2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
21 #include <linux/kvm_host.h>
22 #include <linux/kernel.h>
23 #include <linux/err.h>
24 #include <linux/slab.h>
25 #include <linux/preempt.h>
26 #include <linux/sched/signal.h>
27 #include <linux/sched/stat.h>
28 #include <linux/delay.h>
29 #include <linux/export.h>
31 #include <linux/anon_inodes.h>
32 #include <linux/cpu.h>
33 #include <linux/cpumask.h>
34 #include <linux/spinlock.h>
35 #include <linux/page-flags.h>
36 #include <linux/srcu.h>
37 #include <linux/miscdevice.h>
38 #include <linux/debugfs.h>
39 #include <linux/gfp.h>
40 #include <linux/vmalloc.h>
41 #include <linux/highmem.h>
42 #include <linux/hugetlb.h>
43 #include <linux/kvm_irqfd.h>
44 #include <linux/irqbypass.h>
45 #include <linux/module.h>
46 #include <linux/compiler.h>
50 #include <asm/ppc-opcode.h>
51 #include <asm/asm-prototypes.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <asm/tlbflush.h>
56 #include <linux/uaccess.h>
58 #include <asm/kvm_ppc.h>
59 #include <asm/kvm_book3s.h>
60 #include <asm/mmu_context.h>
61 #include <asm/lppaca.h>
62 #include <asm/processor.h>
63 #include <asm/cputhreads.h>
65 #include <asm/hvcall.h>
66 #include <asm/switch_to.h>
68 #include <asm/dbell.h>
70 #include <asm/pnv-pci.h>
78 #define CREATE_TRACE_POINTS
81 /* #define EXIT_DEBUG */
82 /* #define EXIT_DEBUG_SIMPLE */
83 /* #define EXIT_DEBUG_INT */
85 /* Used to indicate that a guest page fault needs to be handled */
86 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
87 /* Used to indicate that a guest passthrough interrupt needs to be handled */
88 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
90 /* Used as a "null" value for timebase values */
91 #define TB_NIL (~(u64)0)
93 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
95 static int dynamic_mt_modes = 6;
96 module_param(dynamic_mt_modes, int, 0644);
97 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
98 static int target_smt_mode;
99 module_param(target_smt_mode, int, 0644);
100 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
102 static bool indep_threads_mode = true;
103 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
106 #ifdef CONFIG_KVM_XICS
107 static struct kernel_param_ops module_param_ops = {
108 .set = param_set_int,
109 .get = param_get_int,
112 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
113 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
115 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
116 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
119 /* If set, the threads on each CPU core have to be in the same MMU mode */
120 static bool no_mixing_hpt_and_radix;
122 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
123 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
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_ptrs[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.
501 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
502 if (len < sizeof(struct lppaca))
504 vpap = &tvcpu->arch.vpa;
508 case H_VPA_REG_DTL: /* register DTL */
509 if (len < sizeof(struct dtl_entry))
511 len -= len % sizeof(struct dtl_entry);
513 /* Check that they have previously registered a VPA */
515 if (!vpa_is_registered(&tvcpu->arch.vpa))
518 vpap = &tvcpu->arch.dtl;
522 case H_VPA_REG_SLB: /* register SLB shadow buffer */
523 /* Check that they have previously registered a VPA */
525 if (!vpa_is_registered(&tvcpu->arch.vpa))
528 vpap = &tvcpu->arch.slb_shadow;
532 case H_VPA_DEREG_VPA: /* deregister VPA */
533 /* Check they don't still have a DTL or SLB buf registered */
535 if (vpa_is_registered(&tvcpu->arch.dtl) ||
536 vpa_is_registered(&tvcpu->arch.slb_shadow))
539 vpap = &tvcpu->arch.vpa;
543 case H_VPA_DEREG_DTL: /* deregister DTL */
544 vpap = &tvcpu->arch.dtl;
548 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
549 vpap = &tvcpu->arch.slb_shadow;
555 vpap->next_gpa = vpa;
557 vpap->update_pending = 1;
560 spin_unlock(&tvcpu->arch.vpa_update_lock);
565 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
567 struct kvm *kvm = vcpu->kvm;
573 * We need to pin the page pointed to by vpap->next_gpa,
574 * but we can't call kvmppc_pin_guest_page under the lock
575 * as it does get_user_pages() and down_read(). So we
576 * have to drop the lock, pin the page, then get the lock
577 * again and check that a new area didn't get registered
581 gpa = vpap->next_gpa;
582 spin_unlock(&vcpu->arch.vpa_update_lock);
586 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
587 spin_lock(&vcpu->arch.vpa_update_lock);
588 if (gpa == vpap->next_gpa)
590 /* sigh... unpin that one and try again */
592 kvmppc_unpin_guest_page(kvm, va, gpa, false);
595 vpap->update_pending = 0;
596 if (va && nb < vpap->len) {
598 * If it's now too short, it must be that userspace
599 * has changed the mappings underlying guest memory,
600 * so unregister the region.
