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, S_IRUGO | S_IWUSR);
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, S_IRUGO | S_IWUSR);
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,
114 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
116 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
118 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
121 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
122 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
123 static void kvmppc_setup_partition_table(struct kvm *kvm);
125 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
129 struct kvm_vcpu *vcpu;
131 while (++i < MAX_SMT_THREADS) {
132 vcpu = READ_ONCE(vc->runnable_threads[i]);
141 /* Used to traverse the list of runnable threads for a given vcore */
142 #define for_each_runnable_thread(i, vcpu, vc) \
143 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
145 static bool kvmppc_ipi_thread(int cpu)
147 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
149 /* On POWER9 we can use msgsnd to IPI any cpu */
150 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
151 msg |= get_hard_smp_processor_id(cpu);
153 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
157 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
158 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
160 if (cpu_first_thread_sibling(cpu) ==
161 cpu_first_thread_sibling(smp_processor_id())) {
162 msg |= cpu_thread_in_core(cpu);
164 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
171 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
172 if (cpu >= 0 && cpu < nr_cpu_ids) {
173 if (paca[cpu].kvm_hstate.xics_phys) {
177 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
185 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
188 struct swait_queue_head *wqp;
190 wqp = kvm_arch_vcpu_wq(vcpu);
191 if (swq_has_sleeper(wqp)) {
193 ++vcpu->stat.halt_wakeup;
196 cpu = READ_ONCE(vcpu->arch.thread_cpu);
197 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
200 /* CPU points to the first thread of the core */
202 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
203 smp_send_reschedule(cpu);
207 * We use the vcpu_load/put functions to measure stolen time.
208 * Stolen time is counted as time when either the vcpu is able to
209 * run as part of a virtual core, but the task running the vcore
210 * is preempted or sleeping, or when the vcpu needs something done
211 * in the kernel by the task running the vcpu, but that task is
212 * preempted or sleeping. Those two things have to be counted
213 * separately, since one of the vcpu tasks will take on the job
214 * of running the core, and the other vcpu tasks in the vcore will
215 * sleep waiting for it to do that, but that sleep shouldn't count
218 * Hence we accumulate stolen time when the vcpu can run as part of
219 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
220 * needs its task to do other things in the kernel (for example,
221 * service a page fault) in busy_stolen. We don't accumulate
222 * stolen time for a vcore when it is inactive, or for a vcpu
223 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
224 * a misnomer; it means that the vcpu task is not executing in
225 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
226 * the kernel. We don't have any way of dividing up that time
227 * between time that the vcpu is genuinely stopped, time that
228 * the task is actively working on behalf of the vcpu, and time
229 * that the task is preempted, so we don't count any of it as
232 * Updates to busy_stolen are protected by arch.tbacct_lock;
233 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
234 * lock. The stolen times are measured in units of timebase ticks.
235 * (Note that the != TB_NIL checks below are purely defensive;
236 * they should never fail.)
239 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
243 spin_lock_irqsave(&vc->stoltb_lock, flags);
244 vc->preempt_tb = mftb();
245 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
248 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
252 spin_lock_irqsave(&vc->stoltb_lock, flags);
253 if (vc->preempt_tb != TB_NIL) {
254 vc->stolen_tb += mftb() - vc->preempt_tb;
255 vc->preempt_tb = TB_NIL;
257 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
260 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
262 struct kvmppc_vcore *vc = vcpu->arch.vcore;
266 * We can test vc->runner without taking the vcore lock,
267 * because only this task ever sets vc->runner to this
268 * vcpu, and once it is set to this vcpu, only this task
269 * ever sets it to NULL.
271 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
272 kvmppc_core_end_stolen(vc);
274 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
275 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
276 vcpu->arch.busy_preempt != TB_NIL) {
277 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
278 vcpu->arch.busy_preempt = TB_NIL;
280 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
283 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
285 struct kvmppc_vcore *vc = vcpu->arch.vcore;
288 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
289 kvmppc_core_start_stolen(vc);
291 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
292 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
293 vcpu->arch.busy_preempt = mftb();
294 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
297 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
300 * Check for illegal transactional state bit combination
301 * and if we find it, force the TS field to a safe state.
303 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
305 vcpu->arch.shregs.msr = msr;
306 kvmppc_end_cede(vcpu);
309 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
311 vcpu->arch.pvr = pvr;
314 /* Dummy value used in computing PCR value below */
315 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
317 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
319 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
320 struct kvmppc_vcore *vc = vcpu->arch.vcore;
322 /* We can (emulate) our own architecture version and anything older */
323 if (cpu_has_feature(CPU_FTR_ARCH_300))
324 host_pcr_bit = PCR_ARCH_300;
325 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
326 host_pcr_bit = PCR_ARCH_207;
327 else if (cpu_has_feature(CPU_FTR_ARCH_206))
328 host_pcr_bit = PCR_ARCH_206;
330 host_pcr_bit = PCR_ARCH_205;
332 /* Determine lowest PCR bit needed to run guest in given PVR level */
333 guest_pcr_bit = host_pcr_bit;
335 switch (arch_compat) {
337 guest_pcr_bit = PCR_ARCH_205;
341 guest_pcr_bit = PCR_ARCH_206;
344 guest_pcr_bit = PCR_ARCH_207;
347 guest_pcr_bit = PCR_ARCH_300;
354 /* Check requested PCR bits don't exceed our capabilities */
355 if (guest_pcr_bit > host_pcr_bit)
358 spin_lock(&vc->lock);
359 vc->arch_compat = arch_compat;
360 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
361 vc->pcr = host_pcr_bit - guest_pcr_bit;
362 spin_unlock(&vc->lock);
367 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
371 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
372 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
373 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
374 for (r = 0; r < 16; ++r)
375 pr_err("r%2d = %.16lx r%d = %.16lx\n",
376 r, kvmppc_get_gpr(vcpu, r),
377 r+16, kvmppc_get_gpr(vcpu, r+16));
378 pr_err("ctr = %.16lx lr = %.16lx\n",
379 vcpu->arch.ctr, vcpu->arch.lr);
380 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
381 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
382 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
383 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
384 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
385 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
386 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
387 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
388 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
389 pr_err("fault dar = %.16lx dsisr = %.8x\n",
390 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
391 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
392 for (r = 0; r < vcpu->arch.slb_max; ++r)
393 pr_err(" ESID = %.16llx VSID = %.16llx\n",
394 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
395 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
396 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
397 vcpu->arch.last_inst);
400 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
402 struct kvm_vcpu *ret;
404 mutex_lock(&kvm->lock);
405 ret = kvm_get_vcpu_by_id(kvm, id);
406 mutex_unlock(&kvm->lock);
410 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
412 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
413 vpa->yield_count = cpu_to_be32(1);
416 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
417 unsigned long addr, unsigned long len)
419 /* check address is cacheline aligned */
420 if (addr & (L1_CACHE_BYTES - 1))
422 spin_lock(&vcpu->arch.vpa_update_lock);
423 if (v->next_gpa != addr || v->len != len) {
425 v->len = addr ? len : 0;
426 v->update_pending = 1;
428 spin_unlock(&vcpu->arch.vpa_update_lock);
432 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
441 static int vpa_is_registered(struct kvmppc_vpa *vpap)
443 if (vpap->update_pending)
444 return vpap->next_gpa != 0;
445 return vpap->pinned_addr != NULL;
448 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
450 unsigned long vcpuid, unsigned long vpa)
452 struct kvm *kvm = vcpu->kvm;
453 unsigned long len, nb;
455 struct kvm_vcpu *tvcpu;
458 struct kvmppc_vpa *vpap;
460 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
464 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
465 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
466 subfunc == H_VPA_REG_SLB) {
467 /* Registering new area - address must be cache-line aligned */
468 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
471 /* convert logical addr to kernel addr and read length */
472 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
475 if (subfunc == H_VPA_REG_VPA)
476 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
478 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
479 kvmppc_unpin_guest_page(kvm, va, vpa, false);
482 if (len > nb || len < sizeof(struct reg_vpa))
491 spin_lock(&tvcpu->arch.vpa_update_lock);
494 case H_VPA_REG_VPA: /* register VPA */
496 * The size of our lppaca is 1kB because of the way we align
497 * it for the guest to avoid crossing a 4kB boundary. We only
498 * use 640 bytes of the structure though, so we should accept
499 * clients that set a size of 640.
503 vpap = &tvcpu->arch.vpa;
507 case H_VPA_REG_DTL: /* register DTL */
508 if (len < sizeof(struct dtl_entry))
510 len -= len % sizeof(struct dtl_entry);
512 /* Check that they have previously registered a VPA */
514 if (!vpa_is_registered(&tvcpu->arch.vpa))
517 vpap = &tvcpu->arch.dtl;
521 case H_VPA_REG_SLB: /* register SLB shadow buffer */
522 /* Check that they have previously registered a VPA */
524 if (!vpa_is_registered(&tvcpu->arch.vpa))
527 vpap = &tvcpu->arch.slb_shadow;
531 case H_VPA_DEREG_VPA: /* deregister VPA */
532 /* Check they don't still have a DTL or SLB buf registered */
534 if (vpa_is_registered(&tvcpu->arch.dtl) ||
535 vpa_is_registered(&tvcpu->arch.slb_shadow))
538 vpap = &tvcpu->arch.vpa;
542 case H_VPA_DEREG_DTL: /* deregister DTL */
543 vpap = &tvcpu->arch.dtl;
547 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
548 vpap = &tvcpu->arch.slb_shadow;
554 vpap->next_gpa = vpa;
556 vpap->update_pending = 1;
559 spin_unlock(&tvcpu->arch.vpa_update_lock);
564 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
566 struct kvm *kvm = vcpu->kvm;
572 * We need to pin the page pointed to by vpap->next_gpa,
573 * but we can't call kvmppc_pin_guest_page under the lock
574 * as it does get_user_pages() and down_read(). So we
575 * have to drop the lock, pin the page, then get the lock
576 * again and check that a new area didn't get registered
580 gpa = vpap->next_gpa;
581 spin_unlock(&vcpu->arch.vpa_update_lock);
585 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
586 spin_lock(&vcpu->arch.vpa_update_lock);
587 if (gpa == vpap->next_gpa)
589 /* sigh... unpin that one and try again */
591 kvmppc_unpin_guest_page(kvm, va, gpa, false);
594 vpap->update_pending = 0;
595 if (va && nb < vpap->len) {
597 * If it's now too short, it must be that userspace
598 * has changed the mappings underlying guest memory,
599 * so unregister the region.
601 kvmppc_unpin_guest_page(kvm, va, gpa, false);
604 if (vpap->pinned_addr)
605 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
608 vpap->pinned_addr = va;
611 vpap->pinned_end = va + vpap->len;
614 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
616 if (!(vcpu->arch.vpa.update_pending ||
617 vcpu->arch.slb_shadow.update_pending ||
618 vcpu->arch.dtl.update_pending))
621 spin_lock(&vcpu->arch.vpa_update_lock);
622 if (vcpu->arch.vpa.update_pending) {
623 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
624 if (vcpu->arch.vpa.pinned_addr)
625 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
627 if (vcpu->arch.dtl.update_pending) {
628 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
629 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
630 vcpu->arch.dtl_index = 0;
632 if (vcpu->arch.slb_shadow.update_pending)
633 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
634 spin_unlock(&vcpu->arch.vpa_update_lock);
638 * Return the accumulated stolen time for the vcore up until `now'.
639 * The caller should hold the vcore lock.
641 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
646 spin_lock_irqsave(&vc->stoltb_lock, flags);
648 if (vc->vcore_state != VCORE_INACTIVE &&
649 vc->preempt_tb != TB_NIL)
650 p += now - vc->preempt_tb;
651 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
655 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
656 struct kvmppc_vcore *vc)
658 struct dtl_entry *dt;
660 unsigned long stolen;
661 unsigned long core_stolen;
665 dt = vcpu->arch.dtl_ptr;
666 vpa = vcpu->arch.vpa.pinned_addr;
668 core_stolen = vcore_stolen_time(vc, now);
669 stolen = core_stolen - vcpu->arch.stolen_logged;
670 vcpu->arch.stolen_logged = core_stolen;
671 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
672 stolen += vcpu->arch.busy_stolen;
673 vcpu->arch.busy_stolen = 0;
674 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
677 memset(dt, 0, sizeof(struct dtl_entry));
678 dt->dispatch_reason = 7;
679 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
680 dt->timebase = cpu_to_be64(now + vc->tb_offset);
681 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
682 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
683 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
685 if (dt == vcpu->arch.dtl.pinned_end)
686 dt = vcpu->arch.dtl.pinned_addr;
687 vcpu->arch.dtl_ptr = dt;
688 /* order writing *dt vs. writing vpa->dtl_idx */
690 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
691 vcpu->arch.dtl.dirty = true;
694 /* See if there is a doorbell interrupt pending for a vcpu */
695 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
698 struct kvmppc_vcore *vc;
700 if (vcpu->arch.doorbell_request)
703 * Ensure that the read of vcore->dpdes comes after the read
704 * of vcpu->doorbell_request. This barrier matches the
705 * lwsync in book3s_hv_rmhandlers.S just before the
706 * fast_guest_return label.
