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/kernel.h>
23 #include <linux/err.h>
24 #include <linux/slab.h>
25 #include <linux/preempt.h>
26 #include <linux/sched/signal.h>
27 #include <linux/sched/stat.h>
28 #include <linux/delay.h>
29 #include <linux/export.h>
31 #include <linux/anon_inodes.h>
32 #include <linux/cpu.h>
33 #include <linux/cpumask.h>
34 #include <linux/spinlock.h>
35 #include <linux/page-flags.h>
36 #include <linux/srcu.h>
37 #include <linux/miscdevice.h>
38 #include <linux/debugfs.h>
39 #include <linux/gfp.h>
40 #include <linux/vmalloc.h>
41 #include <linux/highmem.h>
42 #include <linux/hugetlb.h>
43 #include <linux/kvm_irqfd.h>
44 #include <linux/irqbypass.h>
45 #include <linux/module.h>
46 #include <linux/compiler.h>
50 #include <asm/ppc-opcode.h>
51 #include <asm/asm-prototypes.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <asm/tlbflush.h>
56 #include <linux/uaccess.h>
58 #include <asm/kvm_ppc.h>
59 #include <asm/kvm_book3s.h>
60 #include <asm/mmu_context.h>
61 #include <asm/lppaca.h>
62 #include <asm/processor.h>
63 #include <asm/cputhreads.h>
65 #include <asm/hvcall.h>
66 #include <asm/switch_to.h>
68 #include <asm/dbell.h>
70 #include <asm/pnv-pci.h>
78 #define CREATE_TRACE_POINTS
81 /* #define EXIT_DEBUG */
82 /* #define EXIT_DEBUG_SIMPLE */
83 /* #define EXIT_DEBUG_INT */
85 /* Used to indicate that a guest page fault needs to be handled */
86 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
87 /* Used to indicate that a guest passthrough interrupt needs to be handled */
88 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
90 /* Used as a "null" value for timebase values */
91 #define TB_NIL (~(u64)0)
93 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
95 static int dynamic_mt_modes = 6;
96 module_param(dynamic_mt_modes, int, 0644);
97 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
98 static int target_smt_mode;
99 module_param(target_smt_mode, int, 0644);
100 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
102 static bool indep_threads_mode = true;
103 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
106 #ifdef CONFIG_KVM_XICS
107 static struct kernel_param_ops module_param_ops = {
108 .set = param_set_int,
109 .get = param_get_int,
112 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
113 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
115 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
116 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
119 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
120 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
122 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
126 struct kvm_vcpu *vcpu;
128 while (++i < MAX_SMT_THREADS) {
129 vcpu = READ_ONCE(vc->runnable_threads[i]);
138 /* Used to traverse the list of runnable threads for a given vcore */
139 #define for_each_runnable_thread(i, vcpu, vc) \
140 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
142 static bool kvmppc_ipi_thread(int cpu)
144 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
146 /* On POWER9 we can use msgsnd to IPI any cpu */
147 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
148 msg |= get_hard_smp_processor_id(cpu);
150 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
154 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
155 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
157 if (cpu_first_thread_sibling(cpu) ==
158 cpu_first_thread_sibling(smp_processor_id())) {
159 msg |= cpu_thread_in_core(cpu);
161 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
168 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
169 if (cpu >= 0 && cpu < nr_cpu_ids) {
170 if (paca[cpu].kvm_hstate.xics_phys) {
174 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
182 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
185 struct swait_queue_head *wqp;
187 wqp = kvm_arch_vcpu_wq(vcpu);
188 if (swq_has_sleeper(wqp)) {
190 ++vcpu->stat.halt_wakeup;
193 cpu = READ_ONCE(vcpu->arch.thread_cpu);
194 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
197 /* CPU points to the first thread of the core */
199 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
200 smp_send_reschedule(cpu);
204 * We use the vcpu_load/put functions to measure stolen time.
205 * Stolen time is counted as time when either the vcpu is able to
206 * run as part of a virtual core, but the task running the vcore
207 * is preempted or sleeping, or when the vcpu needs something done
208 * in the kernel by the task running the vcpu, but that task is
209 * preempted or sleeping. Those two things have to be counted
210 * separately, since one of the vcpu tasks will take on the job
211 * of running the core, and the other vcpu tasks in the vcore will
212 * sleep waiting for it to do that, but that sleep shouldn't count
215 * Hence we accumulate stolen time when the vcpu can run as part of
216 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
217 * needs its task to do other things in the kernel (for example,
218 * service a page fault) in busy_stolen. We don't accumulate
219 * stolen time for a vcore when it is inactive, or for a vcpu
220 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
221 * a misnomer; it means that the vcpu task is not executing in
222 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
223 * the kernel. We don't have any way of dividing up that time
224 * between time that the vcpu is genuinely stopped, time that
225 * the task is actively working on behalf of the vcpu, and time
226 * that the task is preempted, so we don't count any of it as
229 * Updates to busy_stolen are protected by arch.tbacct_lock;
230 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
231 * lock. The stolen times are measured in units of timebase ticks.
232 * (Note that the != TB_NIL checks below are purely defensive;
233 * they should never fail.)
236 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
240 spin_lock_irqsave(&vc->stoltb_lock, flags);
241 vc->preempt_tb = mftb();
242 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
245 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
249 spin_lock_irqsave(&vc->stoltb_lock, flags);
250 if (vc->preempt_tb != TB_NIL) {
251 vc->stolen_tb += mftb() - vc->preempt_tb;
252 vc->preempt_tb = TB_NIL;
254 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
257 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
259 struct kvmppc_vcore *vc = vcpu->arch.vcore;
263 * We can test vc->runner without taking the vcore lock,
264 * because only this task ever sets vc->runner to this
265 * vcpu, and once it is set to this vcpu, only this task
266 * ever sets it to NULL.
268 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
269 kvmppc_core_end_stolen(vc);
271 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
272 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
273 vcpu->arch.busy_preempt != TB_NIL) {
274 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
275 vcpu->arch.busy_preempt = TB_NIL;
277 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
280 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
282 struct kvmppc_vcore *vc = vcpu->arch.vcore;
285 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
286 kvmppc_core_start_stolen(vc);
288 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
289 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
290 vcpu->arch.busy_preempt = mftb();
291 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
294 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
297 * Check for illegal transactional state bit combination
298 * and if we find it, force the TS field to a safe state.
300 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
302 vcpu->arch.shregs.msr = msr;
303 kvmppc_end_cede(vcpu);
306 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
308 vcpu->arch.pvr = pvr;
311 /* Dummy value used in computing PCR value below */
312 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
314 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
316 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
317 struct kvmppc_vcore *vc = vcpu->arch.vcore;
319 /* We can (emulate) our own architecture version and anything older */
320 if (cpu_has_feature(CPU_FTR_ARCH_300))
321 host_pcr_bit = PCR_ARCH_300;
322 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
323 host_pcr_bit = PCR_ARCH_207;
324 else if (cpu_has_feature(CPU_FTR_ARCH_206))
325 host_pcr_bit = PCR_ARCH_206;
327 host_pcr_bit = PCR_ARCH_205;
329 /* Determine lowest PCR bit needed to run guest in given PVR level */
330 guest_pcr_bit = host_pcr_bit;
332 switch (arch_compat) {
334 guest_pcr_bit = PCR_ARCH_205;
338 guest_pcr_bit = PCR_ARCH_206;
341 guest_pcr_bit = PCR_ARCH_207;
344 guest_pcr_bit = PCR_ARCH_300;
351 /* Check requested PCR bits don't exceed our capabilities */
352 if (guest_pcr_bit > host_pcr_bit)
355 spin_lock(&vc->lock);
356 vc->arch_compat = arch_compat;
357 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
358 vc->pcr = host_pcr_bit - guest_pcr_bit;
359 spin_unlock(&vc->lock);
364 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
368 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
369 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
370 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
371 for (r = 0; r < 16; ++r)
372 pr_err("r%2d = %.16lx r%d = %.16lx\n",
373 r, kvmppc_get_gpr(vcpu, r),
374 r+16, kvmppc_get_gpr(vcpu, r+16));
375 pr_err("ctr = %.16lx lr = %.16lx\n",
376 vcpu->arch.ctr, vcpu->arch.lr);
377 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
378 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
379 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
380 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
381 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
382 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
383 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
384 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
385 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
386 pr_err("fault dar = %.16lx dsisr = %.8x\n",
387 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
388 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
389 for (r = 0; r < vcpu->arch.slb_max; ++r)
390 pr_err(" ESID = %.16llx VSID = %.16llx\n",
391 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
392 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
393 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
394 vcpu->arch.last_inst);
397 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
399 struct kvm_vcpu *ret;
401 mutex_lock(&kvm->lock);
402 ret = kvm_get_vcpu_by_id(kvm, id);
403 mutex_unlock(&kvm->lock);
407 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
409 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
410 vpa->yield_count = cpu_to_be32(1);
413 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
414 unsigned long addr, unsigned long len)
416 /* check address is cacheline aligned */
417 if (addr & (L1_CACHE_BYTES - 1))
419 spin_lock(&vcpu->arch.vpa_update_lock);
420 if (v->next_gpa != addr || v->len != len) {
422 v->len = addr ? len : 0;
423 v->update_pending = 1;
425 spin_unlock(&vcpu->arch.vpa_update_lock);
429 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
438 static int vpa_is_registered(struct kvmppc_vpa *vpap)
440 if (vpap->update_pending)
441 return vpap->next_gpa != 0;
442 return vpap->pinned_addr != NULL;
445 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
447 unsigned long vcpuid, unsigned long vpa)
449 struct kvm *kvm = vcpu->kvm;
450 unsigned long len, nb;
452 struct kvm_vcpu *tvcpu;
455 struct kvmppc_vpa *vpap;
457 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
461 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
462 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
463 subfunc == H_VPA_REG_SLB) {
464 /* Registering new area - address must be cache-line aligned */
465 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
468 /* convert logical addr to kernel addr and read length */
469 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
472 if (subfunc == H_VPA_REG_VPA)
473 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
475 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
476 kvmppc_unpin_guest_page(kvm, va, vpa, false);
479 if (len > nb || len < sizeof(struct reg_vpa))
488 spin_lock(&tvcpu->arch.vpa_update_lock);
491 case H_VPA_REG_VPA: /* register VPA */
493 * The size of our lppaca is 1kB because of the way we align
494 * it for the guest to avoid crossing a 4kB boundary. We only
495 * use 640 bytes of the structure though, so we should accept
496 * clients that set a size of 640.
500 vpap = &tvcpu->arch.vpa;
504 case H_VPA_REG_DTL: /* register DTL */
505 if (len < sizeof(struct dtl_entry))
507 len -= len % sizeof(struct dtl_entry);
509 /* Check that they have previously registered a VPA */
511 if (!vpa_is_registered(&tvcpu->arch.vpa))
514 vpap = &tvcpu->arch.dtl;
518 case H_VPA_REG_SLB: /* register SLB shadow buffer */
519 /* Check that they have previously registered a VPA */
521 if (!vpa_is_registered(&tvcpu->arch.vpa))
524 vpap = &tvcpu->arch.slb_shadow;
528 case H_VPA_DEREG_VPA: /* deregister VPA */
529 /* Check they don't still have a DTL or SLB buf registered */
531 if (vpa_is_registered(&tvcpu->arch.dtl) ||
532 vpa_is_registered(&tvcpu->arch.slb_shadow))
535 vpap = &tvcpu->arch.vpa;
539 case H_VPA_DEREG_DTL: /* deregister DTL */
540 vpap = &tvcpu->arch.dtl;
544 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
545 vpap = &tvcpu->arch.slb_shadow;
551 vpap->next_gpa = vpa;
553 vpap->update_pending = 1;
556 spin_unlock(&tvcpu->arch.vpa_update_lock);
561 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
563 struct kvm *kvm = vcpu->kvm;
569 * We need to pin the page pointed to by vpap->next_gpa,
570 * but we can't call kvmppc_pin_guest_page under the lock
571 * as it does get_user_pages() and down_read(). So we
572 * have to drop the lock, pin the page, then get the lock
573 * again and check that a new area didn't get registered
577 gpa = vpap->next_gpa;
578 spin_unlock(&vcpu->arch.vpa_update_lock);
582 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
583 spin_lock(&vcpu->arch.vpa_update_lock);
584 if (gpa == vpap->next_gpa)
586 /* sigh... unpin that one and try again */
588 kvmppc_unpin_guest_page(kvm, va, gpa, false);
591 vpap->update_pending = 0;
592 if (va && nb < vpap->len) {
594 * If it's now too short, it must be that userspace
595 * has changed the mappings underlying guest memory,
596 * so unregister the region.
598 kvmppc_unpin_guest_page(kvm, va, gpa, false);
601 if (vpap->pinned_addr)
602 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
605 vpap->pinned_addr = va;
608 vpap->pinned_end = va + vpap->len;
611 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
613 if (!(vcpu->arch.vpa.update_pending ||
614 vcpu->arch.slb_shadow.update_pending ||
615 vcpu->arch.dtl.update_pending))
618 spin_lock(&vcpu->arch.vpa_update_lock);
619 if (vcpu->arch.vpa.update_pending) {
620 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
621 if (vcpu->arch.vpa.pinned_addr)
622 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
624 if (vcpu->arch.dtl.update_pending) {
625 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
626 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
627 vcpu->arch.dtl_index = 0;
629 if (vcpu->arch.slb_shadow.update_pending)
630 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
631 spin_unlock(&vcpu->arch.vpa_update_lock);
635 * Return the accumulated stolen time for the vcore up until `now'.