602 kvmppc_unpin_guest_page(kvm, va, gpa, false);
605 if (vpap->pinned_addr)
606 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
609 vpap->pinned_addr = va;
612 vpap->pinned_end = va + vpap->len;
615 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
617 if (!(vcpu->arch.vpa.update_pending ||
618 vcpu->arch.slb_shadow.update_pending ||
619 vcpu->arch.dtl.update_pending))
622 spin_lock(&vcpu->arch.vpa_update_lock);
623 if (vcpu->arch.vpa.update_pending) {
624 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
625 if (vcpu->arch.vpa.pinned_addr)
626 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
628 if (vcpu->arch.dtl.update_pending) {
629 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
630 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
631 vcpu->arch.dtl_index = 0;
633 if (vcpu->arch.slb_shadow.update_pending)
634 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
635 spin_unlock(&vcpu->arch.vpa_update_lock);
639 * Return the accumulated stolen time for the vcore up until `now'.
640 * The caller should hold the vcore lock.
642 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
647 spin_lock_irqsave(&vc->stoltb_lock, flags);
649 if (vc->vcore_state != VCORE_INACTIVE &&
650 vc->preempt_tb != TB_NIL)
651 p += now - vc->preempt_tb;
652 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
656 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
657 struct kvmppc_vcore *vc)
659 struct dtl_entry *dt;
661 unsigned long stolen;
662 unsigned long core_stolen;
666 dt = vcpu->arch.dtl_ptr;
667 vpa = vcpu->arch.vpa.pinned_addr;
669 core_stolen = vcore_stolen_time(vc, now);
670 stolen = core_stolen - vcpu->arch.stolen_logged;
671 vcpu->arch.stolen_logged = core_stolen;
672 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
673 stolen += vcpu->arch.busy_stolen;
674 vcpu->arch.busy_stolen = 0;
675 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
678 memset(dt, 0, sizeof(struct dtl_entry));
679 dt->dispatch_reason = 7;
680 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
681 dt->timebase = cpu_to_be64(now + vc->tb_offset);
682 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
683 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
684 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
686 if (dt == vcpu->arch.dtl.pinned_end)
687 dt = vcpu->arch.dtl.pinned_addr;
688 vcpu->arch.dtl_ptr = dt;
689 /* order writing *dt vs. writing vpa->dtl_idx */
691 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
692 vcpu->arch.dtl.dirty = true;
695 /* See if there is a doorbell interrupt pending for a vcpu */
696 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
699 struct kvmppc_vcore *vc;
701 if (vcpu->arch.doorbell_request)
704 * Ensure that the read of vcore->dpdes comes after the read
705 * of vcpu->doorbell_request. This barrier matches the
706 * lwsync in book3s_hv_rmhandlers.S just before the
707 * fast_guest_return label.