709 vc = vcpu->arch.vcore;
710 thr = vcpu->vcpu_id - vc->first_vcpuid;
711 return !!(vc->dpdes & (1 << thr));
714 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
716 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
718 if ((!vcpu->arch.vcore->arch_compat) &&
719 cpu_has_feature(CPU_FTR_ARCH_207S))
724 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
725 unsigned long resource, unsigned long value1,
726 unsigned long value2)
729 case H_SET_MODE_RESOURCE_SET_CIABR:
730 if (!kvmppc_power8_compatible(vcpu))
735 return H_UNSUPPORTED_FLAG_START;
736 /* Guests can't breakpoint the hypervisor */
737 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
739 vcpu->arch.ciabr = value1;
741 case H_SET_MODE_RESOURCE_SET_DAWR:
742 if (!kvmppc_power8_compatible(vcpu))
745 return H_UNSUPPORTED_FLAG_START;
746 if (value2 & DABRX_HYP)
748 vcpu->arch.dawr = value1;
749 vcpu->arch.dawrx = value2;
756 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
758 struct kvmppc_vcore *vcore = target->arch.vcore;
761 * We expect to have been called by the real mode handler
762 * (kvmppc_rm_h_confer()) which would have directly returned
763 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
764 * have useful work to do and should not confer) so we don't
768 spin_lock(&vcore->lock);
769 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
770 vcore->vcore_state != VCORE_INACTIVE &&
772 target = vcore->runner;
773 spin_unlock(&vcore->lock);
775 return kvm_vcpu_yield_to(target);
778 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
781 struct lppaca *lppaca;
783 spin_lock(&vcpu->arch.vpa_update_lock);
784 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
786 yield_count = be32_to_cpu(lppaca->yield_count);
787 spin_unlock(&vcpu->arch.vpa_update_lock);
791 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
793 unsigned long req = kvmppc_get_gpr(vcpu, 3);
794 unsigned long target, ret = H_SUCCESS;
796 struct kvm_vcpu *tvcpu;
799 if (req <= MAX_HCALL_OPCODE &&
800 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
807 target = kvmppc_get_gpr(vcpu, 4);
808 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
813 tvcpu->arch.prodded = 1;
815 if (tvcpu->arch.ceded)
816 kvmppc_fast_vcpu_kick_hv(tvcpu);
819 target = kvmppc_get_gpr(vcpu, 4);
822 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
827 yield_count = kvmppc_get_gpr(vcpu, 5);
828 if (kvmppc_get_yield_count(tvcpu) != yield_count)
830 kvm_arch_vcpu_yield_to(tvcpu);
833 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
834 kvmppc_get_gpr(vcpu, 5),
835 kvmppc_get_gpr(vcpu, 6));
838 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
841 idx = srcu_read_lock(&vcpu->kvm->srcu);
842 rc = kvmppc_rtas_hcall(vcpu);
843 srcu_read_unlock(&vcpu->kvm->srcu, idx);
850 /* Send the error out to userspace via KVM_RUN */
852 case H_LOGICAL_CI_LOAD:
853 ret = kvmppc_h_logical_ci_load(vcpu);
854 if (ret == H_TOO_HARD)
857 case H_LOGICAL_CI_STORE:
858 ret = kvmppc_h_logical_ci_store(vcpu);
859 if (ret == H_TOO_HARD)
863 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
864 kvmppc_get_gpr(vcpu, 5),
865 kvmppc_get_gpr(vcpu, 6),
866 kvmppc_get_gpr(vcpu, 7));
867 if (ret == H_TOO_HARD)
876 if (kvmppc_xics_enabled(vcpu)) {
877 if (xive_enabled()) {
878 ret = H_NOT_AVAILABLE;
881 ret = kvmppc_xics_hcall(vcpu, req);
886 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
887 kvmppc_get_gpr(vcpu, 5),
888 kvmppc_get_gpr(vcpu, 6));
889 if (ret == H_TOO_HARD)
892 case H_PUT_TCE_INDIRECT:
893 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
894 kvmppc_get_gpr(vcpu, 5),
895 kvmppc_get_gpr(vcpu, 6),
896 kvmppc_get_gpr(vcpu, 7));
897 if (ret == H_TOO_HARD)
901 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
902 kvmppc_get_gpr(vcpu, 5),
903 kvmppc_get_gpr(vcpu, 6),
904 kvmppc_get_gpr(vcpu, 7));
905 if (ret == H_TOO_HARD)
911 kvmppc_set_gpr(vcpu, 3, ret);
912 vcpu->arch.hcall_needed = 0;
916 static int kvmppc_hcall_impl_hv(unsigned long cmd)
924 case H_LOGICAL_CI_LOAD:
925 case H_LOGICAL_CI_STORE:
926 #ifdef CONFIG_KVM_XICS
937 /* See if it's in the real-mode table */
938 return kvmppc_hcall_impl_hv_realmode(cmd);
941 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
942 struct kvm_vcpu *vcpu)
946 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
949 * Fetch failed, so return to guest and
950 * try executing it again.
955 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
956 run->exit_reason = KVM_EXIT_DEBUG;
957 run->debug.arch.address = kvmppc_get_pc(vcpu);
960 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
965 static void do_nothing(void *x)
969 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
971 int thr, cpu, pcpu, nthreads;
975 nthreads = vcpu->kvm->arch.emul_smt_mode;
977 cpu = vcpu->vcpu_id & ~(nthreads - 1);
978 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
979 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
983 * If the vcpu is currently running on a physical cpu thread,
984 * interrupt it in order to pull it out of the guest briefly,
985 * which will update its vcore->dpdes value.
987 pcpu = READ_ONCE(v->cpu);
989 smp_call_function_single(pcpu, do_nothing, NULL, 1);
990 if (kvmppc_doorbell_pending(v))
997 * On POWER9, emulate doorbell-related instructions in order to
998 * give the guest the illusion of running on a multi-threaded core.
999 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1002 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1006 struct kvm *kvm = vcpu->kvm;
1007 struct kvm_vcpu *tvcpu;
1009 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1010 return EMULATE_FAIL;
1011 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1012 return RESUME_GUEST;
1013 if (get_op(inst) != 31)
1014 return EMULATE_FAIL;
1016 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1017 switch (get_xop(inst)) {
1018 case OP_31_XOP_MSGSNDP:
1019 arg = kvmppc_get_gpr(vcpu, rb);
1020 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1023 if (arg >= kvm->arch.emul_smt_mode)
1025 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1028 if (!tvcpu->arch.doorbell_request) {
1029 tvcpu->arch.doorbell_request = 1;
1030 kvmppc_fast_vcpu_kick_hv(tvcpu);
1033 case OP_31_XOP_MSGCLRP:
1034 arg = kvmppc_get_gpr(vcpu, rb);
1035 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1037 vcpu->arch.vcore->dpdes = 0;
1038 vcpu->arch.doorbell_request = 0;
1040 case OP_31_XOP_MFSPR:
1041 switch (get_sprn(inst)) {
1046 arg = kvmppc_read_dpdes(vcpu);
1049 return EMULATE_FAIL;
1051 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1054 return EMULATE_FAIL;
1056 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1057 return RESUME_GUEST;
1060 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1061 struct task_struct *tsk)
1063 int r = RESUME_HOST;
1065 vcpu->stat.sum_exits++;
1068 * This can happen if an interrupt occurs in the last stages
1069 * of guest entry or the first stages of guest exit (i.e. after
1070 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1071 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1072 * That can happen due to a bug, or due to a machine check
1073 * occurring at just the wrong time.
1075 if (vcpu->arch.shregs.msr & MSR_HV) {
1076 printk(KERN_EMERG "KVM trap in HV mode!\n");
1077 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1078 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1079 vcpu->arch.shregs.msr);
1080 kvmppc_dump_regs(vcpu);
1081 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1082 run->hw.hardware_exit_reason = vcpu->arch.trap;
1085 run->exit_reason = KVM_EXIT_UNKNOWN;
1086 run->ready_for_interrupt_injection = 1;
1087 switch (vcpu->arch.trap) {
1088 /* We're good on these - the host merely wanted to get our attention */
1089 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1090 vcpu->stat.dec_exits++;
1093 case BOOK3S_INTERRUPT_EXTERNAL:
1094 case BOOK3S_INTERRUPT_H_DOORBELL:
1095 case BOOK3S_INTERRUPT_H_VIRT:
1096 vcpu->stat.ext_intr_exits++;
1099 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1100 case BOOK3S_INTERRUPT_HMI:
1101 case BOOK3S_INTERRUPT_PERFMON:
1102 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1105 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1106 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1107 run->exit_reason = KVM_EXIT_NMI;
1108 run->hw.hardware_exit_reason = vcpu->arch.trap;
1109 /* Clear out the old NMI status from run->flags */
1110 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1111 /* Now set the NMI status */
1112 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1113 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1115 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1118 /* Print the MCE event to host console. */
1119 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1121 case BOOK3S_INTERRUPT_PROGRAM:
1125 * Normally program interrupts are delivered directly
1126 * to the guest by the hardware, but we can get here
1127 * as a result of a hypervisor emulation interrupt
1128 * (e40) getting turned into a 700 by BML RTAS.
1130 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1131 kvmppc_core_queue_program(vcpu, flags);
1135 case BOOK3S_INTERRUPT_SYSCALL:
1137 /* hcall - punt to userspace */
1140 /* hypercall with MSR_PR has already been handled in rmode,
1141 * and never reaches here.
1144 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1145 for (i = 0; i < 9; ++i)
1146 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1147 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1148 vcpu->arch.hcall_needed = 1;
1153 * We get these next two if the guest accesses a page which it thinks
1154 * it has mapped but which is not actually present, either because
1155 * it is for an emulated I/O device or because the corresonding
1156 * host page has been paged out. Any other HDSI/HISI interrupts
1157 * have been handled already.
1159 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1160 r = RESUME_PAGE_FAULT;
1162 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1163 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1164 vcpu->arch.fault_dsisr = 0;
1165 r = RESUME_PAGE_FAULT;
1168 * This occurs if the guest executes an illegal instruction.
1169 * If the guest debug is disabled, generate a program interrupt
1170 * to the guest. If guest debug is enabled, we need to check
1171 * whether the instruction is a software breakpoint instruction.
1172 * Accordingly return to Guest or Host.
1174 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1175 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1176 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1177 swab32(vcpu->arch.emul_inst) :
1178 vcpu->arch.emul_inst;
1179 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1180 r = kvmppc_emulate_debug_inst(run, vcpu);
1182 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1187 * This occurs if the guest (kernel or userspace), does something that
1188 * is prohibited by HFSCR.
1189 * On POWER9, this could be a doorbell instruction that we need
1191 * Otherwise, we just generate a program interrupt to the guest.
1193 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1195 if ((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG)
1196 r = kvmppc_emulate_doorbell_instr(vcpu);
1197 if (r == EMULATE_FAIL) {
1198 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1202 case BOOK3S_INTERRUPT_HV_RM_HARD:
1203 r = RESUME_PASSTHROUGH;
1206 kvmppc_dump_regs(vcpu);
1207 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1208 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1209 vcpu->arch.shregs.msr);
1210 run->hw.hardware_exit_reason = vcpu->arch.trap;
1218 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1219 struct kvm_sregs *sregs)
1223 memset(sregs, 0, sizeof(struct kvm_sregs));
1224 sregs->pvr = vcpu->arch.pvr;
1225 for (i = 0; i < vcpu->arch.slb_max; i++) {
1226 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1227 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1233 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1234 struct kvm_sregs *sregs)
1238 /* Only accept the same PVR as the host's, since we can't spoof it */
1239 if (sregs->pvr != vcpu->arch.pvr)
1243 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1244 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1245 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1246 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1250 vcpu->arch.slb_max = j;
1255 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1256 bool preserve_top32)
1258 struct kvm *kvm = vcpu->kvm;
1259 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1262 mutex_lock(&kvm->lock);
1263 spin_lock(&vc->lock);
1265 * If ILE (interrupt little-endian) has changed, update the
1266 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1268 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1269 struct kvm_vcpu *vcpu;
1272 kvm_for_each_vcpu(i, vcpu, kvm) {
1273 if (vcpu->arch.vcore != vc)
1275 if (new_lpcr & LPCR_ILE)
1276 vcpu->arch.intr_msr |= MSR_LE;
1278 vcpu->arch.intr_msr &= ~MSR_LE;
1283 * Userspace can only modify DPFD (default prefetch depth),
1284 * ILE (interrupt little-endian) and TC (translation control).
1285 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1287 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1288 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1291 * On POWER9, allow userspace to enable large decrementer for the
1292 * guest, whether or not the host has it enabled.
