu->arch.dtl.pinned_addr;
562 vcpu->arch.dtl_ptr = dt;
563 /* order writing *dt vs. writing vpa->dtl_idx */
565 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
566 vcpu->arch.dtl.dirty = true;
569 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
571 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
573 if ((!vcpu->arch.vcore->arch_compat) &&
574 cpu_has_feature(CPU_FTR_ARCH_207S))
579 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
580 unsigned long resource, unsigned long value1,
581 unsigned long value2)
584 case H_SET_MODE_RESOURCE_SET_CIABR:
585 if (!kvmppc_power8_compatible(vcpu))
590 return H_UNSUPPORTED_FLAG_START;
591 /* Guests can't breakpoint the hypervisor */
592 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
594 vcpu->arch.ciabr = value1;
596 case H_SET_MODE_RESOURCE_SET_DAWR:
597 if (!kvmppc_power8_compatible(vcpu))
600 return H_UNSUPPORTED_FLAG_START;
601 if (value2 & DABRX_HYP)
603 vcpu->arch.dawr = value1;
604 vcpu->arch.dawrx = value2;
611 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
613 struct kvmppc_vcore *vcore = target->arch.vcore;
616 * We expect to have been called by the real mode handler
617 * (kvmppc_rm_h_confer()) which would have directly returned
618 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
619 * have useful work to do and should not confer) so we don't
623 spin_lock(&vcore->lock);
624 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
625 vcore->vcore_state != VCORE_INACTIVE)
626 target = vcore->runner;
627 spin_unlock(&vcore->lock);
629 return kvm_vcpu_yield_to(target);
632 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
635 struct lppaca *lppaca;
637 spin_lock(&vcpu->arch.vpa_update_lock);
638 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
640 yield_count = be32_to_cpu(lppaca->yield_count);
641 spin_unlock(&vcpu->arch.vpa_update_lock);
645 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
647 unsigned long req = kvmppc_get_gpr(vcpu, 3);
648 unsigned long target, ret = H_SUCCESS;
650 struct kvm_vcpu *tvcpu;
653 if (req <= MAX_HCALL_OPCODE &&
654 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
661 target = kvmppc_get_gpr(vcpu, 4);
662 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
667 tvcpu->arch.prodded = 1;
669 if (vcpu->arch.ceded) {
670 if (waitqueue_active(&vcpu->wq)) {
671 wake_up_interruptible(&vcpu->wq);
672 vcpu->stat.halt_wakeup++;
677 target = kvmppc_get_gpr(vcpu, 4);
680 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
685 yield_count = kvmppc_get_gpr(vcpu, 5);
686 if (kvmppc_get_yield_count(tvcpu) != yield_count)
688 kvm_arch_vcpu_yield_to(tvcpu);
691 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
692 kvmppc_get_gpr(vcpu, 5),
693 kvmppc_get_gpr(vcpu, 6));
696 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
699 idx = srcu_read_lock(&vcpu->kvm->srcu);
700 rc = kvmppc_rtas_hcall(vcpu);
701 srcu_read_unlock(&vcpu->kvm->srcu, idx);
708 /* Send the error out to userspace via KVM_RUN */
710 case H_LOGICAL_CI_LOAD:
711 ret = kvmppc_h_logical_ci_load(vcpu);
712 if (ret == H_TOO_HARD)
715 case H_LOGICAL_CI_STORE:
716 ret = kvmppc_h_logical_ci_store(vcpu);
717 if (ret == H_TOO_HARD)
721 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
722 kvmppc_get_gpr(vcpu, 5),
723 kvmppc_get_gpr(vcpu, 6),
724 kvmppc_get_gpr(vcpu, 7));
725 if (ret == H_TOO_HARD)
734 if (kvmppc_xics_enabled(vcpu)) {
735 ret = kvmppc_xics_hcall(vcpu, req);
741 kvmppc_set_gpr(vcpu, 3, ret);
742 vcpu->arch.hcall_needed = 0;
746 static int kvmppc_hcall_impl_hv(unsigned long cmd)
754 case H_LOGICAL_CI_LOAD:
755 case H_LOGICAL_CI_STORE:
756 #ifdef CONFIG_KVM_XICS
767 /* See if it's in the real-mode table */
768 return kvmppc_hcall_impl_hv_realmode(cmd);
771 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
772 struct kvm_vcpu *vcpu)
776 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
779 * Fetch failed, so return to guest and
780 * try executing it again.
785 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
786 run->exit_reason = KVM_EXIT_DEBUG;
787 run->debug.arch.address = kvmppc_get_pc(vcpu);
790 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
795 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
796 struct task_struct *tsk)
800 vcpu->stat.sum_exits++;
802 run->exit_reason = KVM_EXIT_UNKNOWN;
803 run->ready_for_interrupt_injection = 1;
804 switch (vcpu->arch.trap) {
805 /* We're good on these - the host merely wanted to get our attention */
806 case BOOK3S_INTERRUPT_HV_DECREMENTER:
807 vcpu->stat.dec_exits++;
810 case BOOK3S_INTERRUPT_EXTERNAL:
811 case BOOK3S_INTERRUPT_H_DOORBELL:
812 vcpu->stat.ext_intr_exits++;
815 /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
816 case BOOK3S_INTERRUPT_HMI:
817 case BOOK3S_INTERRUPT_PERFMON:
820 case BOOK3S_INTERRUPT_MACHINE_CHECK:
822 * Deliver a machine check interrupt to the guest.
823 * We have to do this, even if the host has handled the
824 * machine check, because machine checks use SRR0/1 and
825 * the interrupt might have trashed guest state in them.
827 kvmppc_book3s_queue_irqprio(vcpu,
828 BOOK3S_INTERRUPT_MACHINE_CHECK);
831 case BOOK3S_INTERRUPT_PROGRAM:
835 * Normally program interrupts are delivered directly
836 * to the guest by the hardware, but we can get here
837 * as a result of a hypervisor emulation interrupt
838 * (e40) getting turned into a 700 by BML RTAS.
840 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
841 kvmppc_core_queue_program(vcpu, flags);
845 case BOOK3S_INTERRUPT_SYSCALL:
847 /* hcall - punt to userspace */
850 /* hypercall with MSR_PR has already been handled in rmode,
851 * and never reaches here.
854 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
855 for (i = 0; i < 9; ++i)
856 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
857 run->exit_reason = KVM_EXIT_PAPR_HCALL;
858 vcpu->arch.hcall_needed = 1;
863 * We get these next two if the guest accesses a page which it thinks
864 * it has mapped but which is not actually present, either because
865 * it is for an emulated I/O device or because the corresonding
866 * host page has been paged out. Any other HDSI/HISI interrupts
867 * have been handled already.