710 vc = vcpu->arch.vcore;
711 thr = vcpu->vcpu_id - vc->first_vcpuid;
712 return !!(vc->dpdes & (1 << thr));
715 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
717 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
719 if ((!vcpu->arch.vcore->arch_compat) &&
720 cpu_has_feature(CPU_FTR_ARCH_207S))
725 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
726 unsigned long resource, unsigned long value1,
727 unsigned long value2)
730 case H_SET_MODE_RESOURCE_SET_CIABR:
731 if (!kvmppc_power8_compatible(vcpu))
736 return H_UNSUPPORTED_FLAG_START;
737 /* Guests can't breakpoint the hypervisor */
738 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
740 vcpu->arch.ciabr = value1;
742 case H_SET_MODE_RESOURCE_SET_DAWR:
743 if (!kvmppc_power8_compatible(vcpu))
745 if (!ppc_breakpoint_available())
748 return H_UNSUPPORTED_FLAG_START;
749 if (value2 & DABRX_HYP)
751 vcpu->arch.dawr = value1;
752 vcpu->arch.dawrx = value2;
759 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
761 struct kvmppc_vcore *vcore = target->arch.vcore;
764 * We expect to have been called by the real mode handler
765 * (kvmppc_rm_h_confer()) which would have directly returned
766 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
767 * have useful work to do and should not confer) so we don't
771 spin_lock(&vcore->lock);
772 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
773 vcore->vcore_state != VCORE_INACTIVE &&
775 target = vcore->runner;
776 spin_unlock(&vcore->lock);
778 return kvm_vcpu_yield_to(target);
781 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
784 struct lppaca *lppaca;
786 spin_lock(&vcpu->arch.vpa_update_lock);
787 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
789 yield_count = be32_to_cpu(lppaca->yield_count);
790 spin_unlock(&vcpu->arch.vpa_update_lock);
794 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
796 unsigned long req = kvmppc_get_gpr(vcpu, 3);
797 unsigned long target, ret = H_SUCCESS;
799 struct kvm_vcpu *tvcpu;
802 if (req <= MAX_HCALL_OPCODE &&
803 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
810 target = kvmppc_get_gpr(vcpu, 4);
811 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
816 tvcpu->arch.prodded = 1;
818 if (tvcpu->arch.ceded)
819 kvmppc_fast_vcpu_kick_hv(tvcpu);
822 target = kvmppc_get_gpr(vcpu, 4);
825 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
830 yield_count = kvmppc_get_gpr(vcpu, 5);
831 if (kvmppc_get_yield_count(tvcpu) != yield_count)
833 kvm_arch_vcpu_yield_to(tvcpu);
836 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
837 kvmppc_get_gpr(vcpu, 5),
838 kvmppc_get_gpr(vcpu, 6));
841 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
844 idx = srcu_read_lock(&vcpu->kvm->srcu);
845 rc = kvmppc_rtas_hcall(vcpu);
846 srcu_read_unlock(&vcpu->kvm->srcu, idx);
853 /* Send the error out to userspace via KVM_RUN */
855 case H_LOGICAL_CI_LOAD:
856 ret = kvmppc_h_logical_ci_load(vcpu);
857 if (ret == H_TOO_HARD)
860 case H_LOGICAL_CI_STORE:
861 ret = kvmppc_h_logical_ci_store(vcpu);
862 if (ret == H_TOO_HARD)
866 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
867 kvmppc_get_gpr(vcpu, 5),
868 kvmppc_get_gpr(vcpu, 6),
869 kvmppc_get_gpr(vcpu, 7));
870 if (ret == H_TOO_HARD)
879 if (kvmppc_xics_enabled(vcpu)) {
880 if (xive_enabled()) {
881 ret = H_NOT_AVAILABLE;
884 ret = kvmppc_xics_hcall(vcpu, req);
889 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
890 kvmppc_get_gpr(vcpu, 5),
891 kvmppc_get_gpr(vcpu, 6));
892 if (ret == H_TOO_HARD)
895 case H_PUT_TCE_INDIRECT:
896 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
897 kvmppc_get_gpr(vcpu, 5),
898 kvmppc_get_gpr(vcpu, 6),
899 kvmppc_get_gpr(vcpu, 7));
900 if (ret == H_TOO_HARD)
904 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
905 kvmppc_get_gpr(vcpu, 5),
906 kvmppc_get_gpr(vcpu, 6),
907 kvmppc_get_gpr(vcpu, 7));
908 if (ret == H_TOO_HARD)
914 kvmppc_set_gpr(vcpu, 3, ret);
915 vcpu->arch.hcall_needed = 0;
919 static int kvmppc_hcall_impl_hv(unsigned long cmd)
927 case H_LOGICAL_CI_LOAD:
928 case H_LOGICAL_CI_STORE:
929 #ifdef CONFIG_KVM_XICS
940 /* See if it's in the real-mode table */
941 return kvmppc_hcall_impl_hv_realmode(cmd);
944 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
945 struct kvm_vcpu *vcpu)
949 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
952 * Fetch failed, so return to guest and
953 * try executing it again.