1294 if (cpu_has_feature(CPU_FTR_ARCH_300))
1297 /* Broken 32-bit version of LPCR must not clear top bits */
1300 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1301 spin_unlock(&vc->lock);
1302 mutex_unlock(&kvm->lock);
1305 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1306 union kvmppc_one_reg *val)
1312 case KVM_REG_PPC_DEBUG_INST:
1313 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1315 case KVM_REG_PPC_HIOR:
1316 *val = get_reg_val(id, 0);
1318 case KVM_REG_PPC_DABR:
1319 *val = get_reg_val(id, vcpu->arch.dabr);
1321 case KVM_REG_PPC_DABRX:
1322 *val = get_reg_val(id, vcpu->arch.dabrx);
1324 case KVM_REG_PPC_DSCR:
1325 *val = get_reg_val(id, vcpu->arch.dscr);
1327 case KVM_REG_PPC_PURR:
1328 *val = get_reg_val(id, vcpu->arch.purr);
1330 case KVM_REG_PPC_SPURR:
1331 *val = get_reg_val(id, vcpu->arch.spurr);
1333 case KVM_REG_PPC_AMR:
1334 *val = get_reg_val(id, vcpu->arch.amr);
1336 case KVM_REG_PPC_UAMOR:
1337 *val = get_reg_val(id, vcpu->arch.uamor);
1339 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1340 i = id - KVM_REG_PPC_MMCR0;
1341 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1343 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1344 i = id - KVM_REG_PPC_PMC1;
1345 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1347 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1348 i = id - KVM_REG_PPC_SPMC1;
1349 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1351 case KVM_REG_PPC_SIAR:
1352 *val = get_reg_val(id, vcpu->arch.siar);
1354 case KVM_REG_PPC_SDAR:
1355 *val = get_reg_val(id, vcpu->arch.sdar);
1357 case KVM_REG_PPC_SIER:
1358 *val = get_reg_val(id, vcpu->arch.sier);
1360 case KVM_REG_PPC_IAMR:
1361 *val = get_reg_val(id, vcpu->arch.iamr);
1363 case KVM_REG_PPC_PSPB:
1364 *val = get_reg_val(id, vcpu->arch.pspb);
1366 case KVM_REG_PPC_DPDES:
1367 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1369 case KVM_REG_PPC_VTB:
1370 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1372 case KVM_REG_PPC_DAWR:
1373 *val = get_reg_val(id, vcpu->arch.dawr);
1375 case KVM_REG_PPC_DAWRX:
1376 *val = get_reg_val(id, vcpu->arch.dawrx);
1378 case KVM_REG_PPC_CIABR:
1379 *val = get_reg_val(id, vcpu->arch.ciabr);
1381 case KVM_REG_PPC_CSIGR:
1382 *val = get_reg_val(id, vcpu->arch.csigr);
1384 case KVM_REG_PPC_TACR:
1385 *val = get_reg_val(id, vcpu->arch.tacr);
1387 case KVM_REG_PPC_TCSCR:
1388 *val = get_reg_val(id, vcpu->arch.tcscr);
1390 case KVM_REG_PPC_PID:
1391 *val = get_reg_val(id, vcpu->arch.pid);
1393 case KVM_REG_PPC_ACOP:
1394 *val = get_reg_val(id, vcpu->arch.acop);
1396 case KVM_REG_PPC_WORT:
1397 *val = get_reg_val(id, vcpu->arch.wort);
1399 case KVM_REG_PPC_TIDR:
1400 *val = get_reg_val(id, vcpu->arch.tid);
1402 case KVM_REG_PPC_PSSCR:
1403 *val = get_reg_val(id, vcpu->arch.psscr);
1405 case KVM_REG_PPC_VPA_ADDR:
1406 spin_lock(&vcpu->arch.vpa_update_lock);
1407 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1408 spin_unlock(&vcpu->arch.vpa_update_lock);
1410 case KVM_REG_PPC_VPA_SLB:
1411 spin_lock(&vcpu->arch.vpa_update_lock);
1412 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1413 val->vpaval.length = vcpu->arch.slb_shadow.len;
1414 spin_unlock(&vcpu->arch.vpa_update_lock);
1416 case KVM_REG_PPC_VPA_DTL:
1417 spin_lock(&vcpu->arch.vpa_update_lock);
1418 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1419 val->vpaval.length = vcpu->arch.dtl.len;
1420 spin_unlock(&vcpu->arch.vpa_update_lock);
1422 case KVM_REG_PPC_TB_OFFSET:
1423 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1425 case KVM_REG_PPC_LPCR:
1426 case KVM_REG_PPC_LPCR_64:
1427 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1429 case KVM_REG_PPC_PPR:
1430 *val = get_reg_val(id, vcpu->arch.ppr);
1432 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1433 case KVM_REG_PPC_TFHAR:
1434 *val = get_reg_val(id, vcpu->arch.tfhar);
1436 case KVM_REG_PPC_TFIAR:
1437 *val = get_reg_val(id, vcpu->arch.tfiar);
1439 case KVM_REG_PPC_TEXASR:
1440 *val = get_reg_val(id, vcpu->arch.texasr);
1442 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1443 i = id - KVM_REG_PPC_TM_GPR0;
1444 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1446 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1449 i = id - KVM_REG_PPC_TM_VSR0;
1451 for (j = 0; j < TS_FPRWIDTH; j++)
1452 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1454 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1455 val->vval = vcpu->arch.vr_tm.vr[i-32];
1461 case KVM_REG_PPC_TM_CR:
1462 *val = get_reg_val(id, vcpu->arch.cr_tm);
1464 case KVM_REG_PPC_TM_XER:
1465 *val = get_reg_val(id, vcpu->arch.xer_tm);
1467 case KVM_REG_PPC_TM_LR:
1468 *val = get_reg_val(id, vcpu->arch.lr_tm);
1470 case KVM_REG_PPC_TM_CTR:
1471 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1473 case KVM_REG_PPC_TM_FPSCR:
1474 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1476 case KVM_REG_PPC_TM_AMR:
1477 *val = get_reg_val(id, vcpu->arch.amr_tm);
1479 case KVM_REG_PPC_TM_PPR:
1480 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1482 case KVM_REG_PPC_TM_VRSAVE:
1483 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1485 case KVM_REG_PPC_TM_VSCR:
1486 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1487 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1491 case KVM_REG_PPC_TM_DSCR:
1492 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1494 case KVM_REG_PPC_TM_TAR:
1495 *val = get_reg_val(id, vcpu->arch.tar_tm);
1498 case KVM_REG_PPC_ARCH_COMPAT:
1499 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1509 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1510 union kvmppc_one_reg *val)
1514 unsigned long addr, len;
1517 case KVM_REG_PPC_HIOR:
1518 /* Only allow this to be set to zero */
1519 if (set_reg_val(id, *val))
1522 case KVM_REG_PPC_DABR:
1523 vcpu->arch.dabr = set_reg_val(id, *val);
1525 case KVM_REG_PPC_DABRX:
1526 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1528 case KVM_REG_PPC_DSCR:
1529 vcpu->arch.dscr = set_reg_val(id, *val);
1531 case KVM_REG_PPC_PURR:
1532 vcpu->arch.purr = set_reg_val(id, *val);
1534 case KVM_REG_PPC_SPURR:
1535 vcpu->arch.spurr = set_reg_val(id, *val);
1537 case KVM_REG_PPC_AMR:
1538 vcpu->arch.amr = set_reg_val(id, *val);
1540 case KVM_REG_PPC_UAMOR:
1541 vcpu->arch.uamor = set_reg_val(id, *val);
1543 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1544 i = id - KVM_REG_PPC_MMCR0;
1545 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1547 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1548 i = id - KVM_REG_PPC_PMC1;
1549 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1551 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1552 i = id - KVM_REG_PPC_SPMC1;
1553 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1555 case KVM_REG_PPC_SIAR:
1556 vcpu->arch.siar = set_reg_val(id, *val);
1558 case KVM_REG_PPC_SDAR:
1559 vcpu->arch.sdar = set_reg_val(id, *val);
1561 case KVM_REG_PPC_SIER:
1562 vcpu->arch.sier = set_reg_val(id, *val);
1564 case KVM_REG_PPC_IAMR:
1565 vcpu->arch.iamr = set_reg_val(id, *val);
1567 case KVM_REG_PPC_PSPB:
1568 vcpu->arch.pspb = set_reg_val(id, *val);
1570 case KVM_REG_PPC_DPDES:
1571 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1573 case KVM_REG_PPC_VTB:
1574 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1576 case KVM_REG_PPC_DAWR:
1577 vcpu->arch.dawr = set_reg_val(id, *val);
1579 case KVM_REG_PPC_DAWRX:
1580 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1582 case KVM_REG_PPC_CIABR:
1583 vcpu->arch.ciabr = set_reg_val(id, *val);
1584 /* Don't allow setting breakpoints in hypervisor code */
1585 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1586 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1588 case KVM_REG_PPC_CSIGR:
1589 vcpu->arch.csigr = set_reg_val(id, *val);
1591 case KVM_REG_PPC_TACR:
1592 vcpu->arch.tacr = set_reg_val(id, *val);
1594 case KVM_REG_PPC_TCSCR:
1595 vcpu->arch.tcscr = set_reg_val(id, *val);
1597 case KVM_REG_PPC_PID:
1598 vcpu->arch.pid = set_reg_val(id, *val);
1600 case KVM_REG_PPC_ACOP:
1601 vcpu->arch.acop = set_reg_val(id, *val);
1603 case KVM_REG_PPC_WORT:
1604 vcpu->arch.wort = set_reg_val(id, *val);
1606 case KVM_REG_PPC_TIDR:
1607 vcpu->arch.tid = set_reg_val(id, *val);
1609 case KVM_REG_PPC_PSSCR:
1610 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1612 case KVM_REG_PPC_VPA_ADDR:
1613 addr = set_reg_val(id, *val);
1615 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1616 vcpu->arch.dtl.next_gpa))
1618 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1620 case KVM_REG_PPC_VPA_SLB:
1621 addr = val->vpaval.addr;
1622 len = val->vpaval.length;
1624 if (addr && !vcpu->arch.vpa.next_gpa)
1626 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1628 case KVM_REG_PPC_VPA_DTL:
1629 addr = val->vpaval.addr;
1630 len = val->vpaval.length;
1632 if (addr && (len < sizeof(struct dtl_entry) ||
1633 !vcpu->arch.vpa.next_gpa))
1635 len -= len % sizeof(struct dtl_entry);
1636 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1638 case KVM_REG_PPC_TB_OFFSET:
1640 * POWER9 DD1 has an erratum where writing TBU40 causes
1641 * the timebase to lose ticks. So we don't let the
1642 * timebase offset be changed on P9 DD1. (It is
1643 * initialized to zero.)
1645 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1647 /* round up to multiple of 2^24 */
1648 vcpu->arch.vcore->tb_offset =
1649 ALIGN(set_reg_val(id, *val), 1UL << 24);
1651 case KVM_REG_PPC_LPCR:
1652 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1654 case KVM_REG_PPC_LPCR_64:
1655 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1657 case KVM_REG_PPC_PPR:
1658 vcpu->arch.ppr = set_reg_val(id, *val);
1660 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1661 case KVM_REG_PPC_TFHAR:
1662 vcpu->arch.tfhar = set_reg_val(id, *val);
1664 case KVM_REG_PPC_TFIAR:
1665 vcpu->arch.tfiar = set_reg_val(id, *val);
1667 case KVM_REG_PPC_TEXASR:
1668 vcpu->arch.texasr = set_reg_val(id, *val);
1670 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1671 i = id - KVM_REG_PPC_TM_GPR0;
1672 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1674 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1677 i = id - KVM_REG_PPC_TM_VSR0;
1679 for (j = 0; j < TS_FPRWIDTH; j++)
1680 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1682 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1683 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1688 case KVM_REG_PPC_TM_CR:
1689 vcpu->arch.cr_tm = set_reg_val(id, *val);
1691 case KVM_REG_PPC_TM_XER:
1692 vcpu->arch.xer_tm = set_reg_val(id, *val);
1694 case KVM_REG_PPC_TM_LR:
1695 vcpu->arch.lr_tm = set_reg_val(id, *val);
1697 case KVM_REG_PPC_TM_CTR:
1698 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1700 case KVM_REG_PPC_TM_FPSCR:
1701 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1703 case KVM_REG_PPC_TM_AMR:
1704 vcpu->arch.amr_tm = set_reg_val(id, *val);
1706 case KVM_REG_PPC_TM_PPR:
1707 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1709 case KVM_REG_PPC_TM_VRSAVE:
1710 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1712 case KVM_REG_PPC_TM_VSCR:
1713 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1714 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1718 case KVM_REG_PPC_TM_DSCR:
1719 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1721 case KVM_REG_PPC_TM_TAR:
1722 vcpu->arch.tar_tm = set_reg_val(id, *val);
1725 case KVM_REG_PPC_ARCH_COMPAT:
1726 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1737 * On POWER9, threads are independent and can be in different partitions.
1738 * Therefore we consider each thread to be a subcore.