869 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
870 r = RESUME_PAGE_FAULT;
872 case BOOK3S_INTERRUPT_H_INST_STORAGE:
873 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
874 vcpu->arch.fault_dsisr = 0;
875 r = RESUME_PAGE_FAULT;
878 * This occurs if the guest executes an illegal instruction.
879 * If the guest debug is disabled, generate a program interrupt
880 * to the guest. If guest debug is enabled, we need to check
881 * whether the instruction is a software breakpoint instruction.
882 * Accordingly return to Guest or Host.
884 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
885 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
886 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
887 swab32(vcpu->arch.emul_inst) :
888 vcpu->arch.emul_inst;
889 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
890 r = kvmppc_emulate_debug_inst(run, vcpu);
892 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
897 * This occurs if the guest (kernel or userspace), does something that
898 * is prohibited by HFSCR. We just generate a program interrupt to
901 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
902 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
906 kvmppc_dump_regs(vcpu);
907 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
908 vcpu->arch.trap, kvmppc_get_pc(vcpu),
909 vcpu->arch.shregs.msr);
910 run->hw.hardware_exit_reason = vcpu->arch.trap;
918 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
919 struct kvm_sregs *sregs)
923 memset(sregs, 0, sizeof(struct kvm_sregs));
924 sregs->pvr = vcpu->arch.pvr;
925 for (i = 0; i < vcpu->arch.slb_max; i++) {
926 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
927 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
933 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
934 struct kvm_sregs *sregs)
938 /* Only accept the same PVR as the host's, since we can't spoof it */
939 if (sregs->pvr != vcpu->arch.pvr)
943 for (i = 0; i < vcpu->arch.slb_nr; i++) {
944 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
945 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
946 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
950 vcpu->arch.slb_max = j;
955 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
958 struct kvm *kvm = vcpu->kvm;
959 struct kvmppc_vcore *vc = vcpu->arch.vcore;
962 mutex_lock(&kvm->lock);
963 spin_lock(&vc->lock);
965 * If ILE (interrupt little-endian) has changed, update the
966 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
968 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
969 struct kvm_vcpu *vcpu;
972 kvm_for_each_vcpu(i, vcpu, kvm) {
973 if (vcpu->arch.vcore != vc)
975 if (new_lpcr & LPCR_ILE)
976 vcpu->arch.intr_msr |= MSR_LE;
978 vcpu->arch.intr_msr &= ~MSR_LE;
983 * Userspace can only modify DPFD (default prefetch depth),
984 * ILE (interrupt little-endian) and TC (translation control).
985 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
987 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
988 if (cpu_has_feature(CPU_FTR_ARCH_207S))
991 /* Broken 32-bit version of LPCR must not clear top bits */
994 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
995 spin_unlock(&vc->lock);
996 mutex_unlock(&kvm->lock);
999 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1000 union kvmppc_one_reg *val)
1006 case KVM_REG_PPC_DEBUG_INST:
1007 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1009 case KVM_REG_PPC_HIOR:
1010 *val = get_reg_val(id, 0);
1012 case KVM_REG_PPC_DABR:
1013 *val = get_reg_val(id, vcpu->arch.dabr);
1015 case KVM_REG_PPC_DABRX:
1016 *val = get_reg_val(id, vcpu->arch.dabrx);
1018 case KVM_REG_PPC_DSCR:
1019 *val = get_reg_val(id, vcpu->arch.dscr);
1021 case KVM_REG_PPC_PURR:
1022 *val = get_reg_val(id, vcpu->arch.purr);
1024 case KVM_REG_PPC_SPURR:
1025 *val = get_reg_val(id, vcpu->arch.spurr);
1027 case KVM_REG_PPC_AMR:
1028 *val = get_reg_val(id, vcpu->arch.amr);
1030 case KVM_REG_PPC_UAMOR:
1031 *val = get_reg_val(id, vcpu->arch.uamor);
1033 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1034 i = id - KVM_REG_PPC_MMCR0;
1035 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1037 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1038 i = id - KVM_REG_PPC_PMC1;
1039 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1041 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1042 i = id - KVM_REG_PPC_SPMC1;
1043 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1045 case KVM_REG_PPC_SIAR:
1046 *val = get_reg_val(id, vcpu->arch.siar);
1048 case KVM_REG_PPC_SDAR:
1049 *val = get_reg_val(id, vcpu->arch.sdar);
1051 case KVM_REG_PPC_SIER:
1052 *val = get_reg_val(id, vcpu->arch.sier);
1054 case KVM_REG_PPC_IAMR:
1055 *val = get_reg_val(id, vcpu->arch.iamr);
1057 case KVM_REG_PPC_PSPB:
1058 *val = get_reg_val(id, vcpu->arch.pspb);
1060 case KVM_REG_PPC_DPDES:
1061 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1063 case KVM_REG_PPC_DAWR:
1064 *val = get_reg_val(id, vcpu->arch.dawr);
1066 case KVM_REG_PPC_DAWRX:
1067 *val = get_reg_val(id, vcpu->arch.dawrx);
1069 case KVM_REG_PPC_CIABR:
1070 *val = get_reg_val(id, vcpu->arch.ciabr);
1072 case KVM_REG_PPC_CSIGR:
1073 *val = get_reg_val(id, vcpu->arch.csigr);
1075 case KVM_REG_PPC_TACR:
1076 *val = get_reg_val(id, vcpu->arch.tacr);
1078 case KVM_REG_PPC_TCSCR:
1079 *val = get_reg_val(id, vcpu->arch.tcscr);
1081 case KVM_REG_PPC_PID:
1082 *val = get_reg_val(id, vcpu->arch.pid);
1084 case KVM_REG_PPC_ACOP:
1085 *val = get_reg_val(id, vcpu->arch.acop);
1087 case KVM_REG_PPC_WORT:
1088 *val = get_reg_val(id, vcpu->arch.wort);
1090 case KVM_REG_PPC_VPA_ADDR:
1091 spin_lock(&vcpu->arch.vpa_update_lock);
1092 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1093 spin_unlock(&vcpu->arch.vpa_update_lock);
1095 case KVM_REG_PPC_VPA_SLB:
1096 spin_lock(&vcpu->arch.vpa_update_lock);
1097 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1098 val->vpaval.length = vcpu->arch.slb_shadow.len;
1099 spin_unlock(&vcpu->arch.vpa_update_lock);
1101 case KVM_REG_PPC_VPA_DTL:
1102 spin_lock(&vcpu->arch.vpa_update_lock);
1103 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1104 val->vpaval.length = vcpu->arch.dtl.len;
1105 spin_unlock(&vcpu->arch.vpa_update_lock);
1107 case KVM_REG_PPC_TB_OFFSET:
1108 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1110 case KVM_REG_PPC_LPCR:
1111 case KVM_REG_PPC_LPCR_64:
1112 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1114 case KVM_REG_PPC_PPR:
1115 *val = get_reg_val(id, vcpu->arch.ppr);
1117 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1118 case KVM_REG_PPC_TFHAR:
1119 *val = get_reg_val(id, vcpu->arch.tfhar);
1121 case KVM_REG_PPC_TFIAR:
1122 *val = get_reg_val(id, vcpu->arch.tfiar);
1124 case KVM_REG_PPC_TEXASR:
1125 *val = get_reg_val(id, vcpu->arch.texasr);
1127 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1128 i = id - KVM_REG_PPC_TM_GPR0;
1129 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1131 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1134 i = id - KVM_REG_PPC_TM_VSR0;
1136 for (j = 0; j < TS_FPRWIDTH; j++)
1137 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1139 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1140 val->vval = vcpu->arch.vr_tm.vr[i-32];
1146 case KVM_REG_PPC_TM_CR:
1147 *val = get_reg_val(id, vcpu->arch.cr_tm);
1149 case KVM_REG_PPC_TM_LR:
1150 *val = get_reg_val(id, vcpu->arch.lr_tm);
1152 case KVM_REG_PPC_TM_CTR:
1153 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1155 case KVM_REG_PPC_TM_FPSCR:
1156 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1158 case KVM_REG_PPC_TM_AMR:
1159 *val = get_reg_val(id, vcpu->arch.amr_tm);
1161 case KVM_REG_PPC_TM_PPR:
1162 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1164 case KVM_REG_PPC_TM_VRSAVE:
1165 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1167 case KVM_REG_PPC_TM_VSCR:
1168 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1169 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1173 case KVM_REG_PPC_TM_DSCR:
1174 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1176 case KVM_REG_PPC_TM_TAR:
1177 *val = get_reg_val(id, vcpu->arch.tar_tm);
1180 case KVM_REG_PPC_ARCH_COMPAT:
1181 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1191 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1192 union kvmppc_one_reg *val)
1196 unsigned long addr, len;
1199 case KVM_REG_PPC_HIOR:
1200 /* Only allow this to be set to zero */
1201 if (set_reg_val(id, *val))
1204 case KVM_REG_PPC_DABR:
1205 vcpu->arch.dabr = set_reg_val(id, *val);
1207 case KVM_REG_PPC_DABRX:
1208 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1210 case KVM_REG_PPC_DSCR:
1211 vcpu->arch.dscr = set_reg_val(id, *val);
1213 case KVM_REG_PPC_PURR:
1214 vcpu->arch.purr = set_reg_val(id, *val);
1216 case KVM_REG_PPC_SPURR:
1217 vcpu->arch.spurr = set_reg_val(id, *val);
1219 case KVM_REG_PPC_AMR:
1220 vcpu->arch.amr = set_reg_val(id, *val);
1222 case KVM_REG_PPC_UAMOR:
1223 vcpu->arch.uamor = set_reg_val(id, *val);
1225 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1226 i = id - KVM_REG_PPC_MMCR0;
1227 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1229 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1230 i = id - KVM_REG_PPC_PMC1;
1231 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1233 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1234 i = id - KVM_REG_PPC_SPMC1;
1235 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1237 case KVM_REG_PPC_SIAR:
1238 vcpu->arch.siar = set_reg_val(id, *val);
1240 case KVM_REG_PPC_SDAR:
1241 vcpu->arch.sdar = set_reg_val(id, *val);
1243 case KVM_REG_PPC_SIER:
1244 vcpu->arch.sier = set_reg_val(id, *val);
1246 case KVM_REG_PPC_IAMR:
1247 vcpu->arch.iamr = set_reg_val(id, *val);
1249 case KVM_REG_PPC_PSPB:
1250 vcpu->arch.pspb = set_reg_val(id, *val);
1252 case KVM_REG_PPC_DPDES:
1253 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1255 case KVM_REG_PPC_DAWR:
1256 vcpu->arch.dawr = set_reg_val(id, *val);
1258 case KVM_REG_PPC_DAWRX:
1259 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1261 case KVM_REG_PPC_CIABR:
1262 vcpu->arch.ciabr = set_reg_val(id, *val);
1263 /* Don't allow setting breakpoints in hypervisor code */
1264 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1265 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1267 case KVM_REG_PPC_CSIGR:
1268 vcpu->arch.csigr = set_reg_val(id, *val);
1270 case KVM_REG_PPC_TACR:
1271 vcpu->arch.tacr = set_reg_val(id, *val);
1273 case KVM_REG_PPC_TCSCR:
1274 vcpu->arch.tcscr = set_reg_val(id, *val);
1276 case KVM_REG_PPC_PID:
1277 vcpu->arch.pid = set_reg_val(id, *val);
1279 case KVM_REG_PPC_ACOP:
1280 vcpu->arch.acop = set_reg_val(id, *val);
1282 case KVM_REG_PPC_WORT:
1283 vcpu->arch.wort = set_reg_val(id, *val);
1285 case KVM_REG_PPC_VPA_ADDR:
1286 addr = set_reg_val(id, *val);
1288 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1289 vcpu->arch.dtl.next_gpa))
1291 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1293 case KVM_REG_PPC_VPA_SLB:
1294 addr = val->vpaval.addr;
1295 len = val->vpaval.length;
1297 if (addr && !vcpu->arch.vpa.next_gpa)
1299 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1301 case KVM_REG_PPC_VPA_DTL:
1302 addr = val->vpaval.addr;
1303 len = val->vpaval.length;
1305 if (addr && (len < sizeof(struct dtl_entry) ||
1306 !vcpu->arch.vpa.next_gpa))
1308 len -= len % sizeof(struct dtl_entry);
1309 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1311 case KVM_REG_PPC_TB_OFFSET:
1312 /* round up to multiple of 2^24 */
1313 vcpu->arch.vcore->tb_offset =
1314 ALIGN(set_reg_val(id, *val), 1UL << 24);
1316 case KVM_REG_PPC_LPCR:
1317 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1319 case KVM_REG_PPC_LPCR_64:
1320 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1322 case KVM_REG_PPC_PPR:
1323 vcpu->arch.ppr = set_reg_val(id, *val);
1325 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1326 case KVM_REG_PPC_TFHAR:
1327 vcpu->arch.tfhar = set_reg_val(id, *val);
1329 case KVM_REG_PPC_TFIAR:
1330 vcpu->arch.tfiar = set_reg_val(id, *val);
1332 case KVM_REG_PPC_TEXASR:
1333 vcpu->arch.texasr = set_reg_val(id, *val);
1335 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1336 i = id - KVM_REG_PPC_TM_GPR0;
1337 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1339 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1342 i = id - KVM_REG_PPC_TM_VSR0;
1344 for (j = 0; j < TS_FPRWIDTH; j++)
1345 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1347 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1348 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1353 case KVM_REG_PPC_TM_CR:
1354 vcpu->arch.cr_tm = set_reg_val(id, *val);