958 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
959 run->exit_reason = KVM_EXIT_DEBUG;
960 run->debug.arch.address = kvmppc_get_pc(vcpu);
963 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
968 static void do_nothing(void *x)
972 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
974 int thr, cpu, pcpu, nthreads;
978 nthreads = vcpu->kvm->arch.emul_smt_mode;
980 cpu = vcpu->vcpu_id & ~(nthreads - 1);
981 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
982 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
986 * If the vcpu is currently running on a physical cpu thread,
987 * interrupt it in order to pull it out of the guest briefly,
988 * which will update its vcore->dpdes value.
990 pcpu = READ_ONCE(v->cpu);
992 smp_call_function_single(pcpu, do_nothing, NULL, 1);
993 if (kvmppc_doorbell_pending(v))
1000 * On POWER9, emulate doorbell-related instructions in order to
1001 * give the guest the illusion of running on a multi-threaded core.
1002 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1005 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1009 struct kvm *kvm = vcpu->kvm;
1010 struct kvm_vcpu *tvcpu;
1012 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1013 return RESUME_GUEST;
1014 if (get_op(inst) != 31)
1015 return EMULATE_FAIL;
1017 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1018 switch (get_xop(inst)) {
1019 case OP_31_XOP_MSGSNDP:
1020 arg = kvmppc_get_gpr(vcpu, rb);
1021 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1024 if (arg >= kvm->arch.emul_smt_mode)
1026 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1029 if (!tvcpu->arch.doorbell_request) {
1030 tvcpu->arch.doorbell_request = 1;
1031 kvmppc_fast_vcpu_kick_hv(tvcpu);
1034 case OP_31_XOP_MSGCLRP:
1035 arg = kvmppc_get_gpr(vcpu, rb);
1036 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1038 vcpu->arch.vcore->dpdes = 0;
1039 vcpu->arch.doorbell_request = 0;
1041 case OP_31_XOP_MFSPR:
1042 switch (get_sprn(inst)) {
1047 arg = kvmppc_read_dpdes(vcpu);
1050 return EMULATE_FAIL;
1052 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1055 return EMULATE_FAIL;
1057 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1058 return RESUME_GUEST;
1061 /* Called with vcpu->arch.vcore->lock held */
1062 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1063 struct task_struct *tsk)
1065 int r = RESUME_HOST;
1067 vcpu->stat.sum_exits++;
1070 * This can happen if an interrupt occurs in the last stages
1071 * of guest entry or the first stages of guest exit (i.e. after
1072 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1073 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1074 * That can happen due to a bug, or due to a machine check
1075 * occurring at just the wrong time.