1739 * There is a restriction that all threads have to be in the same
1740 * MMU mode (radix or HPT), unfortunately, but since we only support
1741 * HPT guests on a HPT host so far, that isn't an impediment yet.
1743 static int threads_per_vcore(struct kvm *kvm)
1745 if (kvm->arch.threads_indep)
1747 return threads_per_subcore;
1750 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1752 struct kvmppc_vcore *vcore;
1754 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1759 spin_lock_init(&vcore->lock);
1760 spin_lock_init(&vcore->stoltb_lock);
1761 init_swait_queue_head(&vcore->wq);
1762 vcore->preempt_tb = TB_NIL;
1763 vcore->lpcr = kvm->arch.lpcr;
1764 vcore->first_vcpuid = core * kvm->arch.smt_mode;
1766 INIT_LIST_HEAD(&vcore->preempt_list);
1771 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1772 static struct debugfs_timings_element {
1776 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1777 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1778 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1779 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1780 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1783 #define N_TIMINGS (ARRAY_SIZE(timings))
1785 struct debugfs_timings_state {
1786 struct kvm_vcpu *vcpu;
1787 unsigned int buflen;
1788 char buf[N_TIMINGS * 100];
1791 static int debugfs_timings_open(struct inode *inode, struct file *file)
1793 struct kvm_vcpu *vcpu = inode->i_private;
1794 struct debugfs_timings_state *p;
1796 p = kzalloc(sizeof(*p), GFP_KERNEL);
1800 kvm_get_kvm(vcpu->kvm);
1802 file->private_data = p;
1804 return nonseekable_open(inode, file);
1807 static int debugfs_timings_release(struct inode *inode, struct file *file)
1809 struct debugfs_timings_state *p = file->private_data;
1811 kvm_put_kvm(p->vcpu->kvm);
1816 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1817 size_t len, loff_t *ppos)
1819 struct debugfs_timings_state *p = file->private_data;
1820 struct kvm_vcpu *vcpu = p->vcpu;
1822 struct kvmhv_tb_accumulator tb;
1831 buf_end = s + sizeof(p->buf);
1832 for (i = 0; i < N_TIMINGS; ++i) {
1833 struct kvmhv_tb_accumulator *acc;
1835 acc = (struct kvmhv_tb_accumulator *)
1836 ((unsigned long)vcpu + timings[i].offset);
1838 for (loops = 0; loops < 1000; ++loops) {
1839 count = acc->seqcount;
1844 if (count == acc->seqcount) {
1852 snprintf(s, buf_end - s, "%s: stuck\n",
1855 snprintf(s, buf_end - s,
1856 "%s: %llu %llu %llu %llu\n",
1857 timings[i].name, count / 2,
1858 tb_to_ns(tb.tb_total),
1859 tb_to_ns(tb.tb_min),
1860 tb_to_ns(tb.tb_max));
1863 p->buflen = s - p->buf;
1867 if (pos >= p->buflen)
1869 if (len > p->buflen - pos)
1870 len = p->buflen - pos;
1871 n = copy_to_user(buf, p->buf + pos, len);
1881 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1882 size_t len, loff_t *ppos)
1887 static const struct file_operations debugfs_timings_ops = {
1888 .owner = THIS_MODULE,
1889 .open = debugfs_timings_open,
1890 .release = debugfs_timings_release,
1891 .read = debugfs_timings_read,
1892 .write = debugfs_timings_write,
1893 .llseek = generic_file_llseek,
1896 /* Create a debugfs directory for the vcpu */
1897 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1900 struct kvm *kvm = vcpu->kvm;
1902 snprintf(buf, sizeof(buf), "vcpu%u", id);
1903 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1905 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1906 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1908 vcpu->arch.debugfs_timings =
1909 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1910 vcpu, &debugfs_timings_ops);
1913 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1914 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1917 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1919 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1922 struct kvm_vcpu *vcpu;
1925 struct kvmppc_vcore *vcore;
1928 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1932 err = kvm_vcpu_init(vcpu, kvm, id);
1936 vcpu->arch.shared = &vcpu->arch.shregs;
1937 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1939 * The shared struct is never shared on HV,
1940 * so we can always use host endianness
1942 #ifdef __BIG_ENDIAN__
1943 vcpu->arch.shared_big_endian = true;
1945 vcpu->arch.shared_big_endian = false;
1948 vcpu->arch.mmcr[0] = MMCR0_FC;
1949 vcpu->arch.ctrl = CTRL_RUNLATCH;
1950 /* default to host PVR, since we can't spoof it */
1951 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1952 spin_lock_init(&vcpu->arch.vpa_update_lock);
1953 spin_lock_init(&vcpu->arch.tbacct_lock);
1954 vcpu->arch.busy_preempt = TB_NIL;
1955 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1958 * Set the default HFSCR for the guest from the host value.
1959 * This value is only used on POWER9.
1960 * On POWER9 DD1, TM doesn't work, so we make sure to
1961 * prevent the guest from using it.
1962 * On POWER9, we want to virtualize the doorbell facility, so we
1963 * turn off the HFSCR bit, which causes those instructions to trap.
1965 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1966 if (!cpu_has_feature(CPU_FTR_TM))
1967 vcpu->arch.hfscr &= ~HFSCR_TM;
1968 if (cpu_has_feature(CPU_FTR_ARCH_300))
1969 vcpu->arch.hfscr &= ~HFSCR_MSGP;
1971 kvmppc_mmu_book3s_hv_init(vcpu);
1973 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1975 init_waitqueue_head(&vcpu->arch.cpu_run);
1977 mutex_lock(&kvm->lock);
1980 core = id / kvm->arch.smt_mode;
1981 if (core < KVM_MAX_VCORES) {
1982 vcore = kvm->arch.vcores[core];
1985 vcore = kvmppc_vcore_create(kvm, core);
1986 kvm->arch.vcores[core] = vcore;
1987 kvm->arch.online_vcores++;
1990 mutex_unlock(&kvm->lock);
1995 spin_lock(&vcore->lock);
1996 ++vcore->num_threads;
1997 spin_unlock(&vcore->lock);
1998 vcpu->arch.vcore = vcore;
1999 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2000 vcpu->arch.thread_cpu = -1;
2001 vcpu->arch.prev_cpu = -1;
2003 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2004 kvmppc_sanity_check(vcpu);
2006 debugfs_vcpu_init(vcpu, id);
2011 kmem_cache_free(kvm_vcpu_cache, vcpu);
2013 return ERR_PTR(err);
2016 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2017 unsigned long flags)
2024 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2026 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2028 * On POWER8 (or POWER7), the threading mode is "strict",
2029 * so we pack smt_mode vcpus per vcore.
2031 if (smt_mode > threads_per_subcore)
2035 * On POWER9, the threading mode is "loose",
2036 * so each vcpu gets its own vcore.
2041 mutex_lock(&kvm->lock);
2043 if (!kvm->arch.online_vcores) {
2044 kvm->arch.smt_mode = smt_mode;
2045 kvm->arch.emul_smt_mode = esmt;
2048 mutex_unlock(&kvm->lock);
2053 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2055 if (vpa->pinned_addr)
2056 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2060 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2062 spin_lock(&vcpu->arch.vpa_update_lock);
2063 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2064 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2065 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2066 spin_unlock(&vcpu->arch.vpa_update_lock);
2067 kvm_vcpu_uninit(vcpu);
2068 kmem_cache_free(kvm_vcpu_cache, vcpu);
2071 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2073 /* Indicate we want to get back into the guest */
2077 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2079 unsigned long dec_nsec, now;
2082 if (now > vcpu->arch.dec_expires) {
2083 /* decrementer has already gone negative */
2084 kvmppc_core_queue_dec(vcpu);
2085 kvmppc_core_prepare_to_enter(vcpu);
2088 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2090 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2091 vcpu->arch.timer_running = 1;
2094 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2096 vcpu->arch.ceded = 0;
2097 if (vcpu->arch.timer_running) {
2098 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2099 vcpu->arch.timer_running = 0;
2103 extern int __kvmppc_vcore_entry(void);
2105 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2106 struct kvm_vcpu *vcpu)
2110 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2112 spin_lock_irq(&vcpu->arch.tbacct_lock);
2114 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2115 vcpu->arch.stolen_logged;
2116 vcpu->arch.busy_preempt = now;
2117 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2118 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2120 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2123 static int kvmppc_grab_hwthread(int cpu)
2125 struct paca_struct *tpaca;
2126 long timeout = 10000;
2130 /* Ensure the thread won't go into the kernel if it wakes */
2131 tpaca->kvm_hstate.kvm_vcpu = NULL;
2132 tpaca->kvm_hstate.kvm_vcore = NULL;
2133 tpaca->kvm_hstate.napping = 0;
2135 tpaca->kvm_hstate.hwthread_req = 1;
2138 * If the thread is already executing in the kernel (e.g. handling
2139 * a stray interrupt), wait for it to get back to nap mode.
2140 * The smp_mb() is to ensure that our setting of hwthread_req
2141 * is visible before we look at hwthread_state, so if this
2142 * races with the code at system_reset_pSeries and the thread
2143 * misses our setting of hwthread_req, we are sure to see its
2144 * setting of hwthread_state, and vice versa.
2147 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2148 if (--timeout <= 0) {
2149 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2157 static void kvmppc_release_hwthread(int cpu)
2159 struct paca_struct *tpaca;
2162 tpaca->kvm_hstate.hwthread_req = 0;
2163 tpaca->kvm_hstate.kvm_vcpu = NULL;
2164 tpaca->kvm_hstate.kvm_vcore = NULL;
2165 tpaca->kvm_hstate.kvm_split_mode = NULL;
2168 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2172 cpu = cpu_first_thread_sibling(cpu);
2173 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2175 * Make sure setting of bit in need_tlb_flush precedes
2176 * testing of cpu_in_guest bits. The matching barrier on
2177 * the other side is the first smp_mb() in kvmppc_run_core().
2180 for (i = 0; i < threads_per_core; ++i)
2181 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2182 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2185 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2187 struct kvm *kvm = vcpu->kvm;
2190 * With radix, the guest can do TLB invalidations itself,
2191 * and it could choose to use the local form (tlbiel) if
2192 * it is invalidating a translation that has only ever been
2193 * used on one vcpu. However, that doesn't mean it has
2194 * only ever been used on one physical cpu, since vcpus
2195 * can move around between pcpus. To cope with this, when
2196 * a vcpu moves from one pcpu to another, we need to tell
2197 * any vcpus running on the same core as this vcpu previously
2198 * ran to flush the TLB. The TLB is shared between threads,
2199 * so we use a single bit in .need_tlb_flush for all 4 threads.
2201 if (vcpu->arch.prev_cpu != pcpu) {
2202 if (vcpu->arch.prev_cpu >= 0 &&
2203 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2204 cpu_first_thread_sibling(pcpu))
2205 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2206 vcpu->arch.prev_cpu = pcpu;
2210 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2213 struct paca_struct *tpaca;
2214 struct kvm *kvm = vc->kvm;
2218 if (vcpu->arch.timer_running) {
2219 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2220 vcpu->arch.timer_running = 0;
2222 cpu += vcpu->arch.ptid;
2223 vcpu->cpu = vc->pcpu;
2224 vcpu->arch.thread_cpu = cpu;
2225 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2228 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2229 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2230 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2232 tpaca->kvm_hstate.kvm_vcore = vc;
2233 if (cpu != smp_processor_id())
2234 kvmppc_ipi_thread(cpu);
2237 static void kvmppc_wait_for_nap(int n_threads)
2239 int cpu = smp_processor_id();
2244 for (loops = 0; loops < 1000000; ++loops) {
2246 * Check if all threads are finished.
2247 * We set the vcore pointer when starting a thread
2248 * and the thread clears it when finished, so we look
2249 * for any threads that still have a non-NULL vcore ptr.
2251 for (i = 1; i < n_threads; ++i)
2252 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2254 if (i == n_threads) {
2261 for (i = 1; i < n_threads; ++i)
2262 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2263 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2267 * Check that we are on thread 0 and that any other threads in
2268 * this core are off-line. Then grab the threads so they can't
2271 static int on_primary_thread(void)
2273 int cpu = smp_processor_id();
2276 /* Are we on a primary subcore? */
2277 if (cpu_thread_in_subcore(cpu))
2281 while (++thr < threads_per_subcore)
2282 if (cpu_online(cpu + thr))
2285 /* Grab all hw threads so they can't go into the kernel */
2286 for (thr = 1; thr < threads_per_subcore; ++thr) {
2287 if (kvmppc_grab_hwthread(cpu + thr)) {
2288 /* Couldn't grab one; let the others go */
2290 kvmppc_release_hwthread(cpu + thr);
2291 } while (--thr > 0);
2299 * A list of virtual cores for each physical CPU.
2300 * These are vcores that could run but their runner VCPU tasks are
2301 * (or may be) preempted.