1356 case KVM_REG_PPC_TM_LR:
1357 vcpu->arch.lr_tm = set_reg_val(id, *val);
1359 case KVM_REG_PPC_TM_CTR:
1360 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1362 case KVM_REG_PPC_TM_FPSCR:
1363 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1365 case KVM_REG_PPC_TM_AMR:
1366 vcpu->arch.amr_tm = set_reg_val(id, *val);
1368 case KVM_REG_PPC_TM_PPR:
1369 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1371 case KVM_REG_PPC_TM_VRSAVE:
1372 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1374 case KVM_REG_PPC_TM_VSCR:
1375 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1376 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1380 case KVM_REG_PPC_TM_DSCR:
1381 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1383 case KVM_REG_PPC_TM_TAR:
1384 vcpu->arch.tar_tm = set_reg_val(id, *val);
1387 case KVM_REG_PPC_ARCH_COMPAT:
1388 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1398 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1400 struct kvmppc_vcore *vcore;
1402 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1407 INIT_LIST_HEAD(&vcore->runnable_threads);
1408 spin_lock_init(&vcore->lock);
1409 spin_lock_init(&vcore->stoltb_lock);
1410 init_waitqueue_head(&vcore->wq);
1411 vcore->preempt_tb = TB_NIL;
1412 vcore->lpcr = kvm->arch.lpcr;
1413 vcore->first_vcpuid = core * threads_per_subcore;
1416 vcore->mpp_buffer_is_valid = false;
1418 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1419 vcore->mpp_buffer = (void *)__get_free_pages(
1420 GFP_KERNEL|__GFP_ZERO,
1426 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1427 static struct debugfs_timings_element {
1431 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1432 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1433 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1434 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1435 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1438 #define N_TIMINGS (sizeof(timings) / sizeof(timings[0]))
1440 struct debugfs_timings_state {
1441 struct kvm_vcpu *vcpu;
1442 unsigned int buflen;
1443 char buf[N_TIMINGS * 100];
1446 static int debugfs_timings_open(struct inode *inode, struct file *file)
1448 struct kvm_vcpu *vcpu = inode->i_private;
1449 struct debugfs_timings_state *p;
1451 p = kzalloc(sizeof(*p), GFP_KERNEL);
1455 kvm_get_kvm(vcpu->kvm);
1457 file->private_data = p;
1459 return nonseekable_open(inode, file);
1462 static int debugfs_timings_release(struct inode *inode, struct file *file)
1464 struct debugfs_timings_state *p = file->private_data;
1466 kvm_put_kvm(p->vcpu->kvm);
1471 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1472 size_t len, loff_t *ppos)
1474 struct debugfs_timings_state *p = file->private_data;
1475 struct kvm_vcpu *vcpu = p->vcpu;
1477 struct kvmhv_tb_accumulator tb;
1486 buf_end = s + sizeof(p->buf);
1487 for (i = 0; i < N_TIMINGS; ++i) {
1488 struct kvmhv_tb_accumulator *acc;
1490 acc = (struct kvmhv_tb_accumulator *)
1491 ((unsigned long)vcpu + timings[i].offset);
1493 for (loops = 0; loops < 1000; ++loops) {
1494 count = acc->seqcount;
1499 if (count == acc->seqcount) {
1507 snprintf(s, buf_end - s, "%s: stuck\n",
1510 snprintf(s, buf_end - s,
1511 "%s: %llu %llu %llu %llu\n",
1512 timings[i].name, count / 2,
1513 tb_to_ns(tb.tb_total),
1514 tb_to_ns(tb.tb_min),
1515 tb_to_ns(tb.tb_max));
1518 p->buflen = s - p->buf;
1522 if (pos >= p->buflen)
1524 if (len > p->buflen - pos)
1525 len = p->buflen - pos;
1526 n = copy_to_user(buf, p->buf + pos, len);
1536 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1537 size_t len, loff_t *ppos)
1542 static const struct file_operations debugfs_timings_ops = {
1543 .owner = THIS_MODULE,
1544 .open = debugfs_timings_open,
1545 .release = debugfs_timings_release,
1546 .read = debugfs_timings_read,
1547 .write = debugfs_timings_write,
1548 .llseek = generic_file_llseek,
1551 /* Create a debugfs directory for the vcpu */
1552 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1555 struct kvm *kvm = vcpu->kvm;
1557 snprintf(buf, sizeof(buf), "vcpu%u", id);
1558 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1560 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1561 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1563 vcpu->arch.debugfs_timings =
1564 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1565 vcpu, &debugfs_timings_ops);
1568 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1569 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1572 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1574 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1577 struct kvm_vcpu *vcpu;
1580 struct kvmppc_vcore *vcore;
1582 core = id / threads_per_subcore;
1583 if (core >= KVM_MAX_VCORES)
1587 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1591 err = kvm_vcpu_init(vcpu, kvm, id);
1595 vcpu->arch.shared = &vcpu->arch.shregs;
1596 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1598 * The shared struct is never shared on HV,
1599 * so we can always use host endianness
1601 #ifdef __BIG_ENDIAN__
1602 vcpu->arch.shared_big_endian = true;
1604 vcpu->arch.shared_big_endian = false;
1607 vcpu->arch.mmcr[0] = MMCR0_FC;
1608 vcpu->arch.ctrl = CTRL_RUNLATCH;
1609 /* default to host PVR, since we can't spoof it */
1610 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1611 spin_lock_init(&vcpu->arch.vpa_update_lock);
1612 spin_lock_init(&vcpu->arch.tbacct_lock);
1613 vcpu->arch.busy_preempt = TB_NIL;
1614 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1616 kvmppc_mmu_book3s_hv_init(vcpu);
1618 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1620 init_waitqueue_head(&vcpu->arch.cpu_run);
1622 mutex_lock(&kvm->lock);
1623 vcore = kvm->arch.vcores[core];
1625 vcore = kvmppc_vcore_create(kvm, core);
1626 kvm->arch.vcores[core] = vcore;
1627 kvm->arch.online_vcores++;
1629 mutex_unlock(&kvm->lock);
1634 spin_lock(&vcore->lock);
1635 ++vcore->num_threads;
1636 spin_unlock(&vcore->lock);
1637 vcpu->arch.vcore = vcore;
1638 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1640 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1641 kvmppc_sanity_check(vcpu);
1643 debugfs_vcpu_init(vcpu, id);
1648 kmem_cache_free(kvm_vcpu_cache, vcpu);
1650 return ERR_PTR(err);
1653 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1655 if (vpa->pinned_addr)