636 * The caller should hold the vcore lock.
638 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
643 spin_lock_irqsave(&vc->stoltb_lock, flags);
645 if (vc->vcore_state != VCORE_INACTIVE &&
646 vc->preempt_tb != TB_NIL)
647 p += now - vc->preempt_tb;
648 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
652 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
653 struct kvmppc_vcore *vc)
655 struct dtl_entry *dt;
657 unsigned long stolen;
658 unsigned long core_stolen;
662 dt = vcpu->arch.dtl_ptr;
663 vpa = vcpu->arch.vpa.pinned_addr;
665 core_stolen = vcore_stolen_time(vc, now);
666 stolen = core_stolen - vcpu->arch.stolen_logged;
667 vcpu->arch.stolen_logged = core_stolen;
668 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
669 stolen += vcpu->arch.busy_stolen;
670 vcpu->arch.busy_stolen = 0;
671 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
674 memset(dt, 0, sizeof(struct dtl_entry));
675 dt->dispatch_reason = 7;
676 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
677 dt->timebase = cpu_to_be64(now + vc->tb_offset);
678 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
679 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
680 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
682 if (dt == vcpu->arch.dtl.pinned_end)
683 dt = vcpu->arch.dtl.pinned_addr;
684 vcpu->arch.dtl_ptr = dt;
685 /* order writing *dt vs. writing vpa->dtl_idx */
687 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
688 vcpu->arch.dtl.dirty = true;
691 /* See if there is a doorbell interrupt pending for a vcpu */
692 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
695 struct kvmppc_vcore *vc;
697 if (vcpu->arch.doorbell_request)
700 * Ensure that the read of vcore->dpdes comes after the read
701 * of vcpu->doorbell_request. This barrier matches the
702 * lwsync in book3s_hv_rmhandlers.S just before the
703 * fast_guest_return label.
706 vc = vcpu->arch.vcore;
707 thr = vcpu->vcpu_id - vc->first_vcpuid;
708 return !!(vc->dpdes & (1 << thr));
711 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
713 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
715 if ((!vcpu->arch.vcore->arch_compat) &&
716 cpu_has_feature(CPU_FTR_ARCH_207S))
721 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
722 unsigned long resource, unsigned long value1,
723 unsigned long value2)
726 case H_SET_MODE_RESOURCE_SET_CIABR:
727 if (!kvmppc_power8_compatible(vcpu))
732 return H_UNSUPPORTED_FLAG_START;
733 /* Guests can't breakpoint the hypervisor */
734 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
736 vcpu->arch.ciabr = value1;
738 case H_SET_MODE_RESOURCE_SET_DAWR:
739 if (!kvmppc_power8_compatible(vcpu))
742 return H_UNSUPPORTED_FLAG_START;
743 if (value2 & DABRX_HYP)
745 vcpu->arch.dawr = value1;
746 vcpu->arch.dawrx = value2;
753 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
755 struct kvmppc_vcore *vcore = target->arch.vcore;
758 * We expect to have been called by the real mode handler
759 * (kvmppc_rm_h_confer()) which would have directly returned
760 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
761 * have useful work to do and should not confer) so we don't
765 spin_lock(&vcore->lock);
766 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
767 vcore->vcore_state != VCORE_INACTIVE &&
769 target = vcore->runner;
770 spin_unlock(&vcore->lock);
772 return kvm_vcpu_yield_to(target);
775 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
778 struct lppaca *lppaca;
780 spin_lock(&vcpu->arch.vpa_update_lock);
781 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
783 yield_count = be32_to_cpu(lppaca->yield_count);
784 spin_unlock(&vcpu->arch.vpa_update_lock);
788 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
790 unsigned long req = kvmppc_get_gpr(vcpu, 3);
791 unsigned long target, ret = H_SUCCESS;
793 struct kvm_vcpu *tvcpu;
796 if (req <= MAX_HCALL_OPCODE &&
797 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
804 target = kvmppc_get_gpr(vcpu, 4);
805 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
810 tvcpu->arch.prodded = 1;
812 if (tvcpu->arch.ceded)
813 kvmppc_fast_vcpu_kick_hv(tvcpu);
816 target = kvmppc_get_gpr(vcpu, 4);
819 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
824 yield_count = kvmppc_get_gpr(vcpu, 5);
825 if (kvmppc_get_yield_count(tvcpu) != yield_count)
827 kvm_arch_vcpu_yield_to(tvcpu);
830 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
831 kvmppc_get_gpr(vcpu, 5),
832 kvmppc_get_gpr(vcpu, 6));
835 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
838 idx = srcu_read_lock(&vcpu->kvm->srcu);
839 rc = kvmppc_rtas_hcall(vcpu);
840 srcu_read_unlock(&vcpu->kvm->srcu, idx);
847 /* Send the error out to userspace via KVM_RUN */
849 case H_LOGICAL_CI_LOAD:
850 ret = kvmppc_h_logical_ci_load(vcpu);
851 if (ret == H_TOO_HARD)
854 case H_LOGICAL_CI_STORE:
855 ret = kvmppc_h_logical_ci_store(vcpu);
856 if (ret == H_TOO_HARD)
860 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
861 kvmppc_get_gpr(vcpu, 5),
862 kvmppc_get_gpr(vcpu, 6),
863 kvmppc_get_gpr(vcpu, 7));
864 if (ret == H_TOO_HARD)
873 if (kvmppc_xics_enabled(vcpu)) {
874 if (xive_enabled()) {
875 ret = H_NOT_AVAILABLE;
878 ret = kvmppc_xics_hcall(vcpu, req);
883 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
884 kvmppc_get_gpr(vcpu, 5),
885 kvmppc_get_gpr(vcpu, 6));
886 if (ret == H_TOO_HARD)
889 case H_PUT_TCE_INDIRECT:
890 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
891 kvmppc_get_gpr(vcpu, 5),
892 kvmppc_get_gpr(vcpu, 6),
893 kvmppc_get_gpr(vcpu, 7));
894 if (ret == H_TOO_HARD)
898 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
899 kvmppc_get_gpr(vcpu, 5),
900 kvmppc_get_gpr(vcpu, 6),
901 kvmppc_get_gpr(vcpu, 7));
902 if (ret == H_TOO_HARD)
908 kvmppc_set_gpr(vcpu, 3, ret);
909 vcpu->arch.hcall_needed = 0;
913 static int kvmppc_hcall_impl_hv(unsigned long cmd)
921 case H_LOGICAL_CI_LOAD:
922 case H_LOGICAL_CI_STORE:
923 #ifdef CONFIG_KVM_XICS
934 /* See if it's in the real-mode table */
935 return kvmppc_hcall_impl_hv_realmode(cmd);
938 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
939 struct kvm_vcpu *vcpu)
943 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
946 * Fetch failed, so return to guest and
947 * try executing it again.
952 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
953 run->exit_reason = KVM_EXIT_DEBUG;
954 run->debug.arch.address = kvmppc_get_pc(vcpu);
957 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
962 static void do_nothing(void *x)
966 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
968 int thr, cpu, pcpu, nthreads;
972 nthreads = vcpu->kvm->arch.emul_smt_mode;
974 cpu = vcpu->vcpu_id & ~(nthreads - 1);
975 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
976 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
980 * If the vcpu is currently running on a physical cpu thread,
981 * interrupt it in order to pull it out of the guest briefly,
982 * which will update its vcore->dpdes value.
984 pcpu = READ_ONCE(v->cpu);
986 smp_call_function_single(pcpu, do_nothing, NULL, 1);
987 if (kvmppc_doorbell_pending(v))
994 * On POWER9, emulate doorbell-related instructions in order to
995 * give the guest the illusion of running on a multi-threaded core.
996 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
999 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1003 struct kvm *kvm = vcpu->kvm;
1004 struct kvm_vcpu *tvcpu;
1006 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1007 return EMULATE_FAIL;
1008 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1009 return RESUME_GUEST;
1010 if (get_op(inst) != 31)
1011 return EMULATE_FAIL;
1013 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1014 switch (get_xop(inst)) {
1015 case OP_31_XOP_MSGSNDP:
1016 arg = kvmppc_get_gpr(vcpu, rb);
1017 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1020 if (arg >= kvm->arch.emul_smt_mode)
1022 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1025 if (!tvcpu->arch.doorbell_request) {
1026 tvcpu->arch.doorbell_request = 1;
1027 kvmppc_fast_vcpu_kick_hv(tvcpu);
1030 case OP_31_XOP_MSGCLRP:
1031 arg = kvmppc_get_gpr(vcpu, rb);
1032 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1034 vcpu->arch.vcore->dpdes = 0;
1035 vcpu->arch.doorbell_request = 0;
1037 case OP_31_XOP_MFSPR:
1038 switch (get_sprn(inst)) {
1043 arg = kvmppc_read_dpdes(vcpu);
1046 return EMULATE_FAIL;
1048 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1051 return EMULATE_FAIL;
1053 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1054 return RESUME_GUEST;
1057 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1058 struct task_struct *tsk)
1060 int r = RESUME_HOST;
1062 vcpu->stat.sum_exits++;
1065 * This can happen if an interrupt occurs in the last stages
1066 * of guest entry or the first stages of guest exit (i.e. after
1067 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1068 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1069 * That can happen due to a bug, or due to a machine check
1070 * occurring at just the wrong time.
1072 if (vcpu->arch.shregs.msr & MSR_HV) {
1073 printk(KERN_EMERG "KVM trap in HV mode!\n");
1074 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1075 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1076 vcpu->arch.shregs.msr);
1077 kvmppc_dump_regs(vcpu);
1078 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1079 run->hw.hardware_exit_reason = vcpu->arch.trap;
1082 run->exit_reason = KVM_EXIT_UNKNOWN;
1083 run->ready_for_interrupt_injection = 1;
1084 switch (vcpu->arch.trap) {
1085 /* We're good on these - the host merely wanted to get our attention */
1086 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1087 vcpu->stat.dec_exits++;
1090 case BOOK3S_INTERRUPT_EXTERNAL:
1091 case BOOK3S_INTERRUPT_H_DOORBELL:
1092 case BOOK3S_INTERRUPT_H_VIRT:
1093 vcpu->stat.ext_intr_exits++;
1096 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1097 case BOOK3S_INTERRUPT_HMI:
1098 case BOOK3S_INTERRUPT_PERFMON:
1099 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1102 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1103 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1104 run->exit_reason = KVM_EXIT_NMI;
1105 run->hw.hardware_exit_reason = vcpu->arch.trap;
1106 /* Clear out the old NMI status from run->flags */
1107 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1108 /* Now set the NMI status */
1109 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1110 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1112 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1115 /* Print the MCE event to host console. */
1116 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1118 case BOOK3S_INTERRUPT_PROGRAM:
1122 * Normally program interrupts are delivered directly
1123 * to the guest by the hardware, but we can get here
1124 * as a result of a hypervisor emulation interrupt
1125 * (e40) getting turned into a 700 by BML RTAS.
1127 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1128 kvmppc_core_queue_program(vcpu, flags);
1132 case BOOK3S_INTERRUPT_SYSCALL:
1134 /* hcall - punt to userspace */
1137 /* hypercall with MSR_PR has already been handled in rmode,
1138 * and never reaches here.
1141 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1142 for (i = 0; i < 9; ++i)
1143 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1144 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1145 vcpu->arch.hcall_needed = 1;
1150 * We get these next two if the guest accesses a page which it thinks
1151 * it has mapped but which is not actually present, either because
1152 * it is for an emulated I/O device or because the corresonding
1153 * host page has been paged out. Any other HDSI/HISI interrupts
1154 * have been handled already.
1156 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1157 r = RESUME_PAGE_FAULT;
1159 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1160 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1161 vcpu->arch.fault_dsisr = 0;
1162 r = RESUME_PAGE_FAULT;
1165 * This occurs if the guest executes an illegal instruction.
1166 * If the guest debug is disabled, generate a program interrupt
1167 * to the guest. If guest debug is enabled, we need to check
1168 * whether the instruction is a software breakpoint instruction.
1169 * Accordingly return to Guest or Host.
1171 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1172 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1173 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1174 swab32(vcpu->arch.emul_inst) :
1175 vcpu->arch.emul_inst;
1176 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1177 r = kvmppc_emulate_debug_inst(run, vcpu);
1179 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1184 * This occurs if the guest (kernel or userspace), does something that
1185 * is prohibited by HFSCR.
1186 * On POWER9, this could be a doorbell instruction that we need
1188 * Otherwise, we just generate a program interrupt to the guest.