2303 struct preempted_vcore_list {
2304 struct list_head list;
2308 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2310 static void init_vcore_lists(void)
2314 for_each_possible_cpu(cpu) {
2315 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2316 spin_lock_init(&lp->lock);
2317 INIT_LIST_HEAD(&lp->list);
2321 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2323 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2325 vc->vcore_state = VCORE_PREEMPT;
2326 vc->pcpu = smp_processor_id();
2327 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2328 spin_lock(&lp->lock);
2329 list_add_tail(&vc->preempt_list, &lp->list);
2330 spin_unlock(&lp->lock);
2333 /* Start accumulating stolen time */
2334 kvmppc_core_start_stolen(vc);
2337 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2339 struct preempted_vcore_list *lp;
2341 kvmppc_core_end_stolen(vc);
2342 if (!list_empty(&vc->preempt_list)) {
2343 lp = &per_cpu(preempted_vcores, vc->pcpu);
2344 spin_lock(&lp->lock);
2345 list_del_init(&vc->preempt_list);
2346 spin_unlock(&lp->lock);
2348 vc->vcore_state = VCORE_INACTIVE;
2352 * This stores information about the virtual cores currently
2353 * assigned to a physical core.
2357 int max_subcore_threads;
2359 int subcore_threads[MAX_SUBCORES];
2360 struct kvmppc_vcore *vc[MAX_SUBCORES];
2364 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2365 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2367 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2369 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2371 memset(cip, 0, sizeof(*cip));
2372 cip->n_subcores = 1;
2373 cip->max_subcore_threads = vc->num_threads;
2374 cip->total_threads = vc->num_threads;
2375 cip->subcore_threads[0] = vc->num_threads;
2379 static bool subcore_config_ok(int n_subcores, int n_threads)
2382 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way split-core
2383 * mode, with one thread per subcore.
2385 if (cpu_has_feature(CPU_FTR_ARCH_300))
2386 return n_subcores <= 4 && n_threads == 1;
2388 /* On POWER8, can only dynamically split if unsplit to begin with */
2389 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2391 if (n_subcores > MAX_SUBCORES)
2393 if (n_subcores > 1) {
2394 if (!(dynamic_mt_modes & 2))
2396 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2400 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2403 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2405 vc->entry_exit_map = 0;
2407 vc->napping_threads = 0;
2408 vc->conferring_threads = 0;
2411 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2413 int n_threads = vc->num_threads;
2416 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2419 /* POWER9 currently requires all threads to be in the same MMU mode */
2420 if (cpu_has_feature(CPU_FTR_ARCH_300) &&
2421 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2424 if (n_threads < cip->max_subcore_threads)
2425 n_threads = cip->max_subcore_threads;
2426 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2428 cip->max_subcore_threads = n_threads;
2430 sub = cip->n_subcores;
2432 cip->total_threads += vc->num_threads;
2433 cip->subcore_threads[sub] = vc->num_threads;
2435 init_vcore_to_run(vc);
2436 list_del_init(&vc->preempt_list);
2442 * Work out whether it is possible to piggyback the execution of
2443 * vcore *pvc onto the execution of the other vcores described in *cip.
2445 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2448 if (cip->total_threads + pvc->num_threads > target_threads)
2451 return can_dynamic_split(pvc, cip);
2454 static void prepare_threads(struct kvmppc_vcore *vc)
2457 struct kvm_vcpu *vcpu;
2459 for_each_runnable_thread(i, vcpu, vc) {
2460 if (signal_pending(vcpu->arch.run_task))
2461 vcpu->arch.ret = -EINTR;
2462 else if (vcpu->arch.vpa.update_pending ||
2463 vcpu->arch.slb_shadow.update_pending ||
2464 vcpu->arch.dtl.update_pending)
2465 vcpu->arch.ret = RESUME_GUEST;
2468 kvmppc_remove_runnable(vc, vcpu);
2469 wake_up(&vcpu->arch.cpu_run);
2473 static void collect_piggybacks(struct core_info *cip, int target_threads)
2475 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2476 struct kvmppc_vcore *pvc, *vcnext;
2478 spin_lock(&lp->lock);
2479 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2480 if (!spin_trylock(&pvc->lock))
2482 prepare_threads(pvc);
2483 if (!pvc->n_runnable) {
2484 list_del_init(&pvc->preempt_list);
2485 if (pvc->runner == NULL) {
2486 pvc->vcore_state = VCORE_INACTIVE;
2487 kvmppc_core_end_stolen(pvc);
2489 spin_unlock(&pvc->lock);
2492 if (!can_piggyback(pvc, cip, target_threads)) {
2493 spin_unlock(&pvc->lock);
2496 kvmppc_core_end_stolen(pvc);
2497 pvc->vcore_state = VCORE_PIGGYBACK;
2498 if (cip->total_threads >= target_threads)
2501 spin_unlock(&lp->lock);
2504 static bool recheck_signals(struct core_info *cip)
2507 struct kvm_vcpu *vcpu;
2509 for (sub = 0; sub < cip->n_subcores; ++sub)
2510 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2511 if (signal_pending(vcpu->arch.run_task))
2516 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2518 int still_running = 0, i;
2521 struct kvm_vcpu *vcpu;
2523 spin_lock(&vc->lock);
2525 for_each_runnable_thread(i, vcpu, vc) {
2526 /* cancel pending dec exception if dec is positive */
2527 if (now < vcpu->arch.dec_expires &&
2528 kvmppc_core_pending_dec(vcpu))
2529 kvmppc_core_dequeue_dec(vcpu);
2531 trace_kvm_guest_exit(vcpu);
2534 if (vcpu->arch.trap)
2535 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2536 vcpu->arch.run_task);
2538 vcpu->arch.ret = ret;
2539 vcpu->arch.trap = 0;
2541 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2542 if (vcpu->arch.pending_exceptions)
2543 kvmppc_core_prepare_to_enter(vcpu);
2544 if (vcpu->arch.ceded)
2545 kvmppc_set_timer(vcpu);
2549 kvmppc_remove_runnable(vc, vcpu);
2550 wake_up(&vcpu->arch.cpu_run);
2554 if (still_running > 0) {
2555 kvmppc_vcore_preempt(vc);
2556 } else if (vc->runner) {
2557 vc->vcore_state = VCORE_PREEMPT;
2558 kvmppc_core_start_stolen(vc);
2560 vc->vcore_state = VCORE_INACTIVE;
2562 if (vc->n_runnable > 0 && vc->runner == NULL) {
2563 /* make sure there's a candidate runner awake */
2565 vcpu = next_runnable_thread(vc, &i);
2566 wake_up(&vcpu->arch.cpu_run);
2569 spin_unlock(&vc->lock);
2573 * Clear core from the list of active host cores as we are about to
2574 * enter the guest. Only do this if it is the primary thread of the
2575 * core (not if a subcore) that is entering the guest.
2577 static inline int kvmppc_clear_host_core(unsigned int cpu)
2581 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2584 * Memory barrier can be omitted here as we will do a smp_wmb()
2585 * later in kvmppc_start_thread and we need ensure that state is
2586 * visible to other CPUs only after we enter guest.
2588 core = cpu >> threads_shift;
2589 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2594 * Advertise this core as an active host core since we exited the guest
2595 * Only need to do this if it is the primary thread of the core that is
2598 static inline int kvmppc_set_host_core(unsigned int cpu)
2602 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2606 * Memory barrier can be omitted here because we do a spin_unlock
2607 * immediately after this which provides the memory barrier.
2609 core = cpu >> threads_shift;
2610 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2614 static void set_irq_happened(int trap)
2617 case BOOK3S_INTERRUPT_EXTERNAL:
2618 local_paca->irq_happened |= PACA_IRQ_EE;
2620 case BOOK3S_INTERRUPT_H_DOORBELL:
2621 local_paca->irq_happened |= PACA_IRQ_DBELL;
2623 case BOOK3S_INTERRUPT_HMI:
2624 local_paca->irq_happened |= PACA_IRQ_HMI;
2626 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2627 replay_system_reset();
2633 * Run a set of guest threads on a physical core.
2634 * Called with vc->lock held.
2636 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2638 struct kvm_vcpu *vcpu;
2641 struct core_info core_info;
2642 struct kvmppc_vcore *pvc;
2643 struct kvm_split_mode split_info, *sip;
2644 int split, subcore_size, active;
2647 unsigned long cmd_bit, stat_bit;
2650 int controlled_threads;
2656 * Remove from the list any threads that have a signal pending
2657 * or need a VPA update done
2659 prepare_threads(vc);
2661 /* if the runner is no longer runnable, let the caller pick a new one */
2662 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2668 init_vcore_to_run(vc);
2669 vc->preempt_tb = TB_NIL;
2672 * Number of threads that we will be controlling: the same as
2673 * the number of threads per subcore, except on POWER9,
2674 * where it's 1 because the threads are (mostly) independent.
2676 controlled_threads = threads_per_vcore(vc->kvm);
2679 * Make sure we are running on primary threads, and that secondary
2680 * threads are offline. Also check if the number of threads in this
2681 * guest are greater than the current system threads per guest.
2682 * On POWER9, we need to be not in independent-threads mode if
2683 * this is a HPT guest on a radix host.
2685 hpt_on_radix = radix_enabled() && !kvm_is_radix(vc->kvm);
2686 if (((controlled_threads > 1) &&
2687 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2688 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2689 for_each_runnable_thread(i, vcpu, vc) {
2690 vcpu->arch.ret = -EBUSY;
2691 kvmppc_remove_runnable(vc, vcpu);
2692 wake_up(&vcpu->arch.cpu_run);
2698 * See if we could run any other vcores on the physical core
2699 * along with this one.
2701 init_core_info(&core_info, vc);
2702 pcpu = smp_processor_id();
2703 target_threads = controlled_threads;
2704 if (target_smt_mode && target_smt_mode < target_threads)
2705 target_threads = target_smt_mode;
2706 if (vc->num_threads < target_threads)
2707 collect_piggybacks(&core_info, target_threads);
2710 * On radix, arrange for TLB flushing if necessary.
2711 * This has to be done before disabling interrupts since
2712 * it uses smp_call_function().
2714 pcpu = smp_processor_id();
2715 if (kvm_is_radix(vc->kvm)) {
2716 for (sub = 0; sub < core_info.n_subcores; ++sub)
2717 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2718 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2722 * Hard-disable interrupts, and check resched flag and signals.
2723 * If we need to reschedule or deliver a signal, clean up
2724 * and return without going into the guest(s).
2725 * If the mmu_ready flag has been cleared, don't go into the
2726 * guest because that means a HPT resize operation is in progress.
2728 local_irq_disable();
2730 if (lazy_irq_pending() || need_resched() ||
2731 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
2733 vc->vcore_state = VCORE_INACTIVE;
2734 /* Unlock all except the primary vcore */
2735 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2736 pvc = core_info.vc[sub];
2737 /* Put back on to the preempted vcores list */
2738 kvmppc_vcore_preempt(pvc);
2739 spin_unlock(&pvc->lock);
2741 for (i = 0; i < controlled_threads; ++i)
2742 kvmppc_release_hwthread(pcpu + i);
2746 kvmppc_clear_host_core(pcpu);
2748 /* Decide on micro-threading (split-core) mode */
2749 subcore_size = threads_per_subcore;
2750 cmd_bit = stat_bit = 0;
2751 split = core_info.n_subcores;
2753 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2754 && !cpu_has_feature(CPU_FTR_ARCH_300);
2756 if (split > 1 || hpt_on_radix) {
2758 memset(&split_info, 0, sizeof(split_info));
2759 for (sub = 0; sub < core_info.n_subcores; ++sub)
2760 split_info.vc[sub] = core_info.vc[sub];
2763 if (split == 2 && (dynamic_mt_modes & 2)) {
2764 cmd_bit = HID0_POWER8_1TO2LPAR;
2765 stat_bit = HID0_POWER8_2LPARMODE;
2768 cmd_bit = HID0_POWER8_1TO4LPAR;
2769 stat_bit = HID0_POWER8_4LPARMODE;
2771 subcore_size = MAX_SMT_THREADS / split;
2772 split_info.rpr = mfspr(SPRN_RPR);
2773 split_info.pmmar = mfspr(SPRN_PMMAR);
2774 split_info.ldbar = mfspr(SPRN_LDBAR);
2775 split_info.subcore_size = subcore_size;
2777 split_info.subcore_size = 1;
2779 /* Use the split_info for LPCR/LPIDR changes */
2780 split_info.lpcr_req = vc->lpcr;
2781 split_info.lpidr_req = vc->kvm->arch.lpid;
2782 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2783 split_info.do_set = 1;
2787 /* order writes to split_info before kvm_split_mode pointer */
2791 for (thr = 0; thr < controlled_threads; ++thr) {
2792 paca[pcpu + thr].kvm_hstate.tid = thr;
2793 paca[pcpu + thr].kvm_hstate.napping = 0;
2794 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2797 /* Initiate micro-threading (split-core) on POWER8 if required */
2799 unsigned long hid0 = mfspr(SPRN_HID0);
2801 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2803 mtspr(SPRN_HID0, hid0);
2806 hid0 = mfspr(SPRN_HID0);
2807 if (hid0 & stat_bit)
2813 /* Start all the threads */
2815 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2816 thr = is_power8 ? subcore_thread_map[sub] : sub;
2819 pvc = core_info.vc[sub];
2820 pvc->pcpu = pcpu + thr;
2821 for_each_runnable_thread(i, vcpu, pvc) {
2822 kvmppc_start_thread(vcpu, pvc);
2823 kvmppc_create_dtl_entry(vcpu, pvc);
2824 trace_kvm_guest_enter(vcpu);
2825 if (!vcpu->arch.ptid)
2827 active |= 1 << (thr + vcpu->arch.ptid);
2830 * We need to start the first thread of each subcore
2831 * even if it doesn't have a vcpu.