1656 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1660 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1662 spin_lock(&vcpu->arch.vpa_update_lock);
1663 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1664 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1665 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1666 spin_unlock(&vcpu->arch.vpa_update_lock);
1667 kvm_vcpu_uninit(vcpu);
1668 kmem_cache_free(kvm_vcpu_cache, vcpu);
1671 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1673 /* Indicate we want to get back into the guest */
1677 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1679 unsigned long dec_nsec, now;
1682 if (now > vcpu->arch.dec_expires) {
1683 /* decrementer has already gone negative */
1684 kvmppc_core_queue_dec(vcpu);
1685 kvmppc_core_prepare_to_enter(vcpu);
1688 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1690 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1692 vcpu->arch.timer_running = 1;
1695 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1697 vcpu->arch.ceded = 0;
1698 if (vcpu->arch.timer_running) {
1699 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1700 vcpu->arch.timer_running = 0;
1704 extern void __kvmppc_vcore_entry(void);
1706 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1707 struct kvm_vcpu *vcpu)
1711 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1713 spin_lock_irq(&vcpu->arch.tbacct_lock);
1715 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1716 vcpu->arch.stolen_logged;
1717 vcpu->arch.busy_preempt = now;
1718 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1719 spin_unlock_irq(&vcpu->arch.tbacct_lock);
1721 list_del(&vcpu->arch.run_list);
1724 static int kvmppc_grab_hwthread(int cpu)
1726 struct paca_struct *tpaca;
1727 long timeout = 10000;
1731 /* Ensure the thread won't go into the kernel if it wakes */
1732 tpaca->kvm_hstate.hwthread_req = 1;
1733 tpaca->kvm_hstate.kvm_vcpu = NULL;
1736 * If the thread is already executing in the kernel (e.g. handling
1737 * a stray interrupt), wait for it to get back to nap mode.
1738 * The smp_mb() is to ensure that our setting of hwthread_req
1739 * is visible before we look at hwthread_state, so if this
1740 * races with the code at system_reset_pSeries and the thread
1741 * misses our setting of hwthread_req, we are sure to see its
1742 * setting of hwthread_state, and vice versa.
1745 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1746 if (--timeout <= 0) {
1747 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1755 static void kvmppc_release_hwthread(int cpu)
1757 struct paca_struct *tpaca;
1760 tpaca->kvm_hstate.hwthread_req = 0;
1761 tpaca->kvm_hstate.kvm_vcpu = NULL;
1764 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1767 struct paca_struct *tpaca;
1768 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1770 if (vcpu->arch.timer_running) {
1771 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1772 vcpu->arch.timer_running = 0;
1774 cpu = vc->pcpu + vcpu->arch.ptid;
1776 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1777 tpaca->kvm_hstate.kvm_vcore = vc;
1778 tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1779 vcpu->cpu = vc->pcpu;
1781 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1782 if (cpu != smp_processor_id()) {
1784 if (vcpu->arch.ptid)
1790 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1796 while (vc->nap_count < vc->n_woken) {
1797 if (++i >= 1000000) {
1798 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1799 vc->nap_count, vc->n_woken);
1808 * Check that we are on thread 0 and that any other threads in
1809 * this core are off-line. Then grab the threads so they can't
1812 static int on_primary_thread(void)
1814 int cpu = smp_processor_id();
1817 /* Are we on a primary subcore? */
1818 if (cpu_thread_in_subcore(cpu))
1822 while (++thr < threads_per_subcore)
1823 if (cpu_online(cpu + thr))
1826 /* Grab all hw threads so they can't go into the kernel */
1827 for (thr = 1; thr < threads_per_subcore; ++thr) {
1828 if (kvmppc_grab_hwthread(cpu + thr)) {
1829 /* Couldn't grab one; let the others go */
1831 kvmppc_release_hwthread(cpu + thr);
1832 } while (--thr > 0);
1839 static void kvmppc_start_saving_l2_cache(struct kvmppc_vcore *vc)
1841 phys_addr_t phy_addr, mpp_addr;
1843 phy_addr = (phys_addr_t)virt_to_phys(vc->mpp_buffer);
1844 mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1846 mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_ABORT);
1847 logmpp(mpp_addr | PPC_LOGMPP_LOG_L2);
1849 vc->mpp_buffer_is_valid = true;
1852 static void kvmppc_start_restoring_l2_cache(const struct kvmppc_vcore *vc)
1854 phys_addr_t phy_addr, mpp_addr;
1856 phy_addr = virt_to_phys(vc->mpp_buffer);
1857 mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1859 /* We must abort any in-progress save operations to ensure
1860 * the table is valid so that prefetch engine knows when to
1861 * stop prefetching. */
1862 logmpp(mpp_addr | PPC_LOGMPP_LOG_ABORT);
1863 mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_WHOLE_TABLE);
1866 static void prepare_threads(struct kvmppc_vcore *vc)
1868 struct kvm_vcpu *vcpu, *vnext;
1870 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1872 if (signal_pending(vcpu->arch.run_task))
1873 vcpu->arch.ret = -EINTR;
1874 else if (vcpu->arch.vpa.update_pending ||
1875 vcpu->arch.slb_shadow.update_pending ||
1876 vcpu->arch.dtl.update_pending)
1877 vcpu->arch.ret = RESUME_GUEST;
1880 kvmppc_remove_runnable(vc, vcpu);
1881 wake_up(&vcpu->arch.cpu_run);
1885 static void post_guest_process(struct kvmppc_vcore *vc)
1889 struct kvm_vcpu *vcpu, *vnext;
1892 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1894 /* cancel pending dec exception if dec is positive */
1895 if (now < vcpu->arch.dec_expires &&
1896 kvmppc_core_pending_dec(vcpu))
1897 kvmppc_core_dequeue_dec(vcpu);
1899 trace_kvm_guest_exit(vcpu);
1902 if (vcpu->arch.trap)
1903 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1904 vcpu->arch.run_task);
1906 vcpu->arch.ret = ret;
1907 vcpu->arch.trap = 0;
1909 if (vcpu->arch.ceded) {
1910 if (!is_kvmppc_resume_guest(ret))
1911 kvmppc_end_cede(vcpu);
1913 kvmppc_set_timer(vcpu);
1915 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1916 kvmppc_remove_runnable(vc, vcpu);
1917 wake_up(&vcpu->arch.cpu_run);
1923 * Run a set of guest threads on a physical core.