1190 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1192 if ((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG)
1193 r = kvmppc_emulate_doorbell_instr(vcpu);
1194 if (r == EMULATE_FAIL) {
1195 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1199 case BOOK3S_INTERRUPT_HV_RM_HARD:
1200 r = RESUME_PASSTHROUGH;
1203 kvmppc_dump_regs(vcpu);
1204 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1205 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1206 vcpu->arch.shregs.msr);
1207 run->hw.hardware_exit_reason = vcpu->arch.trap;
1215 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1216 struct kvm_sregs *sregs)
1220 memset(sregs, 0, sizeof(struct kvm_sregs));
1221 sregs->pvr = vcpu->arch.pvr;
1222 for (i = 0; i < vcpu->arch.slb_max; i++) {
1223 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1224 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1230 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1231 struct kvm_sregs *sregs)
1235 /* Only accept the same PVR as the host's, since we can't spoof it */
1236 if (sregs->pvr != vcpu->arch.pvr)
1240 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1241 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1242 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1243 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1247 vcpu->arch.slb_max = j;
1252 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1253 bool preserve_top32)
1255 struct kvm *kvm = vcpu->kvm;
1256 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1259 mutex_lock(&kvm->lock);
1260 spin_lock(&vc->lock);
1262 * If ILE (interrupt little-endian) has changed, update the
1263 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1265 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1266 struct kvm_vcpu *vcpu;
1269 kvm_for_each_vcpu(i, vcpu, kvm) {
1270 if (vcpu->arch.vcore != vc)
1272 if (new_lpcr & LPCR_ILE)
1273 vcpu->arch.intr_msr |= MSR_LE;
1275 vcpu->arch.intr_msr &= ~MSR_LE;
1280 * Userspace can only modify DPFD (default prefetch depth),
1281 * ILE (interrupt little-endian) and TC (translation control).
1282 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1284 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1285 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1288 * On POWER9, allow userspace to enable large decrementer for the
1289 * guest, whether or not the host has it enabled.
1291 if (cpu_has_feature(CPU_FTR_ARCH_300))
1294 /* Broken 32-bit version of LPCR must not clear top bits */
1297 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1298 spin_unlock(&vc->lock);
1299 mutex_unlock(&kvm->lock);
1302 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1303 union kvmppc_one_reg *val)
1309 case KVM_REG_PPC_DEBUG_INST:
1310 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1312 case KVM_REG_PPC_HIOR:
1313 *val = get_reg_val(id, 0);
1315 case KVM_REG_PPC_DABR:
1316 *val = get_reg_val(id, vcpu->arch.dabr);
1318 case KVM_REG_PPC_DABRX:
1319 *val = get_reg_val(id, vcpu->arch.dabrx);
1321 case KVM_REG_PPC_DSCR:
1322 *val = get_reg_val(id, vcpu->arch.dscr);
1324 case KVM_REG_PPC_PURR:
1325 *val = get_reg_val(id, vcpu->arch.purr);
1327 case KVM_REG_PPC_SPURR:
1328 *val = get_reg_val(id, vcpu->arch.spurr);
1330 case KVM_REG_PPC_AMR:
1331 *val = get_reg_val(id, vcpu->arch.amr);
1333 case KVM_REG_PPC_UAMOR:
1334 *val = get_reg_val(id, vcpu->arch.uamor);
1336 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1337 i = id - KVM_REG_PPC_MMCR0;
1338 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1340 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1341 i = id - KVM_REG_PPC_PMC1;
1342 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1344 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1345 i = id - KVM_REG_PPC_SPMC1;
1346 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1348 case KVM_REG_PPC_SIAR:
1349 *val = get_reg_val(id, vcpu->arch.siar);
1351 case KVM_REG_PPC_SDAR:
1352 *val = get_reg_val(id, vcpu->arch.sdar);
1354 case KVM_REG_PPC_SIER:
1355 *val = get_reg_val(id, vcpu->arch.sier);
1357 case KVM_REG_PPC_IAMR:
1358 *val = get_reg_val(id, vcpu->arch.iamr);
1360 case KVM_REG_PPC_PSPB:
1361 *val = get_reg_val(id, vcpu->arch.pspb);
1363 case KVM_REG_PPC_DPDES:
1364 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1366 case KVM_REG_PPC_VTB:
1367 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1369 case KVM_REG_PPC_DAWR:
1370 *val = get_reg_val(id, vcpu->arch.dawr);
1372 case KVM_REG_PPC_DAWRX:
1373 *val = get_reg_val(id, vcpu->arch.dawrx);
1375 case KVM_REG_PPC_CIABR:
1376 *val = get_reg_val(id, vcpu->arch.ciabr);
1378 case KVM_REG_PPC_CSIGR:
1379 *val = get_reg_val(id, vcpu->arch.csigr);
1381 case KVM_REG_PPC_TACR:
1382 *val = get_reg_val(id, vcpu->arch.tacr);
1384 case KVM_REG_PPC_TCSCR:
1385 *val = get_reg_val(id, vcpu->arch.tcscr);
1387 case KVM_REG_PPC_PID:
1388 *val = get_reg_val(id, vcpu->arch.pid);
1390 case KVM_REG_PPC_ACOP:
1391 *val = get_reg_val(id, vcpu->arch.acop);
1393 case KVM_REG_PPC_WORT:
1394 *val = get_reg_val(id, vcpu->arch.wort);
1396 case KVM_REG_PPC_TIDR:
1397 *val = get_reg_val(id, vcpu->arch.tid);
1399 case KVM_REG_PPC_PSSCR:
1400 *val = get_reg_val(id, vcpu->arch.psscr);
1402 case KVM_REG_PPC_VPA_ADDR:
1403 spin_lock(&vcpu->arch.vpa_update_lock);
1404 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1405 spin_unlock(&vcpu->arch.vpa_update_lock);
1407 case KVM_REG_PPC_VPA_SLB:
1408 spin_lock(&vcpu->arch.vpa_update_lock);
1409 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1410 val->vpaval.length = vcpu->arch.slb_shadow.len;
1411 spin_unlock(&vcpu->arch.vpa_update_lock);
1413 case KVM_REG_PPC_VPA_DTL:
1414 spin_lock(&vcpu->arch.vpa_update_lock);
1415 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1416 val->vpaval.length = vcpu->arch.dtl.len;
1417 spin_unlock(&vcpu->arch.vpa_update_lock);
1419 case KVM_REG_PPC_TB_OFFSET:
1420 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1422 case KVM_REG_PPC_LPCR:
1423 case KVM_REG_PPC_LPCR_64:
1424 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1426 case KVM_REG_PPC_PPR:
1427 *val = get_reg_val(id, vcpu->arch.ppr);
1429 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1430 case KVM_REG_PPC_TFHAR:
1431 *val = get_reg_val(id, vcpu->arch.tfhar);
1433 case KVM_REG_PPC_TFIAR:
1434 *val = get_reg_val(id, vcpu->arch.tfiar);
1436 case KVM_REG_PPC_TEXASR:
1437 *val = get_reg_val(id, vcpu->arch.texasr);
1439 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1440 i = id - KVM_REG_PPC_TM_GPR0;
1441 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1443 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1446 i = id - KVM_REG_PPC_TM_VSR0;
1448 for (j = 0; j < TS_FPRWIDTH; j++)
1449 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1451 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1452 val->vval = vcpu->arch.vr_tm.vr[i-32];
1458 case KVM_REG_PPC_TM_CR:
1459 *val = get_reg_val(id, vcpu->arch.cr_tm);
1461 case KVM_REG_PPC_TM_XER:
1462 *val = get_reg_val(id, vcpu->arch.xer_tm);
1464 case KVM_REG_PPC_TM_LR:
1465 *val = get_reg_val(id, vcpu->arch.lr_tm);
1467 case KVM_REG_PPC_TM_CTR:
1468 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1470 case KVM_REG_PPC_TM_FPSCR:
1471 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1473 case KVM_REG_PPC_TM_AMR:
1474 *val = get_reg_val(id, vcpu->arch.amr_tm);
1476 case KVM_REG_PPC_TM_PPR:
1477 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1479 case KVM_REG_PPC_TM_VRSAVE:
1480 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1482 case KVM_REG_PPC_TM_VSCR:
1483 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1484 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1488 case KVM_REG_PPC_TM_DSCR:
1489 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1491 case KVM_REG_PPC_TM_TAR:
1492 *val = get_reg_val(id, vcpu->arch.tar_tm);
1495 case KVM_REG_PPC_ARCH_COMPAT:
1496 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1506 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1507 union kvmppc_one_reg *val)
1511 unsigned long addr, len;
1514 case KVM_REG_PPC_HIOR:
1515 /* Only allow this to be set to zero */
1516 if (set_reg_val(id, *val))
1519 case KVM_REG_PPC_DABR:
1520 vcpu->arch.dabr = set_reg_val(id, *val);
1522 case KVM_REG_PPC_DABRX:
1523 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1525 case KVM_REG_PPC_DSCR:
1526 vcpu->arch.dscr = set_reg_val(id, *val);
1528 case KVM_REG_PPC_PURR:
1529 vcpu->arch.purr = set_reg_val(id, *val);
1531 case KVM_REG_PPC_SPURR:
1532 vcpu->arch.spurr = set_reg_val(id, *val);
1534 case KVM_REG_PPC_AMR:
1535 vcpu->arch.amr = set_reg_val(id, *val);
1537 case KVM_REG_PPC_UAMOR:
1538 vcpu->arch.uamor = set_reg_val(id, *val);
1540 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1541 i = id - KVM_REG_PPC_MMCR0;
1542 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1544 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1545 i = id - KVM_REG_PPC_PMC1;
1546 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1548 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1549 i = id - KVM_REG_PPC_SPMC1;
1550 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1552 case KVM_REG_PPC_SIAR:
1553 vcpu->arch.siar = set_reg_val(id, *val);
1555 case KVM_REG_PPC_SDAR:
1556 vcpu->arch.sdar = set_reg_val(id, *val);
1558 case KVM_REG_PPC_SIER:
1559 vcpu->arch.sier = set_reg_val(id, *val);
1561 case KVM_REG_PPC_IAMR:
1562 vcpu->arch.iamr = set_reg_val(id, *val);
1564 case KVM_REG_PPC_PSPB:
1565 vcpu->arch.pspb = set_reg_val(id, *val);
1567 case KVM_REG_PPC_DPDES:
1568 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1570 case KVM_REG_PPC_VTB:
1571 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1573 case KVM_REG_PPC_DAWR:
1574 vcpu->arch.dawr = set_reg_val(id, *val);
1576 case KVM_REG_PPC_DAWRX:
1577 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1579 case KVM_REG_PPC_CIABR:
1580 vcpu->arch.ciabr = set_reg_val(id, *val);
1581 /* Don't allow setting breakpoints in hypervisor code */
1582 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1583 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1585 case KVM_REG_PPC_CSIGR:
1586 vcpu->arch.csigr = set_reg_val(id, *val);
1588 case KVM_REG_PPC_TACR:
1589 vcpu->arch.tacr = set_reg_val(id, *val);
1591 case KVM_REG_PPC_TCSCR:
1592 vcpu->arch.tcscr = set_reg_val(id, *val);
1594 case KVM_REG_PPC_PID:
1595 vcpu->arch.pid = set_reg_val(id, *val);
1597 case KVM_REG_PPC_ACOP:
1598 vcpu->arch.acop = set_reg_val(id, *val);
1600 case KVM_REG_PPC_WORT:
1601 vcpu->arch.wort = set_reg_val(id, *val);
1603 case KVM_REG_PPC_TIDR:
1604 vcpu->arch.tid = set_reg_val(id, *val);
1606 case KVM_REG_PPC_PSSCR:
1607 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1609 case KVM_REG_PPC_VPA_ADDR:
1610 addr = set_reg_val(id, *val);
1612 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1613 vcpu->arch.dtl.next_gpa))
1615 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1617 case KVM_REG_PPC_VPA_SLB:
1618 addr = val->vpaval.addr;
1619 len = val->vpaval.length;
1621 if (addr && !vcpu->arch.vpa.next_gpa)
1623 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1625 case KVM_REG_PPC_VPA_DTL:
1626 addr = val->vpaval.addr;
1627 len = val->vpaval.length;
1629 if (addr && (len < sizeof(struct dtl_entry) ||
1630 !vcpu->arch.vpa.next_gpa))
1632 len -= len % sizeof(struct dtl_entry);
1633 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1635 case KVM_REG_PPC_TB_OFFSET:
1637 * POWER9 DD1 has an erratum where writing TBU40 causes
1638 * the timebase to lose ticks. So we don't let the
1639 * timebase offset be changed on P9 DD1. (It is
1640 * initialized to zero.)