2834 kvmppc_start_thread(NULL, pvc);
2835 thr += pvc->num_threads;
2839 * Ensure that split_info.do_nap is set after setting
2840 * the vcore pointer in the PACA of the secondaries.
2845 * When doing micro-threading, poke the inactive threads as well.
2846 * This gets them to the nap instruction after kvm_do_nap,
2847 * which reduces the time taken to unsplit later.
2848 * For POWER9 HPT guest on radix host, we need all the secondary
2849 * threads woken up so they can do the LPCR/LPIDR change.
2851 if (cmd_bit || hpt_on_radix) {
2852 split_info.do_nap = 1; /* ask secondaries to nap when done */
2853 for (thr = 1; thr < threads_per_subcore; ++thr)
2854 if (!(active & (1 << thr)))
2855 kvmppc_ipi_thread(pcpu + thr);
2858 vc->vcore_state = VCORE_RUNNING;
2861 trace_kvmppc_run_core(vc, 0);
2863 for (sub = 0; sub < core_info.n_subcores; ++sub)
2864 spin_unlock(&core_info.vc[sub]->lock);
2867 * Interrupts will be enabled once we get into the guest,
2868 * so tell lockdep that we're about to enable interrupts.
2870 trace_hardirqs_on();
2874 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2876 trap = __kvmppc_vcore_entry();
2878 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2882 trace_hardirqs_off();
2883 set_irq_happened(trap);
2885 spin_lock(&vc->lock);
2886 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2887 vc->vcore_state = VCORE_EXITING;
2889 /* wait for secondary threads to finish writing their state to memory */
2890 kvmppc_wait_for_nap(controlled_threads);
2892 /* Return to whole-core mode if we split the core earlier */
2894 unsigned long hid0 = mfspr(SPRN_HID0);
2895 unsigned long loops = 0;
2897 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2898 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2900 mtspr(SPRN_HID0, hid0);
2903 hid0 = mfspr(SPRN_HID0);
2904 if (!(hid0 & stat_bit))
2909 } else if (hpt_on_radix) {
2910 /* Wait for all threads to have seen final sync */
2911 for (thr = 1; thr < controlled_threads; ++thr) {
2912 while (paca[pcpu + thr].kvm_hstate.kvm_split_mode) {
2919 split_info.do_nap = 0;
2921 kvmppc_set_host_core(pcpu);
2925 /* Let secondaries go back to the offline loop */
2926 for (i = 0; i < controlled_threads; ++i) {
2927 kvmppc_release_hwthread(pcpu + i);
2928 if (sip && sip->napped[i])
2929 kvmppc_ipi_thread(pcpu + i);
2930 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2933 spin_unlock(&vc->lock);
2935 /* make sure updates to secondary vcpu structs are visible now */
2938 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2939 pvc = core_info.vc[sub];
2940 post_guest_process(pvc, pvc == vc);
2943 spin_lock(&vc->lock);
2947 vc->vcore_state = VCORE_INACTIVE;
2948 trace_kvmppc_run_core(vc, 1);
2952 * Wait for some other vcpu thread to execute us, and
2953 * wake us up when we need to handle something in the host.
2955 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2956 struct kvm_vcpu *vcpu, int wait_state)
2960 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2961 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2962 spin_unlock(&vc->lock);
2964 spin_lock(&vc->lock);
2966 finish_wait(&vcpu->arch.cpu_run, &wait);
2969 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2972 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2973 vc->halt_poll_ns = 10000;
2975 vc->halt_poll_ns *= halt_poll_ns_grow;
2978 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2980 if (halt_poll_ns_shrink == 0)
2981 vc->halt_poll_ns = 0;
2983 vc->halt_poll_ns /= halt_poll_ns_shrink;
2986 #ifdef CONFIG_KVM_XICS
2987 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2989 if (!xive_enabled())
2991 return vcpu->arch.xive_saved_state.pipr <
2992 vcpu->arch.xive_saved_state.cppr;
2995 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2999 #endif /* CONFIG_KVM_XICS */
3001 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3003 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3004 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3011 * Check to see if any of the runnable vcpus on the vcore have pending
3012 * exceptions or are no longer ceded
3014 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3016 struct kvm_vcpu *vcpu;
3019 for_each_runnable_thread(i, vcpu, vc) {
3020 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3028 * All the vcpus in this vcore are idle, so wait for a decrementer
3029 * or external interrupt to one of the vcpus. vc->lock is held.
3031 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3033 ktime_t cur, start_poll, start_wait;
3036 DECLARE_SWAITQUEUE(wait);
3038 /* Poll for pending exceptions and ceded state */
3039 cur = start_poll = ktime_get();
3040 if (vc->halt_poll_ns) {
3041 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3042 ++vc->runner->stat.halt_attempted_poll;
3044 vc->vcore_state = VCORE_POLLING;
3045 spin_unlock(&vc->lock);
3048 if (kvmppc_vcore_check_block(vc)) {
3053 } while (single_task_running() && ktime_before(cur, stop));
3055 spin_lock(&vc->lock);
3056 vc->vcore_state = VCORE_INACTIVE;
3059 ++vc->runner->stat.halt_successful_poll;
3064 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3066 if (kvmppc_vcore_check_block(vc)) {
3067 finish_swait(&vc->wq, &wait);
3069 /* If we polled, count this as a successful poll */
3070 if (vc->halt_poll_ns)
3071 ++vc->runner->stat.halt_successful_poll;
3075 start_wait = ktime_get();
3077 vc->vcore_state = VCORE_SLEEPING;
3078 trace_kvmppc_vcore_blocked(vc, 0);
3079 spin_unlock(&vc->lock);
3081 finish_swait(&vc->wq, &wait);
3082 spin_lock(&vc->lock);
3083 vc->vcore_state = VCORE_INACTIVE;
3084 trace_kvmppc_vcore_blocked(vc, 1);
3085 ++vc->runner->stat.halt_successful_wait;
3090 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3092 /* Attribute wait time */
3094 vc->runner->stat.halt_wait_ns +=
3095 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3096 /* Attribute failed poll time */
3097 if (vc->halt_poll_ns)
3098 vc->runner->stat.halt_poll_fail_ns +=
3099 ktime_to_ns(start_wait) -
3100 ktime_to_ns(start_poll);
3102 /* Attribute successful poll time */
3103 if (vc->halt_poll_ns)
3104 vc->runner->stat.halt_poll_success_ns +=
3106 ktime_to_ns(start_poll);
3109 /* Adjust poll time */
3111 if (block_ns <= vc->halt_poll_ns)
3113 /* We slept and blocked for longer than the max halt time */
3114 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3115 shrink_halt_poll_ns(vc);
3116 /* We slept and our poll time is too small */
3117 else if (vc->halt_poll_ns < halt_poll_ns &&
3118 block_ns < halt_poll_ns)
3119 grow_halt_poll_ns(vc);
3120 if (vc->halt_poll_ns > halt_poll_ns)
3121 vc->halt_poll_ns = halt_poll_ns;
3123 vc->halt_poll_ns = 0;
3125 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3128 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3131 struct kvm *kvm = vcpu->kvm;
3133 mutex_lock(&kvm->lock);
3134 if (!kvm->arch.mmu_ready) {
3135 if (!kvm_is_radix(kvm))
3136 r = kvmppc_hv_setup_htab_rma(vcpu);
3138 if (cpu_has_feature(CPU_FTR_ARCH_300))
3139 kvmppc_setup_partition_table(kvm);
3140 kvm->arch.mmu_ready = 1;
3143 mutex_unlock(&kvm->lock);
3147 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3150 struct kvmppc_vcore *vc;
3153 trace_kvmppc_run_vcpu_enter(vcpu);
3155 kvm_run->exit_reason = 0;
3156 vcpu->arch.ret = RESUME_GUEST;
3157 vcpu->arch.trap = 0;
3158 kvmppc_update_vpas(vcpu);
3161 * Synchronize with other threads in this virtual core
3163 vc = vcpu->arch.vcore;
3164 spin_lock(&vc->lock);
3165 vcpu->arch.ceded = 0;
3166 vcpu->arch.run_task = current;
3167 vcpu->arch.kvm_run = kvm_run;
3168 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3169 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3170 vcpu->arch.busy_preempt = TB_NIL;
3171 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3175 * This happens the first time this is called for a vcpu.
3176 * If the vcore is already running, we may be able to start
3177 * this thread straight away and have it join in.
3179 if (!signal_pending(current)) {
3180 if (vc->vcore_state == VCORE_PIGGYBACK) {
3181 if (spin_trylock(&vc->lock)) {
3182 if (vc->vcore_state == VCORE_RUNNING &&
3183 !VCORE_IS_EXITING(vc)) {
3184 kvmppc_create_dtl_entry(vcpu, vc);
3185 kvmppc_start_thread(vcpu, vc);
3186 trace_kvm_guest_enter(vcpu);
3188 spin_unlock(&vc->lock);
3190 } else if (vc->vcore_state == VCORE_RUNNING &&
3191 !VCORE_IS_EXITING(vc)) {
3192 kvmppc_create_dtl_entry(vcpu, vc);
3193 kvmppc_start_thread(vcpu, vc);
3194 trace_kvm_guest_enter(vcpu);
3195 } else if (vc->vcore_state == VCORE_SLEEPING) {
3201 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3202 !signal_pending(current)) {
3203 /* See if the MMU is ready to go */
3204 if (!vcpu->kvm->arch.mmu_ready) {
3205 spin_unlock(&vc->lock);
3206 r = kvmhv_setup_mmu(vcpu);
3207 spin_lock(&vc->lock);
3209 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3210 kvm_run->fail_entry.
3211 hardware_entry_failure_reason = 0;
3217 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3218 kvmppc_vcore_end_preempt(vc);
3220 if (vc->vcore_state != VCORE_INACTIVE) {
3221 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3224 for_each_runnable_thread(i, v, vc) {
3225 kvmppc_core_prepare_to_enter(v);
3226 if (signal_pending(v->arch.run_task)) {
3227 kvmppc_remove_runnable(vc, v);
3228 v->stat.signal_exits++;
3229 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3230 v->arch.ret = -EINTR;
3231 wake_up(&v->arch.cpu_run);
3234 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3237 for_each_runnable_thread(i, v, vc) {
3238 if (!kvmppc_vcpu_woken(v))
3239 n_ceded += v->arch.ceded;
3244 if (n_ceded == vc->n_runnable) {
3245 kvmppc_vcore_blocked(vc);
3246 } else if (need_resched()) {
3247 kvmppc_vcore_preempt(vc);
3248 /* Let something else run */
3249 cond_resched_lock(&vc->lock);
3250 if (vc->vcore_state == VCORE_PREEMPT)
3251 kvmppc_vcore_end_preempt(vc);
3253 kvmppc_run_core(vc);
3258 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3259 (vc->vcore_state == VCORE_RUNNING ||
3260 vc->vcore_state == VCORE_EXITING ||
3261 vc->vcore_state == VCORE_PIGGYBACK))
3262 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3264 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3265 kvmppc_vcore_end_preempt(vc);
3267 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3268 kvmppc_remove_runnable(vc, vcpu);
3269 vcpu->stat.signal_exits++;
3270 kvm_run->exit_reason = KVM_EXIT_INTR;
3271 vcpu->arch.ret = -EINTR;
3274 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3275 /* Wake up some vcpu to run the core */
3277 v = next_runnable_thread(vc, &i);
3278 wake_up(&v->arch.cpu_run);
3281 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3282 spin_unlock(&vc->lock);
3283 return vcpu->arch.ret;
3286 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3290 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3291 unsigned long user_tar = 0;
3292 unsigned int user_vrsave;
3295 if (!vcpu->arch.sane) {
3296 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3301 * Don't allow entry with a suspended transaction, because
3302 * the guest entry/exit code will lose it.
3303 * If the guest has TM enabled, save away their TM-related SPRs
3304 * (they will get restored by the TM unavailable interrupt).