1077 if (vcpu->arch.shregs.msr & MSR_HV) {
1078 printk(KERN_EMERG "KVM trap in HV mode!\n");
1079 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1080 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1081 vcpu->arch.shregs.msr);
1082 kvmppc_dump_regs(vcpu);
1083 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1084 run->hw.hardware_exit_reason = vcpu->arch.trap;
1087 run->exit_reason = KVM_EXIT_UNKNOWN;
1088 run->ready_for_interrupt_injection = 1;
1089 switch (vcpu->arch.trap) {
1090 /* We're good on these - the host merely wanted to get our attention */
1091 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1092 vcpu->stat.dec_exits++;
1095 case BOOK3S_INTERRUPT_EXTERNAL:
1096 case BOOK3S_INTERRUPT_H_DOORBELL:
1097 case BOOK3S_INTERRUPT_H_VIRT:
1098 vcpu->stat.ext_intr_exits++;
1101 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1102 case BOOK3S_INTERRUPT_HMI:
1103 case BOOK3S_INTERRUPT_PERFMON:
1104 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1107 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1108 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1109 run->exit_reason = KVM_EXIT_NMI;
1110 run->hw.hardware_exit_reason = vcpu->arch.trap;
1111 /* Clear out the old NMI status from run->flags */
1112 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1113 /* Now set the NMI status */
1114 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1115 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1117 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1120 /* Print the MCE event to host console. */
1121 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1123 case BOOK3S_INTERRUPT_PROGRAM:
1127 * Normally program interrupts are delivered directly
1128 * to the guest by the hardware, but we can get here
1129 * as a result of a hypervisor emulation interrupt
1130 * (e40) getting turned into a 700 by BML RTAS.
1132 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1133 kvmppc_core_queue_program(vcpu, flags);
1137 case BOOK3S_INTERRUPT_SYSCALL:
1139 /* hcall - punt to userspace */
1142 /* hypercall with MSR_PR has already been handled in rmode,
1143 * and never reaches here.
1146 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1147 for (i = 0; i < 9; ++i)
1148 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1149 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1150 vcpu->arch.hcall_needed = 1;
1155 * We get these next two if the guest accesses a page which it thinks
1156 * it has mapped but which is not actually present, either because
1157 * it is for an emulated I/O device or because the corresonding
1158 * host page has been paged out. Any other HDSI/HISI interrupts
1159 * have been handled already.
1161 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1162 r = RESUME_PAGE_FAULT;
1164 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1165 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1166 vcpu->arch.fault_dsisr = 0;
1167 r = RESUME_PAGE_FAULT;
1170 * This occurs if the guest executes an illegal instruction.
1171 * If the guest debug is disabled, generate a program interrupt
1172 * to the guest. If guest debug is enabled, we need to check
1173 * whether the instruction is a software breakpoint instruction.
1174 * Accordingly return to Guest or Host.
1176 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1177 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1178 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1179 swab32(vcpu->arch.emul_inst) :
1180 vcpu->arch.emul_inst;
1181 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1182 /* Need vcore unlocked to call kvmppc_get_last_inst */
1183 spin_unlock(&vcpu->arch.vcore->lock);
1184 r = kvmppc_emulate_debug_inst(run, vcpu);
1185 spin_lock(&vcpu->arch.vcore->lock);
1187 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1192 * This occurs if the guest (kernel or userspace), does something that
1193 * is prohibited by HFSCR.
1194 * On POWER9, this could be a doorbell instruction that we need
1196 * Otherwise, we just generate a program interrupt to the guest.
1198 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1200 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1201 cpu_has_feature(CPU_FTR_ARCH_300)) {
1202 /* Need vcore unlocked to call kvmppc_get_last_inst */
1203 spin_unlock(&vcpu->arch.vcore->lock);
1204 r = kvmppc_emulate_doorbell_instr(vcpu);
1205 spin_lock(&vcpu->arch.vcore->lock);
1207 if (r == EMULATE_FAIL) {
1208 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1213 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1214 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1216 * This occurs for various TM-related instructions that
1217 * we need to emulate on POWER9 DD2.2. We have already
1218 * handled the cases where the guest was in real-suspend
1219 * mode and was transitioning to transactional state.