1642 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1644 /* round up to multiple of 2^24 */
1645 vcpu->arch.vcore->tb_offset =
1646 ALIGN(set_reg_val(id, *val), 1UL << 24);
1648 case KVM_REG_PPC_LPCR:
1649 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1651 case KVM_REG_PPC_LPCR_64:
1652 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1654 case KVM_REG_PPC_PPR:
1655 vcpu->arch.ppr = set_reg_val(id, *val);
1657 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1658 case KVM_REG_PPC_TFHAR:
1659 vcpu->arch.tfhar = set_reg_val(id, *val);
1661 case KVM_REG_PPC_TFIAR:
1662 vcpu->arch.tfiar = set_reg_val(id, *val);
1664 case KVM_REG_PPC_TEXASR:
1665 vcpu->arch.texasr = set_reg_val(id, *val);
1667 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1668 i = id - KVM_REG_PPC_TM_GPR0;
1669 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1671 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1674 i = id - KVM_REG_PPC_TM_VSR0;
1676 for (j = 0; j < TS_FPRWIDTH; j++)
1677 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1679 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1680 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1685 case KVM_REG_PPC_TM_CR:
1686 vcpu->arch.cr_tm = set_reg_val(id, *val);
1688 case KVM_REG_PPC_TM_XER:
1689 vcpu->arch.xer_tm = set_reg_val(id, *val);
1691 case KVM_REG_PPC_TM_LR:
1692 vcpu->arch.lr_tm = set_reg_val(id, *val);
1694 case KVM_REG_PPC_TM_CTR:
1695 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1697 case KVM_REG_PPC_TM_FPSCR:
1698 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1700 case KVM_REG_PPC_TM_AMR:
1701 vcpu->arch.amr_tm = set_reg_val(id, *val);
1703 case KVM_REG_PPC_TM_PPR:
1704 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1706 case KVM_REG_PPC_TM_VRSAVE:
1707 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1709 case KVM_REG_PPC_TM_VSCR:
1710 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1711 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1715 case KVM_REG_PPC_TM_DSCR:
1716 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1718 case KVM_REG_PPC_TM_TAR:
1719 vcpu->arch.tar_tm = set_reg_val(id, *val);
1722 case KVM_REG_PPC_ARCH_COMPAT:
1723 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1734 * On POWER9, threads are independent and can be in different partitions.
1735 * Therefore we consider each thread to be a subcore.
1736 * There is a restriction that all threads have to be in the same
1737 * MMU mode (radix or HPT), unfortunately, but since we only support
1738 * HPT guests on a HPT host so far, that isn't an impediment yet.
1740 static int threads_per_vcore(struct kvm *kvm)
1742 if (kvm->arch.threads_indep)
1744 return threads_per_subcore;
1747 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1749 struct kvmppc_vcore *vcore;
1751 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1756 spin_lock_init(&vcore->lock);
1757 spin_lock_init(&vcore->stoltb_lock);
1758 init_swait_queue_head(&vcore->wq);
1759 vcore->preempt_tb = TB_NIL;
1760 vcore->lpcr = kvm->arch.lpcr;
1761 vcore->first_vcpuid = core * kvm->arch.smt_mode;
1763 INIT_LIST_HEAD(&vcore->preempt_list);
1768 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1769 static struct debugfs_timings_element {
1773 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1774 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1775 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1776 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1777 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1780 #define N_TIMINGS (ARRAY_SIZE(timings))
1782 struct debugfs_timings_state {
1783 struct kvm_vcpu *vcpu;
1784 unsigned int buflen;
1785 char buf[N_TIMINGS * 100];
1788 static int debugfs_timings_open(struct inode *inode, struct file *file)
1790 struct kvm_vcpu *vcpu = inode->i_private;
1791 struct debugfs_timings_state *p;
1793 p = kzalloc(sizeof(*p), GFP_KERNEL);
1797 kvm_get_kvm(vcpu->kvm);
1799 file->private_data = p;
1801 return nonseekable_open(inode, file);
1804 static int debugfs_timings_release(struct inode *inode, struct file *file)
1806 struct debugfs_timings_state *p = file->private_data;
1808 kvm_put_kvm(p->vcpu->kvm);
1813 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1814 size_t len, loff_t *ppos)
1816 struct debugfs_timings_state *p = file->private_data;
1817 struct kvm_vcpu *vcpu = p->vcpu;
1819 struct kvmhv_tb_accumulator tb;
1828 buf_end = s + sizeof(p->buf);
1829 for (i = 0; i < N_TIMINGS; ++i) {
1830 struct kvmhv_tb_accumulator *acc;
1832 acc = (struct kvmhv_tb_accumulator *)
1833 ((unsigned long)vcpu + timings[i].offset);
1835 for (loops = 0; loops < 1000; ++loops) {
1836 count = acc->seqcount;
1841 if (count == acc->seqcount) {
1849 snprintf(s, buf_end - s, "%s: stuck\n",
1852 snprintf(s, buf_end - s,
1853 "%s: %llu %llu %llu %llu\n",
1854 timings[i].name, count / 2,
1855 tb_to_ns(tb.tb_total),
1856 tb_to_ns(tb.tb_min),
1857 tb_to_ns(tb.tb_max));
1860 p->buflen = s - p->buf;
1864 if (pos >= p->buflen)
1866 if (len > p->buflen - pos)
1867 len = p->buflen - pos;
1868 n = copy_to_user(buf, p->buf + pos, len);
1878 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1879 size_t len, loff_t *ppos)
1884 static const struct file_operations debugfs_timings_ops = {
1885 .owner = THIS_MODULE,
1886 .open = debugfs_timings_open,
1887 .release = debugfs_timings_release,
1888 .read = debugfs_timings_read,
1889 .write = debugfs_timings_write,
1890 .llseek = generic_file_llseek,
1893 /* Create a debugfs directory for the vcpu */
1894 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1897 struct kvm *kvm = vcpu->kvm;
1899 snprintf(buf, sizeof(buf), "vcpu%u", id);
1900 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1902 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1903 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1905 vcpu->arch.debugfs_timings =
1906 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1907 vcpu, &debugfs_timings_ops);
1910 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1911 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1914 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1916 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1919 struct kvm_vcpu *vcpu;
1922 struct kvmppc_vcore *vcore;
1925 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1929 err = kvm_vcpu_init(vcpu, kvm, id);
1933 vcpu->arch.shared = &vcpu->arch.shregs;
1934 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1936 * The shared struct is never shared on HV,
1937 * so we can always use host endianness
1939 #ifdef __BIG_ENDIAN__
1940 vcpu->arch.shared_big_endian = true;
1942 vcpu->arch.shared_big_endian = false;
1945 vcpu->arch.mmcr[0] = MMCR0_FC;
1946 vcpu->arch.ctrl = CTRL_RUNLATCH;
1947 /* default to host PVR, since we can't spoof it */
1948 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1949 spin_lock_init(&vcpu->arch.vpa_update_lock);
1950 spin_lock_init(&vcpu->arch.tbacct_lock);
1951 vcpu->arch.busy_preempt = TB_NIL;
1952 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1955 * Set the default HFSCR for the guest from the host value.
1956 * This value is only used on POWER9.
1957 * On POWER9 DD1, TM doesn't work, so we make sure to
1958 * prevent the guest from using it.
1959 * On POWER9, we want to virtualize the doorbell facility, so we
1960 * turn off the HFSCR bit, which causes those instructions to trap.
1962 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1963 if (!cpu_has_feature(CPU_FTR_TM))
1964 vcpu->arch.hfscr &= ~HFSCR_TM;
1965 if (cpu_has_feature(CPU_FTR_ARCH_300))
1966 vcpu->arch.hfscr &= ~HFSCR_MSGP;
1968 kvmppc_mmu_book3s_hv_init(vcpu);
1970 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1972 init_waitqueue_head(&vcpu->arch.cpu_run);
1974 mutex_lock(&kvm->lock);
1977 core = id / kvm->arch.smt_mode;
1978 if (core < KVM_MAX_VCORES) {
1979 vcore = kvm->arch.vcores[core];
1982 vcore = kvmppc_vcore_create(kvm, core);
1983 kvm->arch.vcores[core] = vcore;
1984 kvm->arch.online_vcores++;
1987 mutex_unlock(&kvm->lock);
1992 spin_lock(&vcore->lock);
1993 ++vcore->num_threads;
1994 spin_unlock(&vcore->lock);
1995 vcpu->arch.vcore = vcore;
1996 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1997 vcpu->arch.thread_cpu = -1;
1998 vcpu->arch.prev_cpu = -1;
2000 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2001 kvmppc_sanity_check(vcpu);
2003 debugfs_vcpu_init(vcpu, id);
2008 kmem_cache_free(kvm_vcpu_cache, vcpu);
2010 return ERR_PTR(err);
2013 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2014 unsigned long flags)
2021 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2023 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2025 * On POWER8 (or POWER7), the threading mode is "strict",
2026 * so we pack smt_mode vcpus per vcore.
2028 if (smt_mode > threads_per_subcore)
2032 * On POWER9, the threading mode is "loose",
2033 * so each vcpu gets its own vcore.
2038 mutex_lock(&kvm->lock);
2040 if (!kvm->arch.online_vcores) {
2041 kvm->arch.smt_mode = smt_mode;
2042 kvm->arch.emul_smt_mode = esmt;
2045 mutex_unlock(&kvm->lock);
2050 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2052 if (vpa->pinned_addr)
2053 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2057 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2059 spin_lock(&vcpu->arch.vpa_update_lock);
2060 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2061 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2062 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2063 spin_unlock(&vcpu->arch.vpa_update_lock);
2064 kvm_vcpu_uninit(vcpu);
2065 kmem_cache_free(kvm_vcpu_cache, vcpu);
2068 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2070 /* Indicate we want to get back into the guest */
2074 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2076 unsigned long dec_nsec, now;
2079 if (now > vcpu->arch.dec_expires) {
2080 /* decrementer has already gone negative */
2081 kvmppc_core_queue_dec(vcpu);
2082 kvmppc_core_prepare_to_enter(vcpu);
2085 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2087 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2088 vcpu->arch.timer_running = 1;
2091 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2093 vcpu->arch.ceded = 0;
2094 if (vcpu->arch.timer_running) {
2095 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2096 vcpu->arch.timer_running = 0;
2100 extern int __kvmppc_vcore_entry(void);
2102 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2103 struct kvm_vcpu *vcpu)
2107 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2109 spin_lock_irq(&vcpu->arch.tbacct_lock);
2111 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2112 vcpu->arch.stolen_logged;
2113 vcpu->arch.busy_preempt = now;
2114 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2115 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2117 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2120 static int kvmppc_grab_hwthread(int cpu)
2122 struct paca_struct *tpaca;
2123 long timeout = 10000;
2127 /* Ensure the thread won't go into the kernel if it wakes */
2128 tpaca->kvm_hstate.kvm_vcpu = NULL;
2129 tpaca->kvm_hstate.kvm_vcore = NULL;
2130 tpaca->kvm_hstate.napping = 0;
2132 tpaca->kvm_hstate.hwthread_req = 1;
2135 * If the thread is already executing in the kernel (e.g. handling
2136 * a stray interrupt), wait for it to get back to nap mode.
2137 * The smp_mb() is to ensure that our setting of hwthread_req
2138 * is visible before we look at hwthread_state, so if this
2139 * races with the code at system_reset_pSeries and the thread
2140 * misses our setting of hwthread_req, we are sure to see its
2141 * setting of hwthread_state, and vice versa.
2144 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2145 if (--timeout <= 0) {
2146 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2154 static void kvmppc_release_hwthread(int cpu)
2156 struct paca_struct *tpaca;
2159 tpaca->kvm_hstate.hwthread_req = 0;
2160 tpaca->kvm_hstate.kvm_vcpu = NULL;
2161 tpaca->kvm_hstate.kvm_vcore = NULL;
2162 tpaca->kvm_hstate.kvm_split_mode = NULL;
2165 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2169 cpu = cpu_first_thread_sibling(cpu);
2170 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2172 * Make sure setting of bit in need_tlb_flush precedes
2173 * testing of cpu_in_guest bits. The matching barrier on
2174 * the other side is the first smp_mb() in kvmppc_run_core().
2177 for (i = 0; i < threads_per_core; ++i)
2178 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2179 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2182 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2184 struct kvm *kvm = vcpu->kvm;
2187 * With radix, the guest can do TLB invalidations itself,
2188 * and it could choose to use the local form (tlbiel) if
2189 * it is invalidating a translation that has only ever been
2190 * used on one vcpu. However, that doesn't mean it has
2191 * only ever been used on one physical cpu, since vcpus
2192 * can move around between pcpus. To cope with this, when
2193 * a vcpu moves from one pcpu to another, we need to tell
2194 * any vcpus running on the same core as this vcpu previously
2195 * ran to flush the TLB. The TLB is shared between threads,
2196 * so we use a single bit in .need_tlb_flush for all 4 threads.
2198 if (vcpu->arch.prev_cpu != pcpu) {
2199 if (vcpu->arch.prev_cpu >= 0 &&
2200 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2201 cpu_first_thread_sibling(pcpu))
2202 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2203 vcpu->arch.prev_cpu = pcpu;
2207 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2210 struct paca_struct *tpaca;
2211 struct kvm *kvm = vc->kvm;
2215 if (vcpu->arch.timer_running) {
2216 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2217 vcpu->arch.timer_running = 0;
2219 cpu += vcpu->arch.ptid;
2220 vcpu->cpu = vc->pcpu;
2221 vcpu->arch.thread_cpu = cpu;
2222 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2225 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2226 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2227 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2229 tpaca->kvm_hstate.kvm_vcore = vc;
2230 if (cpu != smp_processor_id())
2231 kvmppc_ipi_thread(cpu);
2234 static void kvmppc_wait_for_nap(int n_threads)
2236 int cpu = smp_processor_id();
2241 for (loops = 0; loops < 1000000; ++loops) {
2243 * Check if all threads are finished.
2244 * We set the vcore pointer when starting a thread
2245 * and the thread clears it when finished, so we look
2246 * for any threads that still have a non-NULL vcore ptr.