3306 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3307 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3308 (current->thread.regs->msr & MSR_TM)) {
3309 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3310 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3311 run->fail_entry.hardware_entry_failure_reason = 0;
3314 /* Enable TM so we can read the TM SPRs */
3315 mtmsr(mfmsr() | MSR_TM);
3316 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3317 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3318 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3319 current->thread.regs->msr &= ~MSR_TM;
3323 kvmppc_core_prepare_to_enter(vcpu);
3325 /* No need to go into the guest when all we'll do is come back out */
3326 if (signal_pending(current)) {
3327 run->exit_reason = KVM_EXIT_INTR;
3332 atomic_inc(&kvm->arch.vcpus_running);
3333 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3336 flush_all_to_thread(current);
3338 /* Save userspace EBB and other register values */
3339 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3340 ebb_regs[0] = mfspr(SPRN_EBBHR);
3341 ebb_regs[1] = mfspr(SPRN_EBBRR);
3342 ebb_regs[2] = mfspr(SPRN_BESCR);
3343 user_tar = mfspr(SPRN_TAR);
3345 user_vrsave = mfspr(SPRN_VRSAVE);
3347 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3348 vcpu->arch.pgdir = current->mm->pgd;
3349 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3352 r = kvmppc_run_vcpu(run, vcpu);
3354 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3355 !(vcpu->arch.shregs.msr & MSR_PR)) {
3356 trace_kvm_hcall_enter(vcpu);
3357 r = kvmppc_pseries_do_hcall(vcpu);
3358 trace_kvm_hcall_exit(vcpu, r);
3359 kvmppc_core_prepare_to_enter(vcpu);
3360 } else if (r == RESUME_PAGE_FAULT) {
3361 srcu_idx = srcu_read_lock(&kvm->srcu);
3362 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3363 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3364 srcu_read_unlock(&kvm->srcu, srcu_idx);
3365 } else if (r == RESUME_PASSTHROUGH) {
3366 if (WARN_ON(xive_enabled()))
3369 r = kvmppc_xics_rm_complete(vcpu, 0);
3371 } while (is_kvmppc_resume_guest(r));
3373 /* Restore userspace EBB and other register values */
3374 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3375 mtspr(SPRN_EBBHR, ebb_regs[0]);
3376 mtspr(SPRN_EBBRR, ebb_regs[1]);
3377 mtspr(SPRN_BESCR, ebb_regs[2]);
3378 mtspr(SPRN_TAR, user_tar);
3379 mtspr(SPRN_FSCR, current->thread.fscr);
3381 mtspr(SPRN_VRSAVE, user_vrsave);
3383 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3384 atomic_dec(&kvm->arch.vcpus_running);
3388 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3389 int shift, int sllp)
3391 (*sps)->page_shift = shift;
3392 (*sps)->slb_enc = sllp;
3393 (*sps)->enc[0].page_shift = shift;
3394 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3396 * Add 16MB MPSS support (may get filtered out by userspace)
3399 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3401 (*sps)->enc[1].page_shift = 24;
3402 (*sps)->enc[1].pte_enc = penc;
3408 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3409 struct kvm_ppc_smmu_info *info)
3411 struct kvm_ppc_one_seg_page_size *sps;
3414 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3415 * POWER7 doesn't support keys for instruction accesses,
3416 * POWER8 and POWER9 do.
3418 info->data_keys = 32;
3419 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3421 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3422 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3423 info->slb_size = 32;
3425 /* We only support these sizes for now, and no muti-size segments */
3426 sps = &info->sps[0];
3427 kvmppc_add_seg_page_size(&sps, 12, 0);
3428 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3429 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3435 * Get (and clear) the dirty memory log for a memory slot.
3437 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3438 struct kvm_dirty_log *log)
3440 struct kvm_memslots *slots;
3441 struct kvm_memory_slot *memslot;
3444 unsigned long *buf, *p;
3445 struct kvm_vcpu *vcpu;
3447 mutex_lock(&kvm->slots_lock);
3450 if (log->slot >= KVM_USER_MEM_SLOTS)
3453 slots = kvm_memslots(kvm);
3454 memslot = id_to_memslot(slots, log->slot);
3456 if (!memslot->dirty_bitmap)
3460 * Use second half of bitmap area because both HPT and radix
3461 * accumulate bits in the first half.
3463 n = kvm_dirty_bitmap_bytes(memslot);
3464 buf = memslot->dirty_bitmap + n / sizeof(long);
3467 if (kvm_is_radix(kvm))
3468 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3470 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3475 * We accumulate dirty bits in the first half of the
3476 * memslot's dirty_bitmap area, for when pages are paged
3477 * out or modified by the host directly. Pick up these
3478 * bits and add them to the map.
3480 p = memslot->dirty_bitmap;
3481 for (i = 0; i < n / sizeof(long); ++i)
3482 buf[i] |= xchg(&p[i], 0);
3484 /* Harvest dirty bits from VPA and DTL updates */
3485 /* Note: we never modify the SLB shadow buffer areas */
3486 kvm_for_each_vcpu(i, vcpu, kvm) {
3487 spin_lock(&vcpu->arch.vpa_update_lock);
3488 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3489 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3490 spin_unlock(&vcpu->arch.vpa_update_lock);
3494 if (copy_to_user(log->dirty_bitmap, buf, n))
3499 mutex_unlock(&kvm->slots_lock);
3503 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3504 struct kvm_memory_slot *dont)
3506 if (!dont || free->arch.rmap != dont->arch.rmap) {
3507 vfree(free->arch.rmap);
3508 free->arch.rmap = NULL;
3512 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3513 unsigned long npages)
3515 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3516 if (!slot->arch.rmap)
3522 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3523 struct kvm_memory_slot *memslot,
3524 const struct kvm_userspace_memory_region *mem)
3529 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3530 const struct kvm_userspace_memory_region *mem,
3531 const struct kvm_memory_slot *old,
3532 const struct kvm_memory_slot *new)
3534 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3537 * If we are making a new memslot, it might make
3538 * some address that was previously cached as emulated
3539 * MMIO be no longer emulated MMIO, so invalidate
3540 * all the caches of emulated MMIO translations.
3543 atomic64_inc(&kvm->arch.mmio_update);
3547 * Update LPCR values in kvm->arch and in vcores.
3548 * Caller must hold kvm->lock.
3550 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3555 if ((kvm->arch.lpcr & mask) == lpcr)
3558 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3560 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3561 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3564 spin_lock(&vc->lock);
3565 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3566 spin_unlock(&vc->lock);
3567 if (++cores_done >= kvm->arch.online_vcores)
3572 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3577 static void kvmppc_setup_partition_table(struct kvm *kvm)
3579 unsigned long dw0, dw1;
3581 if (!kvm_is_radix(kvm)) {
3582 /* PS field - page size for VRMA */
3583 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3584 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3585 /* HTABSIZE and HTABORG fields */
3586 dw0 |= kvm->arch.sdr1;
3588 /* Second dword as set by userspace */
3589 dw1 = kvm->arch.process_table;
3591 dw0 = PATB_HR | radix__get_tree_size() |
3592 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3593 dw1 = PATB_GR | kvm->arch.process_table;
3596 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3600 * Set up HPT (hashed page table) and RMA (real-mode area).
3601 * Must be called with kvm->lock held.
3603 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3606 struct kvm *kvm = vcpu->kvm;
3608 struct kvm_memory_slot *memslot;
3609 struct vm_area_struct *vma;
3610 unsigned long lpcr = 0, senc;
3611 unsigned long psize, porder;
3614 /* Allocate hashed page table (if not done already) and reset it */
3615 if (!kvm->arch.hpt.virt) {
3616 int order = KVM_DEFAULT_HPT_ORDER;
3617 struct kvm_hpt_info info;
3619 err = kvmppc_allocate_hpt(&info, order);
3620 /* If we get here, it means userspace didn't specify a
3621 * size explicitly. So, try successively smaller
3622 * sizes if the default failed. */
3623 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3624 err = kvmppc_allocate_hpt(&info, order);
3627 pr_err("KVM: Couldn't alloc HPT\n");
3631 kvmppc_set_hpt(kvm, &info);
3634 /* Look up the memslot for guest physical address 0 */
3635 srcu_idx = srcu_read_lock(&kvm->srcu);
3636 memslot = gfn_to_memslot(kvm, 0);
3638 /* We must have some memory at 0 by now */
3640 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3643 /* Look up the VMA for the start of this memory slot */
3644 hva = memslot->userspace_addr;
3645 down_read(¤t->mm->mmap_sem);
3646 vma = find_vma(current->mm, hva);
3647 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3650 psize = vma_kernel_pagesize(vma);
3651 porder = __ilog2(psize);
3653 up_read(¤t->mm->mmap_sem);
3655 /* We can handle 4k, 64k or 16M pages in the VRMA */
3657 if (!(psize == 0x1000 || psize == 0x10000 ||
3658 psize == 0x1000000))
3661 senc = slb_pgsize_encoding(psize);
3662 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3663 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3664 /* Create HPTEs in the hash page table for the VRMA */
3665 kvmppc_map_vrma(vcpu, memslot, porder);
3667 /* Update VRMASD field in the LPCR */
3668 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3669 /* the -4 is to account for senc values starting at 0x10 */
3670 lpcr = senc << (LPCR_VRMASD_SH - 4);
3671 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3674 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3678 srcu_read_unlock(&kvm->srcu, srcu_idx);
3683 up_read(¤t->mm->mmap_sem);
3687 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3688 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3690 kvmppc_free_radix(kvm);
3691 kvmppc_update_lpcr(kvm, LPCR_VPM1,
3692 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3693 kvmppc_rmap_reset(kvm);
3694 kvm->arch.radix = 0;
3695 kvm->arch.process_table = 0;
3699 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3700 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3704 err = kvmppc_init_vm_radix(kvm);
3708 kvmppc_free_hpt(&kvm->arch.hpt);
3709 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3710 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3711 kvm->arch.radix = 1;
3715 #ifdef CONFIG_KVM_XICS
3717 * Allocate a per-core structure for managing state about which cores are
3718 * running in the host versus the guest and for exchanging data between
3719 * real mode KVM and CPU running in the host.
3720 * This is only done for the first VM.
3721 * The allocated structure stays even if all VMs have stopped.
3722 * It is only freed when the kvm-hv module is unloaded.
3723 * It's OK for this routine to fail, we just don't support host
3724 * core operations like redirecting H_IPI wakeups.
3726 void kvmppc_alloc_host_rm_ops(void)
3728 struct kvmppc_host_rm_ops *ops;
3729 unsigned long l_ops;
3733 /* Not the first time here ? */
3734 if (kvmppc_host_rm_ops_hv != NULL)
3737 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3741 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3742 ops->rm_core = kzalloc(size, GFP_KERNEL);
3744 if (!ops->rm_core) {
3751 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3752 if (!cpu_online(cpu))
3755 core = cpu >> threads_shift;
3756 ops->rm_core[core].rm_state.in_host = 1;
3759 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3762 * Make the contents of the kvmppc_host_rm_ops structure visible
3763 * to other CPUs before we assign it to the global variable.
3764 * Do an atomic assignment (no locks used here), but if someone
3765 * beats us to it, just free our copy and return.
3768 l_ops = (unsigned long) ops;
3770 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3772 kfree(ops->rm_core);
3777 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3778 "ppc/kvm_book3s:prepare",
3779 kvmppc_set_host_core,
3780 kvmppc_clear_host_core);
3784 void kvmppc_free_host_rm_ops(void)
3786 if (kvmppc_host_rm_ops_hv) {
3787 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3788 kfree(kvmppc_host_rm_ops_hv->rm_core);
3789 kfree(kvmppc_host_rm_ops_hv);
3790 kvmppc_host_rm_ops_hv = NULL;
3795 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3797 unsigned long lpcr, lpid;
3801 /* Allocate the guest's logical partition ID */
3803 lpid = kvmppc_alloc_lpid();
3806 kvm->arch.lpid = lpid;
3808 kvmppc_alloc_host_rm_ops();
3811 * Since we don't flush the TLB when tearing down a VM,
3812 * and this lpid might have previously been used,
3813 * make sure we flush on each core before running the new VM.
3814 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3815 * does this flush for us.
3817 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3818 cpumask_setall(&kvm->arch.need_tlb_flush);
3820 /* Start out with the default set of hcalls enabled */
3821 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3822 sizeof(kvm->arch.enabled_hcalls));
3824 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3825 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3827 /* Init LPCR for virtual RMA mode */
3828 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3829 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3830 lpcr &= LPCR_PECE | LPCR_LPES;
3831 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3832 LPCR_VPM0 | LPCR_VPM1;
3833 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3834 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3835 /* On POWER8 turn on online bit to enable PURR/SPURR */
3836 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3839 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3840 * Set HVICE bit to enable hypervisor virtualization interrupts.
3841 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3842 * be unnecessary but better safe than sorry in case we re-enable
3843 * EE in HV mode with this LPCR still set)
3845 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3847 lpcr |= LPCR_HVICE | LPCR_HEIC;
3850 * If xive is enabled, we route 0x500 interrupts directly
3858 * If the host uses radix, the guest starts out as radix.