1221 r = kvmhv_p9_tm_emulation(vcpu);
1225 case BOOK3S_INTERRUPT_HV_RM_HARD:
1226 r = RESUME_PASSTHROUGH;
1229 kvmppc_dump_regs(vcpu);
1230 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1231 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1232 vcpu->arch.shregs.msr);
1233 run->hw.hardware_exit_reason = vcpu->arch.trap;
1241 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1242 struct kvm_sregs *sregs)
1246 memset(sregs, 0, sizeof(struct kvm_sregs));
1247 sregs->pvr = vcpu->arch.pvr;
1248 for (i = 0; i < vcpu->arch.slb_max; i++) {
1249 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1250 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1256 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1257 struct kvm_sregs *sregs)
1261 /* Only accept the same PVR as the host's, since we can't spoof it */
1262 if (sregs->pvr != vcpu->arch.pvr)
1266 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1267 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1268 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1269 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1273 vcpu->arch.slb_max = j;
1278 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1279 bool preserve_top32)
1281 struct kvm *kvm = vcpu->kvm;
1282 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1285 mutex_lock(&kvm->lock);
1286 spin_lock(&vc->lock);
1288 * If ILE (interrupt little-endian) has changed, update the
1289 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1291 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1292 struct kvm_vcpu *vcpu;
1295 kvm_for_each_vcpu(i, vcpu, kvm) {
1296 if (vcpu->arch.vcore != vc)
1298 if (new_lpcr & LPCR_ILE)
1299 vcpu->arch.intr_msr |= MSR_LE;
1301 vcpu->arch.intr_msr &= ~MSR_LE;
1306 * Userspace can only modify DPFD (default prefetch depth),
1307 * ILE (interrupt little-endian) and TC (translation control).
1308 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1310 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1311 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1314 * On POWER9, allow userspace to enable large decrementer for the
1315 * guest, whether or not the host has it enabled.
1317 if (cpu_has_feature(CPU_FTR_ARCH_300))
1320 /* Broken 32-bit version of LPCR must not clear top bits */
1323 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1324 spin_unlock(&vc->lock);
1325 mutex_unlock(&kvm->lock);
1328 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1329 union kvmppc_one_reg *val)
1335 case KVM_REG_PPC_DEBUG_INST:
1336 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1338 case KVM_REG_PPC_HIOR:
1339 *val = get_reg_val(id, 0);
1341 case KVM_REG_PPC_DABR:
1342 *val = get_reg_val(id, vcpu->arch.dabr);
1344 case KVM_REG_PPC_DABRX:
1345 *val = get_reg_val(id, vcpu->arch.dabrx);
1347 case KVM_REG_PPC_DSCR:
1348 *val = get_reg_val(id, vcpu->arch.dscr);
1350 case KVM_REG_PPC_PURR:
1351 *val = get_reg_val(id, vcpu->arch.purr);
1353 case KVM_REG_PPC_SPURR:
1354 *val = get_reg_val(id, vcpu->arch.spurr);
1356 case KVM_REG_PPC_AMR:
1357 *val = get_reg_val(id, vcpu->arch.amr);
1359 case KVM_REG_PPC_UAMOR:
1360 *val = get_reg_val(id, vcpu->arch.uamor);
1362 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1363 i = id - KVM_REG_PPC_MMCR0;
1364 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1366 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1367 i = id - KVM_REG_PPC_PMC1;
1368 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1370 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1371 i = id - KVM_REG_PPC_SPMC1;
1372 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1374 case KVM_REG_PPC_SIAR:
1375 *val = get_reg_val(id, vcpu->arch.siar);