1924 * Called with vc->lock held.
1926 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1928 struct kvm_vcpu *vcpu;
1933 * Remove from the list any threads that have a signal pending
1934 * or need a VPA update done
1936 prepare_threads(vc);
1938 /* if the runner is no longer runnable, let the caller pick a new one */
1939 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
1947 vc->entry_exit_count = 0;
1948 vc->preempt_tb = TB_NIL;
1950 vc->napping_threads = 0;
1951 vc->conferring_threads = 0;
1954 * Make sure we are running on primary threads, and that secondary
1955 * threads are offline. Also check if the number of threads in this
1956 * guest are greater than the current system threads per guest.
1958 if ((threads_per_core > 1) &&
1959 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1960 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1961 vcpu->arch.ret = -EBUSY;
1962 kvmppc_remove_runnable(vc, vcpu);
1963 wake_up(&vcpu->arch.cpu_run);
1969 vc->pcpu = smp_processor_id();
1970 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1971 kvmppc_start_thread(vcpu);
1972 kvmppc_create_dtl_entry(vcpu, vc);
1973 trace_kvm_guest_enter(vcpu);
1976 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1977 get_paca()->kvm_hstate.kvm_vcore = vc;
1978 get_paca()->kvm_hstate.ptid = 0;
1980 vc->vcore_state = VCORE_RUNNING;
1983 trace_kvmppc_run_core(vc, 0);
1985 spin_unlock(&vc->lock);
1989 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1991 if (vc->mpp_buffer_is_valid)
1992 kvmppc_start_restoring_l2_cache(vc);
1994 __kvmppc_vcore_entry();
1996 spin_lock(&vc->lock);
1999 kvmppc_start_saving_l2_cache(vc);
2001 /* disable sending of IPIs on virtual external irqs */
2002 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
2004 /* wait for secondary threads to finish writing their state to memory */
2005 if (vc->nap_count < vc->n_woken)
2006 kvmppc_wait_for_nap(vc);
2007 for (i = 0; i < threads_per_subcore; ++i)
2008 kvmppc_release_hwthread(vc->pcpu + i);
2009 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2010 vc->vcore_state = VCORE_EXITING;
2011 spin_unlock(&vc->lock);
2013 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2015 /* make sure updates to secondary vcpu structs are visible now */
2021 spin_lock(&vc->lock);
2022 post_guest_process(vc);
2025 vc->vcore_state = VCORE_INACTIVE;
2026 trace_kvmppc_run_core(vc, 1);
2030 * Wait for some other vcpu thread to execute us, and
2031 * wake us up when we need to handle something in the host.
2033 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
2037 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2038 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
2040 finish_wait(&vcpu->arch.cpu_run, &wait);
2044 * All the vcpus in this vcore are idle, so wait for a decrementer
2045 * or external interrupt to one of the vcpus. vc->lock is held.
2047 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2049 struct kvm_vcpu *vcpu;
2054 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2057 * Check one last time for pending exceptions and ceded state after
2058 * we put ourselves on the wait queue
2060 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2061 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2068 finish_wait(&vc->wq, &wait);
2072 vc->vcore_state = VCORE_SLEEPING;
2073 trace_kvmppc_vcore_blocked(vc, 0);
2074 spin_unlock(&vc->lock);
2076 finish_wait(&vc->wq, &wait);
2077 spin_lock(&vc->lock);
2078 vc->vcore_state = VCORE_INACTIVE;
2079 trace_kvmppc_vcore_blocked(vc, 1);
2082 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2085 struct kvmppc_vcore *vc;
2086 struct kvm_vcpu *v, *vn;
2088 trace_kvmppc_run_vcpu_enter(vcpu);
2090 kvm_run->exit_reason = 0;
2091 vcpu->arch.ret = RESUME_GUEST;
2092 vcpu->arch.trap = 0;
2093 kvmppc_update_vpas(vcpu);
2096 * Synchronize with other threads in this virtual core
2098 vc = vcpu->arch.vcore;
2099 spin_lock(&vc->lock);
2100 vcpu->arch.ceded = 0;
2101 vcpu->arch.run_task = current;
2102 vcpu->arch.kvm_run = kvm_run;
2103 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2104 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2105 vcpu->arch.busy_preempt = TB_NIL;
2106 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2110 * This happens the first time this is called for a vcpu.
2111 * If the vcore is already running, we may be able to start
2112 * this thread straight away and have it join in.