3860 if (radix_enabled()) {
3861 kvm->arch.radix = 1;
3862 kvm->arch.mmu_ready = 1;
3864 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3865 ret = kvmppc_init_vm_radix(kvm);
3867 kvmppc_free_lpid(kvm->arch.lpid);
3870 kvmppc_setup_partition_table(kvm);
3873 kvm->arch.lpcr = lpcr;
3875 /* Initialization for future HPT resizes */
3876 kvm->arch.resize_hpt = NULL;
3879 * Work out how many sets the TLB has, for the use of
3880 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3882 if (radix_enabled())
3883 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
3884 else if (cpu_has_feature(CPU_FTR_ARCH_300))
3885 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
3886 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3887 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
3889 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
3892 * Track that we now have a HV mode VM active. This blocks secondary
3893 * CPU threads from coming online.
3894 * On POWER9, we only need to do this if the "indep_threads_mode"
3895 * module parameter has been set to N.
3897 if (cpu_has_feature(CPU_FTR_ARCH_300))
3898 kvm->arch.threads_indep = indep_threads_mode;
3899 if (!kvm->arch.threads_indep)
3900 kvm_hv_vm_activated();
3903 * Initialize smt_mode depending on processor.
3904 * POWER8 and earlier have to use "strict" threading, where
3905 * all vCPUs in a vcore have to run on the same (sub)core,
3906 * whereas on POWER9 the threads can each run a different
3909 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3910 kvm->arch.smt_mode = threads_per_subcore;
3912 kvm->arch.smt_mode = 1;
3913 kvm->arch.emul_smt_mode = 1;
3916 * Create a debugfs directory for the VM
3918 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3919 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3920 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3921 kvmppc_mmu_debugfs_init(kvm);
3926 static void kvmppc_free_vcores(struct kvm *kvm)
3930 for (i = 0; i < KVM_MAX_VCORES; ++i)
3931 kfree(kvm->arch.vcores[i]);
3932 kvm->arch.online_vcores = 0;
3935 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3937 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3939 if (!kvm->arch.threads_indep)
3940 kvm_hv_vm_deactivated();
3942 kvmppc_free_vcores(kvm);
3944 kvmppc_free_lpid(kvm->arch.lpid);
3946 if (kvm_is_radix(kvm))
3947 kvmppc_free_radix(kvm);
3949 kvmppc_free_hpt(&kvm->arch.hpt);
3951 kvmppc_free_pimap(kvm);
3954 /* We don't need to emulate any privileged instructions or dcbz */
3955 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3956 unsigned int inst, int *advance)
3958 return EMULATE_FAIL;
3961 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3964 return EMULATE_FAIL;
3967 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3970 return EMULATE_FAIL;
3973 static int kvmppc_core_check_processor_compat_hv(void)
3975 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3976 !cpu_has_feature(CPU_FTR_ARCH_206))
3982 #ifdef CONFIG_KVM_XICS
3984 void kvmppc_free_pimap(struct kvm *kvm)
3986 kfree(kvm->arch.pimap);
3989 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3991 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3994 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3996 struct irq_desc *desc;
3997 struct kvmppc_irq_map *irq_map;
3998 struct kvmppc_passthru_irqmap *pimap;
3999 struct irq_chip *chip;
4002 if (!kvm_irq_bypass)
4005 desc = irq_to_desc(host_irq);
4009 mutex_lock(&kvm->lock);
4011 pimap = kvm->arch.pimap;
4012 if (pimap == NULL) {
4013 /* First call, allocate structure to hold IRQ map */
4014 pimap = kvmppc_alloc_pimap();
4015 if (pimap == NULL) {
4016 mutex_unlock(&kvm->lock);
4019 kvm->arch.pimap = pimap;
4023 * For now, we only support interrupts for which the EOI operation
4024 * is an OPAL call followed by a write to XIRR, since that's
4025 * what our real-mode EOI code does, or a XIVE interrupt
4027 chip = irq_data_get_irq_chip(&desc->irq_data);
4028 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4029 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4030 host_irq, guest_gsi);
4031 mutex_unlock(&kvm->lock);
4036 * See if we already have an entry for this guest IRQ number.
4037 * If it's mapped to a hardware IRQ number, that's an error,
4038 * otherwise re-use this entry.
4040 for (i = 0; i < pimap->n_mapped; i++) {
4041 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4042 if (pimap->mapped[i].r_hwirq) {
4043 mutex_unlock(&kvm->lock);
4050 if (i == KVMPPC_PIRQ_MAPPED) {
4051 mutex_unlock(&kvm->lock);
4052 return -EAGAIN; /* table is full */
4055 irq_map = &pimap->mapped[i];
4057 irq_map->v_hwirq = guest_gsi;
4058 irq_map->desc = desc;
4061 * Order the above two stores before the next to serialize with
4062 * the KVM real mode handler.
4065 irq_map->r_hwirq = desc->irq_data.hwirq;
4067 if (i == pimap->n_mapped)
4071 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4073 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4075 irq_map->r_hwirq = 0;
4077 mutex_unlock(&kvm->lock);
4082 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4084 struct irq_desc *desc;
4085 struct kvmppc_passthru_irqmap *pimap;
4088 if (!kvm_irq_bypass)
4091 desc = irq_to_desc(host_irq);
4095 mutex_lock(&kvm->lock);
4096 if (!kvm->arch.pimap)
4099 pimap = kvm->arch.pimap;
4101 for (i = 0; i < pimap->n_mapped; i++) {
4102 if (guest_gsi == pimap->mapped[i].v_hwirq)
4106 if (i == pimap->n_mapped) {
4107 mutex_unlock(&kvm->lock);
4112 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4114 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4116 /* invalidate the entry (what do do on error from the above ?) */
4117 pimap->mapped[i].r_hwirq = 0;
4120 * We don't free this structure even when the count goes to
4121 * zero. The structure is freed when we destroy the VM.
4124 mutex_unlock(&kvm->lock);
4128 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4129 struct irq_bypass_producer *prod)
4132 struct kvm_kernel_irqfd *irqfd =
4133 container_of(cons, struct kvm_kernel_irqfd, consumer);
4135 irqfd->producer = prod;
4137 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4139 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4140 prod->irq, irqfd->gsi, ret);
4145 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4146 struct irq_bypass_producer *prod)
4149 struct kvm_kernel_irqfd *irqfd =
4150 container_of(cons, struct kvm_kernel_irqfd, consumer);
4152 irqfd->producer = NULL;
4155 * When producer of consumer is unregistered, we change back to
4156 * default external interrupt handling mode - KVM real mode
4157 * will switch back to host.
4159 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4161 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4162 prod->irq, irqfd->gsi, ret);
4166 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4167 unsigned int ioctl, unsigned long arg)
4169 struct kvm *kvm __maybe_unused = filp->private_data;
4170 void __user *argp = (void __user *)arg;
4175 case KVM_PPC_ALLOCATE_HTAB: {
4179 if (get_user(htab_order, (u32 __user *)argp))
4181 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4188 case KVM_PPC_GET_HTAB_FD: {
4189 struct kvm_get_htab_fd ghf;
4192 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4194 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4198 case KVM_PPC_RESIZE_HPT_PREPARE: {
4199 struct kvm_ppc_resize_hpt rhpt;
4202 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4205 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4209 case KVM_PPC_RESIZE_HPT_COMMIT: {
4210 struct kvm_ppc_resize_hpt rhpt;
4213 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4216 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4228 * List of hcall numbers to enable by default.
4229 * For compatibility with old userspace, we enable by default
4230 * all hcalls that were implemented before the hcall-enabling
4231 * facility was added. Note this list should not include H_RTAS.
4233 static unsigned int default_hcall_list[] = {
4247 #ifdef CONFIG_KVM_XICS
4258 static void init_default_hcalls(void)
4263 for (i = 0; default_hcall_list[i]; ++i) {
4264 hcall = default_hcall_list[i];
4265 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4266 __set_bit(hcall / 4, default_enabled_hcalls);
4270 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4276 /* If not on a POWER9, reject it */
4277 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4280 /* If any unknown flags set, reject it */
4281 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4284 /* GR (guest radix) bit in process_table field must match */
4285 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4286 if (!!(cfg->process_table & PATB_GR) != radix)
4289 /* Process table size field must be reasonable, i.e. <= 24 */
4290 if ((cfg->process_table & PRTS_MASK) > 24)
4293 /* We can change a guest to/from radix now, if the host is radix */
4294 if (radix && !radix_enabled())
4297 mutex_lock(&kvm->lock);
4298 if (radix != kvm_is_radix(kvm)) {
4299 if (kvm->arch.mmu_ready) {
4300 kvm->arch.mmu_ready = 0;
4301 /* order mmu_ready vs. vcpus_running */
4303 if (atomic_read(&kvm->arch.vcpus_running)) {
4304 kvm->arch.mmu_ready = 1;
4310 err = kvmppc_switch_mmu_to_radix(kvm);
4312 err = kvmppc_switch_mmu_to_hpt(kvm);
4317 kvm->arch.process_table = cfg->process_table;
4318 kvmppc_setup_partition_table(kvm);
4320 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4321 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4325 mutex_unlock(&kvm->lock);
4329 static struct kvmppc_ops kvm_ops_hv = {
4330 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4331 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4332 .get_one_reg = kvmppc_get_one_reg_hv,
4333 .set_one_reg = kvmppc_set_one_reg_hv,
4334 .vcpu_load = kvmppc_core_vcpu_load_hv,
4335 .vcpu_put = kvmppc_core_vcpu_put_hv,
4336 .set_msr = kvmppc_set_msr_hv,
4337 .vcpu_run = kvmppc_vcpu_run_hv,
4338 .vcpu_create = kvmppc_core_vcpu_create_hv,
4339 .vcpu_free = kvmppc_core_vcpu_free_hv,
4340 .check_requests = kvmppc_core_check_requests_hv,
4341 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4342 .flush_memslot = kvmppc_core_flush_memslot_hv,
4343 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4344 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4345 .unmap_hva = kvm_unmap_hva_hv,
4346 .unmap_hva_range = kvm_unmap_hva_range_hv,
4347 .age_hva = kvm_age_hva_hv,
4348 .test_age_hva = kvm_test_age_hva_hv,
4349 .set_spte_hva = kvm_set_spte_hva_hv,
4350 .mmu_destroy = kvmppc_mmu_destroy_hv,
4351 .free_memslot = kvmppc_core_free_memslot_hv,
4352 .create_memslot = kvmppc_core_create_memslot_hv,
4353 .init_vm = kvmppc_core_init_vm_hv,
4354 .destroy_vm = kvmppc_core_destroy_vm_hv,
4355 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4356 .emulate_op = kvmppc_core_emulate_op_hv,
4357 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4358 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4359 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4360 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4361 .hcall_implemented = kvmppc_hcall_impl_hv,
4362 #ifdef CONFIG_KVM_XICS
4363 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4364 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4366 .configure_mmu = kvmhv_configure_mmu,
4367 .get_rmmu_info = kvmhv_get_rmmu_info,
4368 .set_smt_mode = kvmhv_set_smt_mode,
4371 static int kvm_init_subcore_bitmap(void)
4374 int nr_cores = cpu_nr_cores();
4375 struct sibling_subcore_state *sibling_subcore_state;
4377 for (i = 0; i < nr_cores; i++) {
4378 int first_cpu = i * threads_per_core;
4379 int node = cpu_to_node(first_cpu);
4381 /* Ignore if it is already allocated. */
4382 if (paca[first_cpu].sibling_subcore_state)
4385 sibling_subcore_state =
4386 kmalloc_node(sizeof(struct sibling_subcore_state),
4388 if (!sibling_subcore_state)
4391 memset(sibling_subcore_state, 0,
4392 sizeof(struct sibling_subcore_state));
4394 for (j = 0; j < threads_per_core; j++) {
4395 int cpu = first_cpu + j;
4397 paca[cpu].sibling_subcore_state = sibling_subcore_state;
4403 static int kvmppc_radix_possible(void)
4405 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4408 static int kvmppc_book3s_init_hv(void)
4412 * FIXME!! Do we need to check on all cpus ?
4414 r = kvmppc_core_check_processor_compat_hv();
4418 r = kvm_init_subcore_bitmap();
4423 * We need a way of accessing the XICS interrupt controller,
4424 * either directly, via paca[cpu].kvm_hstate.xics_phys, or
4425 * indirectly, via OPAL.
4428 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4429 struct device_node *np;
4431 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4433 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4439 kvm_ops_hv.owner = THIS_MODULE;
4440 kvmppc_hv_ops = &kvm_ops_hv;
4442 init_default_hcalls();
4446 r = kvmppc_mmu_hv_init();
4450 if (kvmppc_radix_possible())
4451 r = kvmppc_radix_init();
4455 static void kvmppc_book3s_exit_hv(void)
4457 kvmppc_free_host_rm_ops();
4458 if (kvmppc_radix_possible())
4459 kvmppc_radix_exit();
4460 kvmppc_hv_ops = NULL;
4463 module_init(kvmppc_book3s_init_hv);
4464 module_exit(kvmppc_book3s_exit_hv);
4465 MODULE_LICENSE("GPL");
4466 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4467 MODULE_ALIAS("devname:kvm");