2248 for (i = 1; i < n_threads; ++i)
2249 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2251 if (i == n_threads) {
2258 for (i = 1; i < n_threads; ++i)
2259 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2260 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2264 * Check that we are on thread 0 and that any other threads in
2265 * this core are off-line. Then grab the threads so they can't
2268 static int on_primary_thread(void)
2270 int cpu = smp_processor_id();
2273 /* Are we on a primary subcore? */
2274 if (cpu_thread_in_subcore(cpu))
2278 while (++thr < threads_per_subcore)
2279 if (cpu_online(cpu + thr))
2282 /* Grab all hw threads so they can't go into the kernel */
2283 for (thr = 1; thr < threads_per_subcore; ++thr) {
2284 if (kvmppc_grab_hwthread(cpu + thr)) {
2285 /* Couldn't grab one; let the others go */
2287 kvmppc_release_hwthread(cpu + thr);
2288 } while (--thr > 0);
2296 * A list of virtual cores for each physical CPU.
2297 * These are vcores that could run but their runner VCPU tasks are
2298 * (or may be) preempted.
2300 struct preempted_vcore_list {
2301 struct list_head list;
2305 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2307 static void init_vcore_lists(void)
2311 for_each_possible_cpu(cpu) {
2312 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2313 spin_lock_init(&lp->lock);
2314 INIT_LIST_HEAD(&lp->list);
2318 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2320 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2322 vc->vcore_state = VCORE_PREEMPT;
2323 vc->pcpu = smp_processor_id();
2324 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2325 spin_lock(&lp->lock);
2326 list_add_tail(&vc->preempt_list, &lp->list);
2327 spin_unlock(&lp->lock);
2330 /* Start accumulating stolen time */
2331 kvmppc_core_start_stolen(vc);
2334 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2336 struct preempted_vcore_list *lp;
2338 kvmppc_core_end_stolen(vc);
2339 if (!list_empty(&vc->preempt_list)) {
2340 lp = &per_cpu(preempted_vcores, vc->pcpu);
2341 spin_lock(&lp->lock);
2342 list_del_init(&vc->preempt_list);
2343 spin_unlock(&lp->lock);
2345 vc->vcore_state = VCORE_INACTIVE;
2349 * This stores information about the virtual cores currently
2350 * assigned to a physical core.
2354 int max_subcore_threads;
2356 int subcore_threads[MAX_SUBCORES];
2357 struct kvmppc_vcore *vc[MAX_SUBCORES];
2361 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2362 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2364 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2366 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2368 memset(cip, 0, sizeof(*cip));
2369 cip->n_subcores = 1;
2370 cip->max_subcore_threads = vc->num_threads;
2371 cip->total_threads = vc->num_threads;
2372 cip->subcore_threads[0] = vc->num_threads;
2376 static bool subcore_config_ok(int n_subcores, int n_threads)
2379 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way split-core
2380 * mode, with one thread per subcore.
2382 if (cpu_has_feature(CPU_FTR_ARCH_300))
2383 return n_subcores <= 4 && n_threads == 1;
2385 /* On POWER8, can only dynamically split if unsplit to begin with */
2386 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2388 if (n_subcores > MAX_SUBCORES)
2390 if (n_subcores > 1) {
2391 if (!(dynamic_mt_modes & 2))
2393 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2397 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2400 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2402 vc->entry_exit_map = 0;
2404 vc->napping_threads = 0;
2405 vc->conferring_threads = 0;
2408 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2410 int n_threads = vc->num_threads;
2413 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2416 /* POWER9 currently requires all threads to be in the same MMU mode */
2417 if (cpu_has_feature(CPU_FTR_ARCH_300) &&
2418 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2421 if (n_threads < cip->max_subcore_threads)
2422 n_threads = cip->max_subcore_threads;
2423 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2425 cip->max_subcore_threads = n_threads;
2427 sub = cip->n_subcores;
2429 cip->total_threads += vc->num_threads;
2430 cip->subcore_threads[sub] = vc->num_threads;
2432 init_vcore_to_run(vc);
2433 list_del_init(&vc->preempt_list);
2439 * Work out whether it is possible to piggyback the execution of
2440 * vcore *pvc onto the execution of the other vcores described in *cip.
2442 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2445 if (cip->total_threads + pvc->num_threads > target_threads)
2448 return can_dynamic_split(pvc, cip);
2451 static void prepare_threads(struct kvmppc_vcore *vc)
2454 struct kvm_vcpu *vcpu;
2456 for_each_runnable_thread(i, vcpu, vc) {
2457 if (signal_pending(vcpu->arch.run_task))
2458 vcpu->arch.ret = -EINTR;
2459 else if (vcpu->arch.vpa.update_pending ||
2460 vcpu->arch.slb_shadow.update_pending ||
2461 vcpu->arch.dtl.update_pending)
2462 vcpu->arch.ret = RESUME_GUEST;
2465 kvmppc_remove_runnable(vc, vcpu);
2466 wake_up(&vcpu->arch.cpu_run);
2470 static void collect_piggybacks(struct core_info *cip, int target_threads)
2472 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2473 struct kvmppc_vcore *pvc, *vcnext;
2475 spin_lock(&lp->lock);
2476 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2477 if (!spin_trylock(&pvc->lock))
2479 prepare_threads(pvc);
2480 if (!pvc->n_runnable) {
2481 list_del_init(&pvc->preempt_list);
2482 if (pvc->runner == NULL) {
2483 pvc->vcore_state = VCORE_INACTIVE;
2484 kvmppc_core_end_stolen(pvc);
2486 spin_unlock(&pvc->lock);
2489 if (!can_piggyback(pvc, cip, target_threads)) {
2490 spin_unlock(&pvc->lock);
2493 kvmppc_core_end_stolen(pvc);
2494 pvc->vcore_state = VCORE_PIGGYBACK;
2495 if (cip->total_threads >= target_threads)
2498 spin_unlock(&lp->lock);
2501 static bool recheck_signals(struct core_info *cip)
2504 struct kvm_vcpu *vcpu;
2506 for (sub = 0; sub < cip->n_subcores; ++sub)
2507 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2508 if (signal_pending(vcpu->arch.run_task))
2513 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2515 int still_running = 0, i;
2518 struct kvm_vcpu *vcpu;
2520 spin_lock(&vc->lock);
2522 for_each_runnable_thread(i, vcpu, vc) {
2523 /* cancel pending dec exception if dec is positive */
2524 if (now < vcpu->arch.dec_expires &&
2525 kvmppc_core_pending_dec(vcpu))
2526 kvmppc_core_dequeue_dec(vcpu);
2528 trace_kvm_guest_exit(vcpu);
2531 if (vcpu->arch.trap)
2532 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2533 vcpu->arch.run_task);
2535 vcpu->arch.ret = ret;
2536 vcpu->arch.trap = 0;
2538 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2539 if (vcpu->arch.pending_exceptions)
2540 kvmppc_core_prepare_to_enter(vcpu);
2541 if (vcpu->arch.ceded)
2542 kvmppc_set_timer(vcpu);
2546 kvmppc_remove_runnable(vc, vcpu);
2547 wake_up(&vcpu->arch.cpu_run);
2551 if (still_running > 0) {
2552 kvmppc_vcore_preempt(vc);
2553 } else if (vc->runner) {
2554 vc->vcore_state = VCORE_PREEMPT;
2555 kvmppc_core_start_stolen(vc);
2557 vc->vcore_state = VCORE_INACTIVE;
2559 if (vc->n_runnable > 0 && vc->runner == NULL) {
2560 /* make sure there's a candidate runner awake */
2562 vcpu = next_runnable_thread(vc, &i);
2563 wake_up(&vcpu->arch.cpu_run);
2566 spin_unlock(&vc->lock);
2570 * Clear core from the list of active host cores as we are about to
2571 * enter the guest. Only do this if it is the primary thread of the
2572 * core (not if a subcore) that is entering the guest.
2574 static inline int kvmppc_clear_host_core(unsigned int cpu)
2578 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2581 * Memory barrier can be omitted here as we will do a smp_wmb()
2582 * later in kvmppc_start_thread and we need ensure that state is
2583 * visible to other CPUs only after we enter guest.
2585 core = cpu >> threads_shift;
2586 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2591 * Advertise this core as an active host core since we exited the guest
2592 * Only need to do this if it is the primary thread of the core that is
2595 static inline int kvmppc_set_host_core(unsigned int cpu)
2599 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2603 * Memory barrier can be omitted here because we do a spin_unlock
2604 * immediately after this which provides the memory barrier.
2606 core = cpu >> threads_shift;
2607 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2611 static void set_irq_happened(int trap)
2614 case BOOK3S_INTERRUPT_EXTERNAL:
2615 local_paca->irq_happened |= PACA_IRQ_EE;
2617 case BOOK3S_INTERRUPT_H_DOORBELL:
2618 local_paca->irq_happened |= PACA_IRQ_DBELL;
2620 case BOOK3S_INTERRUPT_HMI:
2621 local_paca->irq_happened |= PACA_IRQ_HMI;
2623 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2624 replay_system_reset();
2630 * Run a set of guest threads on a physical core.
2631 * Called with vc->lock held.
2633 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2635 struct kvm_vcpu *vcpu;
2638 struct core_info core_info;
2639 struct kvmppc_vcore *pvc;
2640 struct kvm_split_mode split_info, *sip;
2641 int split, subcore_size, active;
2644 unsigned long cmd_bit, stat_bit;
2647 int controlled_threads;
2653 * Remove from the list any threads that have a signal pending
2654 * or need a VPA update done
2656 prepare_threads(vc);
2658 /* if the runner is no longer runnable, let the caller pick a new one */
2659 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2665 init_vcore_to_run(vc);
2666 vc->preempt_tb = TB_NIL;
2669 * Number of threads that we will be controlling: the same as
2670 * the number of threads per subcore, except on POWER9,
2671 * where it's 1 because the threads are (mostly) independent.
2673 controlled_threads = threads_per_vcore(vc->kvm);
2676 * Make sure we are running on primary threads, and that secondary
2677 * threads are offline. Also check if the number of threads in this
2678 * guest are greater than the current system threads per guest.
2679 * On POWER9, we need to be not in independent-threads mode if
2680 * this is a HPT guest on a radix host.
2682 hpt_on_radix = radix_enabled() && !kvm_is_radix(vc->kvm);
2683 if (((controlled_threads > 1) &&
2684 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2685 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2686 for_each_runnable_thread(i, vcpu, vc) {
2687 vcpu->arch.ret = -EBUSY;
2688 kvmppc_remove_runnable(vc, vcpu);
2689 wake_up(&vcpu->arch.cpu_run);
2695 * See if we could run any other vcores on the physical core
2696 * along with this one.
2698 init_core_info(&core_info, vc);
2699 pcpu = smp_processor_id();
2700 target_threads = controlled_threads;
2701 if (target_smt_mode && target_smt_mode < target_threads)
2702 target_threads = target_smt_mode;
2703 if (vc->num_threads < target_threads)
2704 collect_piggybacks(&core_info, target_threads);
2707 * On radix, arrange for TLB flushing if necessary.
2708 * This has to be done before disabling interrupts since
2709 * it uses smp_call_function().
2711 pcpu = smp_processor_id();
2712 if (kvm_is_radix(vc->kvm)) {
2713 for (sub = 0; sub < core_info.n_subcores; ++sub)
2714 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2715 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2719 * Hard-disable interrupts, and check resched flag and signals.
2720 * If we need to reschedule or deliver a signal, clean up
2721 * and return without going into the guest(s).
2722 * If the mmu_ready flag has been cleared, don't go into the
2723 * guest because that means a HPT resize operation is in progress.
2725 local_irq_disable();
2727 if (lazy_irq_pending() || need_resched() ||
2728 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
2730 vc->vcore_state = VCORE_INACTIVE;
2731 /* Unlock all except the primary vcore */
2732 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2733 pvc = core_info.vc[sub];
2734 /* Put back on to the preempted vcores list */
2735 kvmppc_vcore_preempt(pvc);
2736 spin_unlock(&pvc->lock);
2738 for (i = 0; i < controlled_threads; ++i)
2739 kvmppc_release_hwthread(pcpu + i);
2743 kvmppc_clear_host_core(pcpu);
2745 /* Decide on micro-threading (split-core) mode */
2746 subcore_size = threads_per_subcore;
2747 cmd_bit = stat_bit = 0;
2748 split = core_info.n_subcores;
2750 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2751 && !cpu_has_feature(CPU_FTR_ARCH_300);
2753 if (split > 1 || hpt_on_radix) {
2755 memset(&split_info, 0, sizeof(split_info));
2756 for (sub = 0; sub < core_info.n_subcores; ++sub)
2757 split_info.vc[sub] = core_info.vc[sub];
2760 if (split == 2 && (dynamic_mt_modes & 2)) {
2761 cmd_bit = HID0_POWER8_1TO2LPAR;
2762 stat_bit = HID0_POWER8_2LPARMODE;
2765 cmd_bit = HID0_POWER8_1TO4LPAR;
2766 stat_bit = HID0_POWER8_4LPARMODE;
2768 subcore_size = MAX_SMT_THREADS / split;
2769 split_info.rpr = mfspr(SPRN_RPR);
2770 split_info.pmmar = mfspr(SPRN_PMMAR);
2771 split_info.ldbar = mfspr(SPRN_LDBAR);
2772 split_info.subcore_size = subcore_size;
2774 split_info.subcore_size = 1;
2776 /* Use the split_info for LPCR/LPIDR changes */
2777 split_info.lpcr_req = vc->lpcr;
2778 split_info.lpidr_req = vc->kvm->arch.lpid;
2779 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2780 split_info.do_set = 1;
2784 /* order writes to split_info before kvm_split_mode pointer */
2788 for (thr = 0; thr < controlled_threads; ++thr) {
2789 paca[pcpu + thr].kvm_hstate.tid = thr;
2790 paca[pcpu + thr].kvm_hstate.napping = 0;
2791 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2794 /* Initiate micro-threading (split-core) on POWER8 if required */
2796 unsigned long hid0 = mfspr(SPRN_HID0);
2798 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2800 mtspr(SPRN_HID0, hid0);
2803 hid0 = mfspr(SPRN_HID0);
2804 if (hid0 & stat_bit)
2810 /* Start all the threads */
2812 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2813 thr = is_power8 ? subcore_thread_map[sub] : sub;
2816 pvc = core_info.vc[sub];
2817 pvc->pcpu = pcpu + thr;
2818 for_each_runnable_thread(i, vcpu, pvc) {
2819 kvmppc_start_thread(vcpu, pvc);
2820 kvmppc_create_dtl_entry(vcpu, pvc);
2821 trace_kvm_guest_enter(vcpu);
2822 if (!vcpu->arch.ptid)
2824 active |= 1 << (thr + vcpu->arch.ptid);
2827 * We need to start the first thread of each subcore
2828 * even if it doesn't have a vcpu.