2114 if (!signal_pending(current)) {
2115 if (vc->vcore_state == VCORE_RUNNING &&
2116 VCORE_EXIT_COUNT(vc) == 0) {
2117 kvmppc_create_dtl_entry(vcpu, vc);
2118 kvmppc_start_thread(vcpu);
2119 trace_kvm_guest_enter(vcpu);
2120 } else if (vc->vcore_state == VCORE_SLEEPING) {
2126 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2127 !signal_pending(current)) {
2128 if (vc->vcore_state != VCORE_INACTIVE) {
2129 spin_unlock(&vc->lock);
2130 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
2131 spin_lock(&vc->lock);
2134 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2136 kvmppc_core_prepare_to_enter(v);
2137 if (signal_pending(v->arch.run_task)) {
2138 kvmppc_remove_runnable(vc, v);
2139 v->stat.signal_exits++;
2140 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2141 v->arch.ret = -EINTR;
2142 wake_up(&v->arch.cpu_run);
2145 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2148 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2149 if (!v->arch.pending_exceptions)
2150 n_ceded += v->arch.ceded;
2155 if (n_ceded == vc->n_runnable) {
2156 kvmppc_vcore_blocked(vc);
2157 } else if (should_resched()) {
2158 vc->vcore_state = VCORE_PREEMPT;
2159 /* Let something else run */
2160 cond_resched_lock(&vc->lock);
2161 vc->vcore_state = VCORE_INACTIVE;
2163 kvmppc_run_core(vc);
2168 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2169 (vc->vcore_state == VCORE_RUNNING ||
2170 vc->vcore_state == VCORE_EXITING)) {
2171 spin_unlock(&vc->lock);
2172 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
2173 spin_lock(&vc->lock);
2176 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2177 kvmppc_remove_runnable(vc, vcpu);
2178 vcpu->stat.signal_exits++;
2179 kvm_run->exit_reason = KVM_EXIT_INTR;
2180 vcpu->arch.ret = -EINTR;
2183 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2184 /* Wake up some vcpu to run the core */
2185 v = list_first_entry(&vc->runnable_threads,
2186 struct kvm_vcpu, arch.run_list);
2187 wake_up(&v->arch.cpu_run);
2190 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2191 spin_unlock(&vc->lock);
2192 return vcpu->arch.ret;
2195 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2200 if (!vcpu->arch.sane) {
2201 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2205 kvmppc_core_prepare_to_enter(vcpu);
2207 /* No need to go into the guest when all we'll do is come back out */
2208 if (signal_pending(current)) {
2209 run->exit_reason = KVM_EXIT_INTR;
2213 atomic_inc(&vcpu->kvm->arch.vcpus_running);
2214 /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2217 /* On the first time here, set up HTAB and VRMA */
2218 if (!vcpu->kvm->arch.hpte_setup_done) {
2219 r = kvmppc_hv_setup_htab_rma(vcpu);
2224 flush_fp_to_thread(current);
2225 flush_altivec_to_thread(current);
2226 flush_vsx_to_thread(current);
2227 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2228 vcpu->arch.pgdir = current->mm->pgd;
2229 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2232 r = kvmppc_run_vcpu(run, vcpu);
2234 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2235 !(vcpu->arch.shregs.msr & MSR_PR)) {
2236 trace_kvm_hcall_enter(vcpu);
2237 r = kvmppc_pseries_do_hcall(vcpu);
2238 trace_kvm_hcall_exit(vcpu, r);
2239 kvmppc_core_prepare_to_enter(vcpu);
2240 } else if (r == RESUME_PAGE_FAULT) {
2241 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2242 r = kvmppc_book3s_hv_page_fault(run, vcpu,
2243 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2244 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2246 } while (is_kvmppc_resume_guest(r));
2249 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2250 atomic_dec(&vcpu->kvm->arch.vcpus_running);
2254 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2257 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2261 (*sps)->page_shift = def->shift;
2262 (*sps)->slb_enc = def->sllp;
2263 (*sps)->enc[0].page_shift = def->shift;
2264 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2266 * Add 16MB MPSS support if host supports it
2268 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2269 (*sps)->enc[1].page_shift = 24;
2270 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2275 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2276 struct kvm_ppc_smmu_info *info)
2278 struct kvm_ppc_one_seg_page_size *sps;
2280 info->flags = KVM_PPC_PAGE_SIZES_REAL;
2281 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2282 info->flags |= KVM_PPC_1T_SEGMENTS;
2283 info->slb_size = mmu_slb_size;
2285 /* We only support these sizes for now, and no muti-size segments */
2286 sps = &info->sps[0];
2287 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2288 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2289 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2295 * Get (and clear) the dirty memory log for a memory slot.
2297 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2298 struct kvm_dirty_log *log)
2300 struct kvm_memory_slot *memslot;
2304 mutex_lock(&kvm->slots_lock);
2307 if (log->slot >= KVM_USER_MEM_SLOTS)
2310 memslot = id_to_memslot(kvm->memslots, log->slot);
2312 if (!memslot->dirty_bitmap)
2315 n = kvm_dirty_bitmap_bytes(memslot);
2316 memset(memslot->dirty_bitmap, 0, n);
2318 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2323 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2328 mutex_unlock(&kvm->slots_lock);
2332 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2333 struct kvm_memory_slot *dont)
2335 if (!dont || free->arch.rmap != dont->arch.rmap) {
2336 vfree(free->arch.rmap);
2337 free->arch.rmap = NULL;
2341 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2342 unsigned long npages)
2344 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2345 if (!slot->arch.rmap)
2351 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2352 struct kvm_memory_slot *memslot,
2353 struct kvm_userspace_memory_region *mem)
2358 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2359 struct kvm_userspace_memory_region *mem,
2360 const struct kvm_memory_slot *old)
2362 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2363 struct kvm_memory_slot *memslot;
2365 if (npages && old->npages) {
2367 * If modifying a memslot, reset all the rmap dirty bits.
2368 * If this is a new memslot, we don't need to do anything
2369 * since the rmap array starts out as all zeroes,
2370 * i.e. no pages are dirty.
2372 memslot = id_to_memslot(kvm->memslots, mem->slot);
2373 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2378 * Update LPCR values in kvm->arch and in vcores.
2379 * Caller must hold kvm->lock.
2381 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2386 if ((kvm->arch.lpcr & mask) == lpcr)
2389 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2391 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2392 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2395 spin_lock(&vc->lock);
2396 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2397 spin_unlock(&vc->lock);
2398 if (++cores_done >= kvm->arch.online_vcores)
2403 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2408 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2411 struct kvm *kvm = vcpu->kvm;
2413 struct kvm_memory_slot *memslot;
2414 struct vm_area_struct *vma;
2415 unsigned long lpcr = 0, senc;
2416 unsigned long psize, porder;
2419 mutex_lock(&kvm->lock);
2420 if (kvm->arch.hpte_setup_done)
2421 goto out; /* another vcpu beat us to it */
2423 /* Allocate hashed page table (if not done already) and reset it */
2424 if (!kvm->arch.hpt_virt) {
2425 err = kvmppc_alloc_hpt(kvm, NULL);
2427 pr_err("KVM: Couldn't alloc HPT\n");
2432 /* Look up the memslot for guest physical address 0 */
2433 srcu_idx = srcu_read_lock(&kvm->srcu);
2434 memslot = gfn_to_memslot(kvm, 0);
2436 /* We must have some memory at 0 by now */
2438 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2441 /* Look up the VMA for the start of this memory slot */
2442 hva = memslot->userspace_addr;
2443 down_read(¤t->mm->mmap_sem);
2444 vma = find_vma(current->mm, hva);
2445 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2448 psize = vma_kernel_pagesize(vma);
2449 porder = __ilog2(psize);
2451 up_read(¤t->mm->mmap_sem);
2453 /* We can handle 4k, 64k or 16M pages in the VRMA */
2455 if (!(psize == 0x1000 || psize == 0x10000 ||
2456 psize == 0x1000000))
2459 /* Update VRMASD field in the LPCR */
2460 senc = slb_pgsize_encoding(psize);
2461 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2462 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2463 /* the -4 is to account for senc values starting at 0x10 */
2464 lpcr = senc << (LPCR_VRMASD_SH - 4);
2466 /* Create HPTEs in the hash page table for the VRMA */
2467 kvmppc_map_vrma(vcpu, memslot, porder);
2469 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2471 /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2473 kvm->arch.hpte_setup_done = 1;
2476 srcu_read_unlock(&kvm->srcu, srcu_idx);
2478 mutex_unlock(&kvm->lock);
2482 up_read(¤t->mm->mmap_sem);
2486 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2488 unsigned long lpcr, lpid;
2491 /* Allocate the guest's logical partition ID */
2493 lpid = kvmppc_alloc_lpid();
2496 kvm->arch.lpid = lpid;
2499 * Since we don't flush the TLB when tearing down a VM,
2500 * and this lpid might have previously been used,
2501 * make sure we flush on each core before running the new VM.
2503 cpumask_setall(&kvm->arch.need_tlb_flush);
2505 /* Start out with the default set of hcalls enabled */
2506 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2507 sizeof(kvm->arch.enabled_hcalls));
2509 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2511 /* Init LPCR for virtual RMA mode */
2512 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2513 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2514 lpcr &= LPCR_PECE | LPCR_LPES;
2515 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2516 LPCR_VPM0 | LPCR_VPM1;
2517 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2518 (VRMA_VSID << SLB_VSID_SHIFT_1T);
2519 /* On POWER8 turn on online bit to enable PURR/SPURR */
2520 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2522 kvm->arch.lpcr = lpcr;
2525 * Track that we now have a HV mode VM active. This blocks secondary
2526 * CPU threads from coming online.
2528 kvm_hv_vm_activated();
2531 * Create a debugfs directory for the VM
2533 snprintf(buf, sizeof(buf), "vm%d", current->pid);
2534 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
2535 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2536 kvmppc_mmu_debugfs_init(kvm);
2541 static void kvmppc_free_vcores(struct kvm *kvm)
2545 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2546 if (kvm->arch.vcores[i] && kvm->arch.vcores[i]->mpp_buffer) {
2547 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2548 free_pages((unsigned long)vc->mpp_buffer,
2551 kfree(kvm->arch.vcores[i]);
2553 kvm->arch.online_vcores = 0;
2556 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2558 debugfs_remove_recursive(kvm->arch.debugfs_dir);
2560 kvm_hv_vm_deactivated();
2562 kvmppc_free_vcores(kvm);
2564 kvmppc_free_hpt(kvm);
2567 /* We don't need to emulate any privileged instructions or dcbz */
2568 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2569 unsigned int inst, int *advance)
2571 return EMULATE_FAIL;
2574 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2577 return EMULATE_FAIL;
2580 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2583 return EMULATE_FAIL;
2586 static int kvmppc_core_check_processor_compat_hv(void)
2588 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
2589 !cpu_has_feature(CPU_FTR_ARCH_206))
2594 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2595 unsigned int ioctl, unsigned long arg)
2597 struct kvm *kvm __maybe_unused = filp->private_data;
2598 void __user *argp = (void __user *)arg;
2603 case KVM_PPC_ALLOCATE_HTAB: {
2607 if (get_user(htab_order, (u32 __user *)argp))
2609 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2613 if (put_user(htab_order, (u32 __user *)argp))
2619 case KVM_PPC_GET_HTAB_FD: {
2620 struct kvm_get_htab_fd ghf;
2623 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2625 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2637 * List of hcall numbers to enable by default.
2638 * For compatibility with old userspace, we enable by default
2639 * all hcalls that were implemented before the hcall-enabling
2640 * facility was added. Note this list should not include H_RTAS.
2642 static unsigned int default_hcall_list[] = {
2656 #ifdef CONFIG_KVM_XICS
2667 static void init_default_hcalls(void)
2672 for (i = 0; default_hcall_list[i]; ++i) {
2673 hcall = default_hcall_list[i];
2674 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2675 __set_bit(hcall / 4, default_enabled_hcalls);
2679 static struct kvmppc_ops kvm_ops_hv = {
2680 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2681 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2682 .get_one_reg = kvmppc_get_one_reg_hv,
2683 .set_one_reg = kvmppc_set_one_reg_hv,
2684 .vcpu_load = kvmppc_core_vcpu_load_hv,
2685 .vcpu_put = kvmppc_core_vcpu_put_hv,
2686 .set_msr = kvmppc_set_msr_hv,
2687 .vcpu_run = kvmppc_vcpu_run_hv,
2688 .vcpu_create = kvmppc_core_vcpu_create_hv,
2689 .vcpu_free = kvmppc_core_vcpu_free_hv,
2690 .check_requests = kvmppc_core_check_requests_hv,
2691 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
2692 .flush_memslot = kvmppc_core_flush_memslot_hv,
2693 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2694 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
2695 .unmap_hva = kvm_unmap_hva_hv,
2696 .unmap_hva_range = kvm_unmap_hva_range_hv,
2697 .age_hva = kvm_age_hva_hv,
2698 .test_age_hva = kvm_test_age_hva_hv,
2699 .set_spte_hva = kvm_set_spte_hva_hv,
2700 .mmu_destroy = kvmppc_mmu_destroy_hv,
2701 .free_memslot = kvmppc_core_free_memslot_hv,
2702 .create_memslot = kvmppc_core_create_memslot_hv,
2703 .init_vm = kvmppc_core_init_vm_hv,
2704 .destroy_vm = kvmppc_core_destroy_vm_hv,
2705 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2706 .emulate_op = kvmppc_core_emulate_op_hv,
2707 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2708 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2709 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2710 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
2711 .hcall_implemented = kvmppc_hcall_impl_hv,
2714 static int kvmppc_book3s_init_hv(void)
2718 * FIXME!! Do we need to check on all cpus ?
2720 r = kvmppc_core_check_processor_compat_hv();
2724 kvm_ops_hv.owner = THIS_MODULE;
2725 kvmppc_hv_ops = &kvm_ops_hv;
2727 init_default_hcalls();
2729 r = kvmppc_mmu_hv_init();
2733 static void kvmppc_book3s_exit_hv(void)
2735 kvmppc_hv_ops = NULL;
2738 module_init(kvmppc_book3s_init_hv);
2739 module_exit(kvmppc_book3s_exit_hv);
2740 MODULE_LICENSE("GPL");
2741 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2742 MODULE_ALIAS("devname:kvm");