2831 kvmppc_start_thread(NULL, pvc);
2832 thr += pvc->num_threads;
2836 * Ensure that split_info.do_nap is set after setting
2837 * the vcore pointer in the PACA of the secondaries.
2842 * When doing micro-threading, poke the inactive threads as well.
2843 * This gets them to the nap instruction after kvm_do_nap,
2844 * which reduces the time taken to unsplit later.
2845 * For POWER9 HPT guest on radix host, we need all the secondary
2846 * threads woken up so they can do the LPCR/LPIDR change.
2848 if (cmd_bit || hpt_on_radix) {
2849 split_info.do_nap = 1; /* ask secondaries to nap when done */
2850 for (thr = 1; thr < threads_per_subcore; ++thr)
2851 if (!(active & (1 << thr)))
2852 kvmppc_ipi_thread(pcpu + thr);
2855 vc->vcore_state = VCORE_RUNNING;
2858 trace_kvmppc_run_core(vc, 0);
2860 for (sub = 0; sub < core_info.n_subcores; ++sub)
2861 spin_unlock(&core_info.vc[sub]->lock);
2864 * Interrupts will be enabled once we get into the guest,
2865 * so tell lockdep that we're about to enable interrupts.
2867 trace_hardirqs_on();
2871 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2873 trap = __kvmppc_vcore_entry();
2875 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2879 trace_hardirqs_off();
2880 set_irq_happened(trap);
2882 spin_lock(&vc->lock);
2883 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2884 vc->vcore_state = VCORE_EXITING;
2886 /* wait for secondary threads to finish writing their state to memory */
2887 kvmppc_wait_for_nap(controlled_threads);
2889 /* Return to whole-core mode if we split the core earlier */
2891 unsigned long hid0 = mfspr(SPRN_HID0);
2892 unsigned long loops = 0;
2894 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2895 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2897 mtspr(SPRN_HID0, hid0);
2900 hid0 = mfspr(SPRN_HID0);
2901 if (!(hid0 & stat_bit))
2906 } else if (hpt_on_radix) {
2907 /* Wait for all threads to have seen final sync */
2908 for (thr = 1; thr < controlled_threads; ++thr) {
2909 while (paca[pcpu + thr].kvm_hstate.kvm_split_mode) {
2916 split_info.do_nap = 0;
2918 kvmppc_set_host_core(pcpu);
2922 /* Let secondaries go back to the offline loop */
2923 for (i = 0; i < controlled_threads; ++i) {
2924 kvmppc_release_hwthread(pcpu + i);
2925 if (sip && sip->napped[i])
2926 kvmppc_ipi_thread(pcpu + i);
2927 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2930 spin_unlock(&vc->lock);
2932 /* make sure updates to secondary vcpu structs are visible now */
2935 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2936 pvc = core_info.vc[sub];
2937 post_guest_process(pvc, pvc == vc);
2940 spin_lock(&vc->lock);
2944 vc->vcore_state = VCORE_INACTIVE;
2945 trace_kvmppc_run_core(vc, 1);
2949 * Wait for some other vcpu thread to execute us, and
2950 * wake us up when we need to handle something in the host.
2952 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2953 struct kvm_vcpu *vcpu, int wait_state)
2957 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2958 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2959 spin_unlock(&vc->lock);
2961 spin_lock(&vc->lock);
2963 finish_wait(&vcpu->arch.cpu_run, &wait);
2966 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2969 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2970 vc->halt_poll_ns = 10000;
2972 vc->halt_poll_ns *= halt_poll_ns_grow;
2975 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2977 if (halt_poll_ns_shrink == 0)
2978 vc->halt_poll_ns = 0;
2980 vc->halt_poll_ns /= halt_poll_ns_shrink;
2983 #ifdef CONFIG_KVM_XICS
2984 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2986 if (!xive_enabled())
2988 return vcpu->arch.xive_saved_state.pipr <
2989 vcpu->arch.xive_saved_state.cppr;
2992 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2996 #endif /* CONFIG_KVM_XICS */
2998 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3000 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3001 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3008 * Check to see if any of the runnable vcpus on the vcore have pending
3009 * exceptions or are no longer ceded
3011 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3013 struct kvm_vcpu *vcpu;
3016 for_each_runnable_thread(i, vcpu, vc) {
3017 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3025 * All the vcpus in this vcore are idle, so wait for a decrementer
3026 * or external interrupt to one of the vcpus. vc->lock is held.
3028 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3030 ktime_t cur, start_poll, start_wait;
3033 DECLARE_SWAITQUEUE(wait);
3035 /* Poll for pending exceptions and ceded state */
3036 cur = start_poll = ktime_get();
3037 if (vc->halt_poll_ns) {
3038 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3039 ++vc->runner->stat.halt_attempted_poll;
3041 vc->vcore_state = VCORE_POLLING;
3042 spin_unlock(&vc->lock);
3045 if (kvmppc_vcore_check_block(vc)) {
3050 } while (single_task_running() && ktime_before(cur, stop));
3052 spin_lock(&vc->lock);
3053 vc->vcore_state = VCORE_INACTIVE;
3056 ++vc->runner->stat.halt_successful_poll;
3061 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3063 if (kvmppc_vcore_check_block(vc)) {
3064 finish_swait(&vc->wq, &wait);
3066 /* If we polled, count this as a successful poll */
3067 if (vc->halt_poll_ns)
3068 ++vc->runner->stat.halt_successful_poll;
3072 start_wait = ktime_get();
3074 vc->vcore_state = VCORE_SLEEPING;
3075 trace_kvmppc_vcore_blocked(vc, 0);
3076 spin_unlock(&vc->lock);
3078 finish_swait(&vc->wq, &wait);
3079 spin_lock(&vc->lock);
3080 vc->vcore_state = VCORE_INACTIVE;
3081 trace_kvmppc_vcore_blocked(vc, 1);
3082 ++vc->runner->stat.halt_successful_wait;
3087 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3089 /* Attribute wait time */
3091 vc->runner->stat.halt_wait_ns +=
3092 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3093 /* Attribute failed poll time */
3094 if (vc->halt_poll_ns)
3095 vc->runner->stat.halt_poll_fail_ns +=
3096 ktime_to_ns(start_wait) -
3097 ktime_to_ns(start_poll);
3099 /* Attribute successful poll time */
3100 if (vc->halt_poll_ns)
3101 vc->runner->stat.halt_poll_success_ns +=
3103 ktime_to_ns(start_poll);
3106 /* Adjust poll time */
3108 if (block_ns <= vc->halt_poll_ns)
3110 /* We slept and blocked for longer than the max halt time */
3111 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3112 shrink_halt_poll_ns(vc);
3113 /* We slept and our poll time is too small */
3114 else if (vc->halt_poll_ns < halt_poll_ns &&
3115 block_ns < halt_poll_ns)
3116 grow_halt_poll_ns(vc);
3117 if (vc->halt_poll_ns > halt_poll_ns)
3118 vc->halt_poll_ns = halt_poll_ns;
3120 vc->halt_poll_ns = 0;
3122 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3125 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3128 struct kvm *kvm = vcpu->kvm;
3130 mutex_lock(&kvm->lock);
3131 if (!kvm->arch.mmu_ready) {
3132 if (!kvm_is_radix(kvm))
3133 r = kvmppc_hv_setup_htab_rma(vcpu);
3135 if (cpu_has_feature(CPU_FTR_ARCH_300))
3136 kvmppc_setup_partition_table(kvm);
3137 kvm->arch.mmu_ready = 1;
3140 mutex_unlock(&kvm->lock);
3144 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3147 struct kvmppc_vcore *vc;
3150 trace_kvmppc_run_vcpu_enter(vcpu);
3152 kvm_run->exit_reason = 0;
3153 vcpu->arch.ret = RESUME_GUEST;
3154 vcpu->arch.trap = 0;
3155 kvmppc_update_vpas(vcpu);
3158 * Synchronize with other threads in this virtual core
3160 vc = vcpu->arch.vcore;
3161 spin_lock(&vc->lock);
3162 vcpu->arch.ceded = 0;
3163 vcpu->arch.run_task = current;
3164 vcpu->arch.kvm_run = kvm_run;
3165 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3166 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3167 vcpu->arch.busy_preempt = TB_NIL;
3168 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3172 * This happens the first time this is called for a vcpu.
3173 * If the vcore is already running, we may be able to start
3174 * this thread straight away and have it join in.
3176 if (!signal_pending(current)) {
3177 if (vc->vcore_state == VCORE_PIGGYBACK) {
3178 if (spin_trylock(&vc->lock)) {
3179 if (vc->vcore_state == VCORE_RUNNING &&
3180 !VCORE_IS_EXITING(vc)) {
3181 kvmppc_create_dtl_entry(vcpu, vc);
3182 kvmppc_start_thread(vcpu, vc);
3183 trace_kvm_guest_enter(vcpu);
3185 spin_unlock(&vc->lock);
3187 } else if (vc->vcore_state == VCORE_RUNNING &&
3188 !VCORE_IS_EXITING(vc)) {
3189 kvmppc_create_dtl_entry(vcpu, vc);
3190 kvmppc_start_thread(vcpu, vc);
3191 trace_kvm_guest_enter(vcpu);
3192 } else if (vc->vcore_state == VCORE_SLEEPING) {
3198 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3199 !signal_pending(current)) {
3200 /* See if the MMU is ready to go */
3201 if (!vcpu->kvm->arch.mmu_ready) {
3202 spin_unlock(&vc->lock);
3203 r = kvmhv_setup_mmu(vcpu);
3204 spin_lock(&vc->lock);
3206 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3207 kvm_run->fail_entry.
3208 hardware_entry_failure_reason = 0;
3214 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3215 kvmppc_vcore_end_preempt(vc);
3217 if (vc->vcore_state != VCORE_INACTIVE) {
3218 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3221 for_each_runnable_thread(i, v, vc) {
3222 kvmppc_core_prepare_to_enter(v);
3223 if (signal_pending(v->arch.run_task)) {
3224 kvmppc_remove_runnable(vc, v);
3225 v->stat.signal_exits++;
3226 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3227 v->arch.ret = -EINTR;
3228 wake_up(&v->arch.cpu_run);
3231 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3234 for_each_runnable_thread(i, v, vc) {
3235 if (!kvmppc_vcpu_woken(v))
3236 n_ceded += v->arch.ceded;
3241 if (n_ceded == vc->n_runnable) {
3242 kvmppc_vcore_blocked(vc);
3243 } else if (need_resched()) {
3244 kvmppc_vcore_preempt(vc);
3245 /* Let something else run */
3246 cond_resched_lock(&vc->lock);
3247 if (vc->vcore_state == VCORE_PREEMPT)
3248 kvmppc_vcore_end_preempt(vc);
3250 kvmppc_run_core(vc);
3255 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3256 (vc->vcore_state == VCORE_RUNNING ||
3257 vc->vcore_state == VCORE_EXITING ||
3258 vc->vcore_state == VCORE_PIGGYBACK))
3259 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3261 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3262 kvmppc_vcore_end_preempt(vc);
3264 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3265 kvmppc_remove_runnable(vc, vcpu);
3266 vcpu->stat.signal_exits++;
3267 kvm_run->exit_reason = KVM_EXIT_INTR;
3268 vcpu->arch.ret = -EINTR;
3271 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3272 /* Wake up some vcpu to run the core */
3274 v = next_runnable_thread(vc, &i);
3275 wake_up(&v->arch.cpu_run);
3278 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3279 spin_unlock(&vc->lock);
3280 return vcpu->arch.ret;
3283 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3287 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3288 unsigned long user_tar = 0;
3289 unsigned int user_vrsave;
3292 if (!vcpu->arch.sane) {
3293 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3298 * Don't allow entry with a suspended transaction, because
3299 * the guest entry/exit code will lose it.
3300 * If the guest has TM enabled, save away their TM-related SPRs
3301 * (they will get restored by the TM unavailable interrupt).
3303 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3304 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3305 (current->thread.regs->msr & MSR_TM)) {
3306 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3307 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3308 run->fail_entry.hardware_entry_failure_reason = 0;
3311 /* Enable TM so we can read the TM SPRs */
3312 mtmsr(mfmsr() | MSR_TM);
3313 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3314 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3315 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3316 current->thread.regs->msr &= ~MSR_TM;
3320 kvmppc_core_prepare_to_enter(vcpu);
3322 /* No need to go into the guest when all we'll do is come back out */
3323 if (signal_pending(current)) {
3324 run->exit_reason = KVM_EXIT_INTR;
3329 atomic_inc(&kvm->arch.vcpus_running);
3330 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3333 flush_all_to_thread(current);
3335 /* Save userspace EBB and other register values */
3336 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3337 ebb_regs[0] = mfspr(SPRN_EBBHR);
3338 ebb_regs[1] = mfspr(SPRN_EBBRR);
3339 ebb_regs[2] = mfspr(SPRN_BESCR);
3340 user_tar = mfspr(SPRN_TAR);
3342 user_vrsave = mfspr(SPRN_VRSAVE);
3344 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3345 vcpu->arch.pgdir = current->mm->pgd;
3346 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3349 r = kvmppc_run_vcpu(run, vcpu);
3351 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3352 !(vcpu->arch.shregs.msr & MSR_PR)) {
3353 trace_kvm_hcall_enter(vcpu);
3354 r = kvmppc_pseries_do_hcall(vcpu);
3355 trace_kvm_hcall_exit(vcpu, r);
3356 kvmppc_core_prepare_to_enter(vcpu);
3357 } else if (r == RESUME_PAGE_FAULT) {
3358 srcu_idx = srcu_read_lock(&kvm->srcu);
3359 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3360 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3361 srcu_read_unlock(&kvm->srcu, srcu_idx);
3362 } else if (r == RESUME_PASSTHROUGH) {
3363 if (WARN_ON(xive_enabled()))
3366 r = kvmppc_xics_rm_complete(vcpu, 0);
3368 } while (is_kvmppc_resume_guest(r));
3370 /* Restore userspace EBB and other register values */
3371 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3372 mtspr(SPRN_EBBHR, ebb_regs[0]);
3373 mtspr(SPRN_EBBRR, ebb_regs[1]);
3374 mtspr(SPRN_BESCR, ebb_regs[2]);
3375 mtspr(SPRN_TAR, user_tar);
3376 mtspr(SPRN_FSCR, current->thread.fscr);
3378 mtspr(SPRN_VRSAVE, user_vrsave);
3380 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3381 atomic_dec(&kvm->arch.vcpus_running);
3385 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3386 int shift, int sllp)
3388 (*sps)->page_shift = shift;
3389 (*sps)->slb_enc = sllp;
3390 (*sps)->enc[0].page_shift = shift;
3391 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3393 * Add 16MB MPSS support (may get filtered out by userspace)
3396 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3398 (*sps)->enc[1].page_shift = 24;
3399 (*sps)->enc[1].pte_enc = penc;
3405 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3406 struct kvm_ppc_smmu_info *info)
3408 struct kvm_ppc_one_seg_page_size *sps;
3411 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3412 * POWER7 doesn't support keys for instruction accesses,
3413 * POWER8 and POWER9 do.
3415 info->data_keys = 32;
3416 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3418 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3419 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3420 info->slb_size = 32;
3422 /* We only support these sizes for now, and no muti-size segments */
3423 sps = &info->sps[0];
3424 kvmppc_add_seg_page_size(&sps, 12, 0);
3425 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3426 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3432 * Get (and clear) the dirty memory log for a memory slot.
3434 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3435 struct kvm_dirty_log *log)
3437 struct kvm_memslots *slots;
3438 struct kvm_memory_slot *memslot;
3441 unsigned long *buf, *p;
3442 struct kvm_vcpu *vcpu;
3444 mutex_lock(&kvm->slots_lock);
3447 if (log->slot >= KVM_USER_MEM_SLOTS)
3450 slots = kvm_memslots(kvm);
3451 memslot = id_to_memslot(slots, log->slot);
3453 if (!memslot->dirty_bitmap)
3457 * Use second half of bitmap area because both HPT and radix
3458 * accumulate bits in the first half.
3460 n = kvm_dirty_bitmap_bytes(memslot);
3461 buf = memslot->dirty_bitmap + n / sizeof(long);
3464 if (kvm_is_radix(kvm))
3465 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3467 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3472 * We accumulate dirty bits in the first half of the
3473 * memslot's dirty_bitmap area, for when pages are paged
3474 * out or modified by the host directly. Pick up these
3475 * bits and add them to the map.
3477 p = memslot->dirty_bitmap;
3478 for (i = 0; i < n / sizeof(long); ++i)
3479 buf[i] |= xchg(&p[i], 0);
3481 /* Harvest dirty bits from VPA and DTL updates */
3482 /* Note: we never modify the SLB shadow buffer areas */
3483 kvm_for_each_vcpu(i, vcpu, kvm) {
3484 spin_lock(&vcpu->arch.vpa_update_lock);
3485 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3486 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3487 spin_unlock(&vcpu->arch.vpa_update_lock);
3491 if (copy_to_user(log->dirty_bitmap, buf, n))
3496 mutex_unlock(&kvm->slots_lock);
3500 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3501 struct kvm_memory_slot *dont)
3503 if (!dont || free->arch.rmap != dont->arch.rmap) {
3504 vfree(free->arch.rmap);
3505 free->arch.rmap = NULL;
3509 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3510 unsigned long npages)
3512 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3513 if (!slot->arch.rmap)
3519 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3520 struct kvm_memory_slot *memslot,
3521 const struct kvm_userspace_memory_region *mem)
3526 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3527 const struct kvm_userspace_memory_region *mem,
3528 const struct kvm_memory_slot *old,
3529 const struct kvm_memory_slot *new)
3531 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3534 * If we are making a new memslot, it might make
3535 * some address that was previously cached as emulated
3536 * MMIO be no longer emulated MMIO, so invalidate
3537 * all the caches of emulated MMIO translations.
3540 atomic64_inc(&kvm->arch.mmio_update);
3544 * Update LPCR values in kvm->arch and in vcores.
3545 * Caller must hold kvm->lock.
3547 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3552 if ((kvm->arch.lpcr & mask) == lpcr)
3555 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3557 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3558 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3561 spin_lock(&vc->lock);
3562 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3563 spin_unlock(&vc->lock);
3564 if (++cores_done >= kvm->arch.online_vcores)
3569 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3574 void kvmppc_setup_partition_table(struct kvm *kvm)
3576 unsigned long dw0, dw1;
3578 if (!kvm_is_radix(kvm)) {
3579 /* PS field - page size for VRMA */
3580 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3581 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3582 /* HTABSIZE and HTABORG fields */
3583 dw0 |= kvm->arch.sdr1;
3585 /* Second dword as set by userspace */
3586 dw1 = kvm->arch.process_table;
3588 dw0 = PATB_HR | radix__get_tree_size() |
3589 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3590 dw1 = PATB_GR | kvm->arch.process_table;
3593 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3597 * Set up HPT (hashed page table) and RMA (real-mode area).
3598 * Must be called with kvm->lock held.
3600 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3603 struct kvm *kvm = vcpu->kvm;
3605 struct kvm_memory_slot *memslot;
3606 struct vm_area_struct *vma;
3607 unsigned long lpcr = 0, senc;
3608 unsigned long psize, porder;
3611 /* Allocate hashed page table (if not done already) and reset it */
3612 if (!kvm->arch.hpt.virt) {
3613 int order = KVM_DEFAULT_HPT_ORDER;
3614 struct kvm_hpt_info info;
3616 err = kvmppc_allocate_hpt(&info, order);
3617 /* If we get here, it means userspace didn't specify a
3618 * size explicitly. So, try successively smaller
3619 * sizes if the default failed. */
3620 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3621 err = kvmppc_allocate_hpt(&info, order);
3624 pr_err("KVM: Couldn't alloc HPT\n");
3628 kvmppc_set_hpt(kvm, &info);
3631 /* Look up the memslot for guest physical address 0 */
3632 srcu_idx = srcu_read_lock(&kvm->srcu);
3633 memslot = gfn_to_memslot(kvm, 0);
3635 /* We must have some memory at 0 by now */
3637 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3640 /* Look up the VMA for the start of this memory slot */
3641 hva = memslot->userspace_addr;
3642 down_read(¤t->mm->mmap_sem);
3643 vma = find_vma(current->mm, hva);
3644 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3647 psize = vma_kernel_pagesize(vma);
3648 porder = __ilog2(psize);
3650 up_read(¤t->mm->mmap_sem);
3652 /* We can handle 4k, 64k or 16M pages in the VRMA */
3654 if (!(psize == 0x1000 || psize == 0x10000 ||
3655 psize == 0x1000000))
3658 senc = slb_pgsize_encoding(psize);
3659 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3660 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3661 /* Create HPTEs in the hash page table for the VRMA */
3662 kvmppc_map_vrma(vcpu, memslot, porder);
3664 /* Update VRMASD field in the LPCR */
3665 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3666 /* the -4 is to account for senc values starting at 0x10 */
3667 lpcr = senc << (LPCR_VRMASD_SH - 4);
3668 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3671 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3675 srcu_read_unlock(&kvm->srcu, srcu_idx);
3680 up_read(¤t->mm->mmap_sem);
3684 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3685 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3687 kvmppc_free_radix(kvm);
3688 kvmppc_update_lpcr(kvm, LPCR_VPM1,
3689 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3690 kvmppc_rmap_reset(kvm);
3691 kvm->arch.radix = 0;
3692 kvm->arch.process_table = 0;
3696 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3697 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3701 err = kvmppc_init_vm_radix(kvm);
3705 kvmppc_free_hpt(&kvm->arch.hpt);
3706 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3707 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3708 kvm->arch.radix = 1;
3712 #ifdef CONFIG_KVM_XICS
3714 * Allocate a per-core structure for managing state about which cores are
3715 * running in the host versus the guest and for exchanging data between
3716 * real mode KVM and CPU running in the host.
3717 * This is only done for the first VM.
3718 * The allocated structure stays even if all VMs have stopped.
3719 * It is only freed when the kvm-hv module is unloaded.
3720 * It's OK for this routine to fail, we just don't support host
3721 * core operations like redirecting H_IPI wakeups.
3723 void kvmppc_alloc_host_rm_ops(void)
3725 struct kvmppc_host_rm_ops *ops;
3726 unsigned long l_ops;
3730 /* Not the first time here ? */
3731 if (kvmppc_host_rm_ops_hv != NULL)
3734 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3738 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3739 ops->rm_core = kzalloc(size, GFP_KERNEL);
3741 if (!ops->rm_core) {
3748 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3749 if (!cpu_online(cpu))
3752 core = cpu >> threads_shift;
3753 ops->rm_core[core].rm_state.in_host = 1;
3756 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3759 * Make the contents of the kvmppc_host_rm_ops structure visible
3760 * to other CPUs before we assign it to the global variable.
3761 * Do an atomic assignment (no locks used here), but if someone
3762 * beats us to it, just free our copy and return.
3765 l_ops = (unsigned long) ops;
3767 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3769 kfree(ops->rm_core);
3774 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3775 "ppc/kvm_book3s:prepare",
3776 kvmppc_set_host_core,
3777 kvmppc_clear_host_core);
3781 void kvmppc_free_host_rm_ops(void)
3783 if (kvmppc_host_rm_ops_hv) {
3784 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3785 kfree(kvmppc_host_rm_ops_hv->rm_core);
3786 kfree(kvmppc_host_rm_ops_hv);
3787 kvmppc_host_rm_ops_hv = NULL;
3792 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3794 unsigned long lpcr, lpid;
3798 /* Allocate the guest's logical partition ID */
3800 lpid = kvmppc_alloc_lpid();
3803 kvm->arch.lpid = lpid;
3805 kvmppc_alloc_host_rm_ops();
3808 * Since we don't flush the TLB when tearing down a VM,
3809 * and this lpid might have previously been used,
3810 * make sure we flush on each core before running the new VM.
3811 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3812 * does this flush for us.
3814 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3815 cpumask_setall(&kvm->arch.need_tlb_flush);
3817 /* Start out with the default set of hcalls enabled */
3818 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3819 sizeof(kvm->arch.enabled_hcalls));
3821 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3822 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3824 /* Init LPCR for virtual RMA mode */
3825 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3826 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3827 lpcr &= LPCR_PECE | LPCR_LPES;
3828 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3829 LPCR_VPM0 | LPCR_VPM1;
3830 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3831 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3832 /* On POWER8 turn on online bit to enable PURR/SPURR */
3833 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3836 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3837 * Set HVICE bit to enable hypervisor virtualization interrupts.
3838 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3839 * be unnecessary but better safe than sorry in case we re-enable
3840 * EE in HV mode with this LPCR still set)
3842 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3844 lpcr |= LPCR_HVICE | LPCR_HEIC;
3847 * If xive is enabled, we route 0x500 interrupts directly
3855 * If the host uses radix, the guest starts out as radix.
3857 if (radix_enabled()) {
3858 kvm->arch.radix = 1;
3859 kvm->arch.mmu_ready = 1;
3861 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3862 ret = kvmppc_init_vm_radix(kvm);
3864 kvmppc_free_lpid(kvm->arch.lpid);
3867 kvmppc_setup_partition_table(kvm);
3870 kvm->arch.lpcr = lpcr;
3872 /* Initialization for future HPT resizes */
3873 kvm->arch.resize_hpt = NULL;
3876 * Work out how many sets the TLB has, for the use of
3877 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3879 if (radix_enabled())
3880 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
3881 else if (cpu_has_feature(CPU_FTR_ARCH_300))
3882 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
3883 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3884 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
3886 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
3889 * Track that we now have a HV mode VM active. This blocks secondary
3890 * CPU threads from coming online.
3891 * On POWER9, we only need to do this if the "indep_threads_mode"
3892 * module parameter has been set to N.
3894 if (cpu_has_feature(CPU_FTR_ARCH_300))
3895 kvm->arch.threads_indep = indep_threads_mode;
3896 if (!kvm->arch.threads_indep)
3897 kvm_hv_vm_activated();
3900 * Initialize smt_mode depending on processor.
3901 * POWER8 and earlier have to use "strict" threading, where
3902 * all vCPUs in a vcore have to run on the same (sub)core,
3903 * whereas on POWER9 the threads can each run a different
3906 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3907 kvm->arch.smt_mode = threads_per_subcore;
3909 kvm->arch.smt_mode = 1;
3910 kvm->arch.emul_smt_mode = 1;
3913 * Create a debugfs directory for the VM
3915 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3916 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3917 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3918 kvmppc_mmu_debugfs_init(kvm);
3923 static void kvmppc_free_vcores(struct kvm *kvm)
3927 for (i = 0; i < KVM_MAX_VCORES; ++i)
3928 kfree(kvm->arch.vcores[i]);
3929 kvm->arch.online_vcores = 0;
3932 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3934 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3936 if (!kvm->arch.threads_indep)
3937 kvm_hv_vm_deactivated();
3939 kvmppc_free_vcores(kvm);
3941 kvmppc_free_lpid(kvm->arch.lpid);
3943 if (kvm_is_radix(kvm))
3944 kvmppc_free_radix(kvm);
3946 kvmppc_free_hpt(&kvm->arch.hpt);
3948 kvmppc_free_pimap(kvm);
3951 /* We don't need to emulate any privileged instructions or dcbz */
3952 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3953 unsigned int inst, int *advance)
3955 return EMULATE_FAIL;
3958 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3961 return EMULATE_FAIL;
3964 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3967 return EMULATE_FAIL;
3970 static int kvmppc_core_check_processor_compat_hv(void)
3972 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3973 !cpu_has_feature(CPU_FTR_ARCH_206))
3979 #ifdef CONFIG_KVM_XICS
3981 void kvmppc_free_pimap(struct kvm *kvm)
3983 kfree(kvm->arch.pimap);
3986 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3988 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3991 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3993 struct irq_desc *desc;
3994 struct kvmppc_irq_map *irq_map;
3995 struct kvmppc_passthru_irqmap *pimap;
3996 struct irq_chip *chip;
3999 if (!kvm_irq_bypass)
4002 desc = irq_to_desc(host_irq);
4006 mutex_lock(&kvm->lock);
4008 pimap = kvm->arch.pimap;
4009 if (pimap == NULL) {
4010 /* First call, allocate structure to hold IRQ map */
4011 pimap = kvmppc_alloc_pimap();
4012 if (pimap == NULL) {
4013 mutex_unlock(&kvm->lock);
4016 kvm->arch.pimap = pimap;
4020 * For now, we only support interrupts for which the EOI operation
4021 * is an OPAL call followed by a write to XIRR, since that's
4022 * what our real-mode EOI code does, or a XIVE interrupt
4024 chip = irq_data_get_irq_chip(&desc->irq_data);
4025 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4026 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4027 host_irq, guest_gsi);
4028 mutex_unlock(&kvm->lock);
4033 * See if we already have an entry for this guest IRQ number.
4034 * If it's mapped to a hardware IRQ number, that's an error,
4035 * otherwise re-use this entry.
4037 for (i = 0; i < pimap->n_mapped; i++) {
4038 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4039 if (pimap->mapped[i].r_hwirq) {
4040 mutex_unlock(&kvm->lock);
4047 if (i == KVMPPC_PIRQ_MAPPED) {
4048 mutex_unlock(&kvm->lock);
4049 return -EAGAIN; /* table is full */
4052 irq_map = &pimap->mapped[i];
4054 irq_map->v_hwirq = guest_gsi;
4055 irq_map->desc = desc;
4058 * Order the above two stores before the next to serialize with
4059 * the KVM real mode handler.
4062 irq_map->r_hwirq = desc->irq_data.hwirq;
4064 if (i == pimap->n_mapped)
4068 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4070 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4072 irq_map->r_hwirq = 0;
4074 mutex_unlock(&kvm->lock);
4079 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4081 struct irq_desc *desc;
4082 struct kvmppc_passthru_irqmap *pimap;
4085 if (!kvm_irq_bypass)
4088 desc = irq_to_desc(host_irq);
4092 mutex_lock(&kvm->lock);
4093 if (!kvm->arch.pimap)
4096 pimap = kvm->arch.pimap;
4098 for (i = 0; i < pimap->n_mapped; i++) {
4099 if (guest_gsi == pimap->mapped[i].v_hwirq)
4103 if (i == pimap->n_mapped) {
4104 mutex_unlock(&kvm->lock);
4109 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4111 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4113 /* invalidate the entry (what do do on error from the above ?) */
4114 pimap->mapped[i].r_hwirq = 0;
4117 * We don't free this structure even when the count goes to
4118 * zero. The structure is freed when we destroy the VM.
4121 mutex_unlock(&kvm->lock);
4125 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4126 struct irq_bypass_producer *prod)
4129 struct kvm_kernel_irqfd *irqfd =
4130 container_of(cons, struct kvm_kernel_irqfd, consumer);
4132 irqfd->producer = prod;
4134 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4136 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4137 prod->irq, irqfd->gsi, ret);
4142 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4143 struct irq_bypass_producer *prod)
4146 struct kvm_kernel_irqfd *irqfd =
4147 container_of(cons, struct kvm_kernel_irqfd, consumer);
4149 irqfd->producer = NULL;
4152 * When producer of consumer is unregistered, we change back to
4153 * default external interrupt handling mode - KVM real mode
4154 * will switch back to host.
4156 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4158 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4159 prod->irq, irqfd->gsi, ret);
4163 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4164 unsigned int ioctl, unsigned long arg)
4166 struct kvm *kvm __maybe_unused = filp->private_data;
4167 void __user *argp = (void __user *)arg;
4172 case KVM_PPC_ALLOCATE_HTAB: {
4176 if (get_user(htab_order, (u32 __user *)argp))
4178 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4185 case KVM_PPC_GET_HTAB_FD: {
4186 struct kvm_get_htab_fd ghf;
4189 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4191 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4195 case KVM_PPC_RESIZE_HPT_PREPARE: {
4196 struct kvm_ppc_resize_hpt rhpt;
4199 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4202 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4206 case KVM_PPC_RESIZE_HPT_COMMIT: {
4207 struct kvm_ppc_resize_hpt rhpt;
4210 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4213 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4225 * List of hcall numbers to enable by default.
4226 * For compatibility with old userspace, we enable by default
4227 * all hcalls that were implemented before the hcall-enabling
4228 * facility was added. Note this list should not include H_RTAS.
4230 static unsigned int default_hcall_list[] = {
4244 #ifdef CONFIG_KVM_XICS
4255 static void init_default_hcalls(void)
4260 for (i = 0; default_hcall_list[i]; ++i) {
4261 hcall = default_hcall_list[i];
4262 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4263 __set_bit(hcall / 4, default_enabled_hcalls);
4267 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4273 /* If not on a POWER9, reject it */
4274 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4277 /* If any unknown flags set, reject it */
4278 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4281 /* GR (guest radix) bit in process_table field must match */
4282 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4283 if (!!(cfg->process_table & PATB_GR) != radix)
4286 /* Process table size field must be reasonable, i.e. <= 24 */
4287 if ((cfg->process_table & PRTS_MASK) > 24)
4290 /* We can change a guest to/from radix now, if the host is radix */
4291 if (radix && !radix_enabled())
4294 mutex_lock(&kvm->lock);
4295 if (radix != kvm_is_radix(kvm)) {
4296 if (kvm->arch.mmu_ready) {
4297 kvm->arch.mmu_ready = 0;
4298 /* order mmu_ready vs. vcpus_running */
4300 if (atomic_read(&kvm->arch.vcpus_running)) {
4301 kvm->arch.mmu_ready = 1;
4307 err = kvmppc_switch_mmu_to_radix(kvm);
4309 err = kvmppc_switch_mmu_to_hpt(kvm);
4314 kvm->arch.process_table = cfg->process_table;
4315 kvmppc_setup_partition_table(kvm);
4317 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4318 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4322 mutex_unlock(&kvm->lock);
4326 static struct kvmppc_ops kvm_ops_hv = {
4327 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4328 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4329 .get_one_reg = kvmppc_get_one_reg_hv,
4330 .set_one_reg = kvmppc_set_one_reg_hv,
4331 .vcpu_load = kvmppc_core_vcpu_load_hv,
4332 .vcpu_put = kvmppc_core_vcpu_put_hv,
4333 .set_msr = kvmppc_set_msr_hv,
4334 .vcpu_run = kvmppc_vcpu_run_hv,
4335 .vcpu_create = kvmppc_core_vcpu_create_hv,
4336 .vcpu_free = kvmppc_core_vcpu_free_hv,
4337 .check_requests = kvmppc_core_check_requests_hv,
4338 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4339 .flush_memslot = kvmppc_core_flush_memslot_hv,
4340 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4341 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4342 .unmap_hva = kvm_unmap_hva_hv,
4343 .unmap_hva_range = kvm_unmap_hva_range_hv,
4344 .age_hva = kvm_age_hva_hv,
4345 .test_age_hva = kvm_test_age_hva_hv,
4346 .set_spte_hva = kvm_set_spte_hva_hv,
4347 .mmu_destroy = kvmppc_mmu_destroy_hv,
4348 .free_memslot = kvmppc_core_free_memslot_hv,
4349 .create_memslot = kvmppc_core_create_memslot_hv,
4350 .init_vm = kvmppc_core_init_vm_hv,
4351 .destroy_vm = kvmppc_core_destroy_vm_hv,
4352 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4353 .emulate_op = kvmppc_core_emulate_op_hv,
4354 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4355 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4356 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4357 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4358 .hcall_implemented = kvmppc_hcall_impl_hv,
4359 #ifdef CONFIG_KVM_XICS
4360 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4361 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4363 .configure_mmu = kvmhv_configure_mmu,
4364 .get_rmmu_info = kvmhv_get_rmmu_info,
4365 .set_smt_mode = kvmhv_set_smt_mode,
4368 static int kvm_init_subcore_bitmap(void)
4371 int nr_cores = cpu_nr_cores();
4372 struct sibling_subcore_state *sibling_subcore_state;
4374 for (i = 0; i < nr_cores; i++) {
4375 int first_cpu = i * threads_per_core;
4376 int node = cpu_to_node(first_cpu);
4378 /* Ignore if it is already allocated. */
4379 if (paca[first_cpu].sibling_subcore_state)
4382 sibling_subcore_state =
4383 kmalloc_node(sizeof(struct sibling_subcore_state),
4385 if (!sibling_subcore_state)
4388 memset(sibling_subcore_state, 0,
4389 sizeof(struct sibling_subcore_state));
4391 for (j = 0; j < threads_per_core; j++) {
4392 int cpu = first_cpu + j;
4394 paca[cpu].sibling_subcore_state = sibling_subcore_state;
4400 static int kvmppc_radix_possible(void)
4402 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4405 static int kvmppc_book3s_init_hv(void)
4409 * FIXME!! Do we need to check on all cpus ?
4411 r = kvmppc_core_check_processor_compat_hv();
4415 r = kvm_init_subcore_bitmap();
4420 * We need a way of accessing the XICS interrupt controller,
4421 * either directly, via paca[cpu].kvm_hstate.xics_phys, or
4422 * indirectly, via OPAL.
4425 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4426 struct device_node *np;
4428 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4430 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4436 kvm_ops_hv.owner = THIS_MODULE;
4437 kvmppc_hv_ops = &kvm_ops_hv;
4439 init_default_hcalls();
4443 r = kvmppc_mmu_hv_init();
4447 if (kvmppc_radix_possible())
4448 r = kvmppc_radix_init();
4452 static void kvmppc_book3s_exit_hv(void)
4454 kvmppc_free_host_rm_ops();
4455 if (kvmppc_radix_possible())
4456 kvmppc_radix_exit();
4457 kvmppc_hv_ops = NULL;
4460 module_init(kvmppc_book3s_init_hv);
4461 module_exit(kvmppc_book3s_exit_hv);
4462 MODULE_LICENSE("GPL");
4463 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4464 MODULE_ALIAS("devname:kvm");