1377 case KVM_REG_PPC_SDAR:
1378 *val = get_reg_val(id, vcpu->arch.sdar);
1380 case KVM_REG_PPC_SIER:
1381 *val = get_reg_val(id, vcpu->arch.sier);
1383 case KVM_REG_PPC_IAMR:
1384 *val = get_reg_val(id, vcpu->arch.iamr);
1386 case KVM_REG_PPC_PSPB:
1387 *val = get_reg_val(id, vcpu->arch.pspb);
1389 case KVM_REG_PPC_DPDES:
1390 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1392 case KVM_REG_PPC_VTB:
1393 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1395 case KVM_REG_PPC_DAWR:
1396 *val = get_reg_val(id, vcpu->arch.dawr);
1398 case KVM_REG_PPC_DAWRX:
1399 *val = get_reg_val(id, vcpu->arch.dawrx);
1401 case KVM_REG_PPC_CIABR:
1402 *val = get_reg_val(id, vcpu->arch.ciabr);
1404 case KVM_REG_PPC_CSIGR:
1405 *val = get_reg_val(id, vcpu->arch.csigr);
1407 case KVM_REG_PPC_TACR:
1408 *val = get_reg_val(id, vcpu->arch.tacr);
1410 case KVM_REG_PPC_TCSCR:
1411 *val = get_reg_val(id, vcpu->arch.tcscr);
1413 case KVM_REG_PPC_PID:
1414 *val = get_reg_val(id, vcpu->arch.pid);
1416 case KVM_REG_PPC_ACOP:
1417 *val = get_reg_val(id, vcpu->arch.acop);
1419 case KVM_REG_PPC_WORT:
1420 *val = get_reg_val(id, vcpu->arch.wort);
1422 case KVM_REG_PPC_TIDR:
1423 *val = get_reg_val(id, vcpu->arch.tid);
1425 case KVM_REG_PPC_PSSCR:
1426 *val = get_reg_val(id, vcpu->arch.psscr);
1428 case KVM_REG_PPC_VPA_ADDR:
1429 spin_lock(&vcpu->arch.vpa_update_lock);
1430 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1431 spin_unlock(&vcpu->arch.vpa_update_lock);
1433 case KVM_REG_PPC_VPA_SLB:
1434 spin_lock(&vcpu->arch.vpa_update_lock);
1435 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1436 val->vpaval.length = vcpu->arch.slb_shadow.len;
1437 spin_unlock(&vcpu->arch.vpa_update_lock);
1439 case KVM_REG_PPC_VPA_DTL:
1440 spin_lock(&vcpu->arch.vpa_update_lock);
1441 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1442 val->vpaval.length = vcpu->arch.dtl.len;
1443 spin_unlock(&vcpu->arch.vpa_update_lock);
1445 case KVM_REG_PPC_TB_OFFSET:
1446 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1448 case KVM_REG_PPC_LPCR:
1449 case KVM_REG_PPC_LPCR_64:
1450 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1452 case KVM_REG_PPC_PPR:
1453 *val = get_reg_val(id, vcpu->arch.ppr);
1455 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1456 case KVM_REG_PPC_TFHAR:
1457 *val = get_reg_val(id, vcpu->arch.tfhar);
1459 case KVM_REG_PPC_TFIAR:
1460 *val = get_reg_val(id, vcpu->arch.tfiar);
1462 case KVM_REG_PPC_TEXASR:
1463 *val = get_reg_val(id, vcpu->arch.texasr);
1465 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1466 i = id - KVM_REG_PPC_TM_GPR0;
1467 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1469 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1472 i = id - KVM_REG_PPC_TM_VSR0;
1474 for (j = 0; j < TS_FPRWIDTH; j++)
1475 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1477 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1478 val->vval = vcpu->arch.vr_tm.vr[i-32];
1484 case KVM_REG_PPC_TM_CR:
1485 *val = get_reg_val(id, vcpu->arch.cr_tm);
1487 case KVM_REG_PPC_TM_XER:
1488 *val = get_reg_val(id, vcpu->arch.xer_tm);
1490 case KVM_REG_PPC_TM_LR:
1491 *val = get_reg_val(id, vcpu->arch.lr_tm);
1493 case KVM_REG_PPC_TM_CTR:
1494 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1496 case KVM_REG_PPC_TM_FPSCR:
1497 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1499 case KVM_REG_PPC_TM_AMR:
1500 *val = get_reg_val(id, vcpu->arch.amr_tm);
1502 case KVM_REG_PPC_TM_PPR:
1503 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1505 case KVM_REG_PPC_TM_VRSAVE:
1506 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1508 case KVM_REG_PPC_TM_VSCR:
1509 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1510 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1514 case KVM_REG_PPC_TM_DSCR:
1515 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1517 case KVM_REG_PPC_TM_TAR:
1518 *val = get_reg_val(id, vcpu->arch.tar_tm);
1521 case KVM_REG_PPC_ARCH_COMPAT:
1522 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1524 case KVM_REG_PPC_DEC_EXPIRY:
1525 *val = get_reg_val(id, vcpu->arch.dec_expires +
1526 vcpu->arch.vcore->tb_offset);
1536 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1537 union kvmppc_one_reg *val)
1541 unsigned long addr, len;
1544 case KVM_REG_PPC_HIOR:
1545 /* Only allow this to be set to zero */
1546 if (set_reg_val(id, *val))
1549 case KVM_REG_PPC_DABR:
1550 vcpu->arch.dabr = set_reg_val(id, *val);
1552 case KVM_REG_PPC_DABRX:
1553 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1555 case KVM_REG_PPC_DSCR:
1556 vcpu->arch.dscr = set_reg_val(id, *val);
1558 case KVM_REG_PPC_PURR:
1559 vcpu->arch.purr = set_reg_val(id, *val);
1561 case KVM_REG_PPC_SPURR:
1562 vcpu->arch.spurr = set_reg_val(id, *val);
1564 case KVM_REG_PPC_AMR:
1565 vcpu->arch.amr = set_reg_val(id, *val);
1567 case KVM_REG_PPC_UAMOR:
1568 vcpu->arch.uamor = set_reg_val(id, *val);
1570 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1571 i = id - KVM_REG_PPC_MMCR0;
1572 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1574 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1575 i = id - KVM_REG_PPC_PMC1;
1576 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1578 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1579 i = id - KVM_REG_PPC_SPMC1;
1580 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1582 case KVM_REG_PPC_SIAR:
1583 vcpu->arch.siar = set_reg_val(id, *val);
1585 case KVM_REG_PPC_SDAR:
1586 vcpu->arch.sdar = set_reg_val(id, *val);
1588 case KVM_REG_PPC_SIER:
1589 vcpu->arch.sier = set_reg_val(id, *val);
1591 case KVM_REG_PPC_IAMR:
1592 vcpu->arch.iamr = set_reg_val(id, *val);
1594 case KVM_REG_PPC_PSPB:
1595 vcpu->arch.pspb = set_reg_val(id, *val);
1597 case KVM_REG_PPC_DPDES:
1598 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1600 case KVM_REG_PPC_VTB:
1601 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1603 case KVM_REG_PPC_DAWR:
1604 vcpu->arch.dawr = set_reg_val(id, *val);
1606 case KVM_REG_PPC_DAWRX:
1607 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1609 case KVM_REG_PPC_CIABR:
1610 vcpu->arch.ciabr = set_reg_val(id, *val);
1611 /* Don't allow setting breakpoints in hypervisor code */
1612 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1613 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1615 case KVM_REG_PPC_CSIGR:
1616 vcpu->arch.csigr = set_reg_val(id, *val);
1618 case KVM_REG_PPC_TACR:
1619 vcpu->arch.tacr = set_reg_val(id, *val);
1621 case KVM_REG_PPC_TCSCR:
1622 vcpu->arch.tcscr = set_reg_val(id, *val);
1624 case KVM_REG_PPC_PID:
1625 vcpu->arch.pid = set_reg_val(id, *val);
1627 case KVM_REG_PPC_ACOP:
1628 vcpu->arch.acop = set_reg_val(id, *val);
1630 case KVM_REG_PPC_WORT:
1631 vcpu->arch.wort = set_reg_val(id, *val);
1633 case KVM_REG_PPC_TIDR:
1634 vcpu->arch.tid = set_reg_val(id, *val);
1636 case KVM_REG_PPC_PSSCR:
1637 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1639 case KVM_REG_PPC_VPA_ADDR:
1640 addr = set_reg_val(id, *val);
1642 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1643 vcpu->arch.dtl.next_gpa))
1645 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1647 case KVM_REG_PPC_VPA_SLB:
1648 addr = val->vpaval.addr;
1649 len = val->vpaval.length;
1651 if (addr && !vcpu->arch.vpa.next_gpa)
1653 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1655 case KVM_REG_PPC_VPA_DTL:
1656 addr = val->vpaval.addr;
1657 len = val->vpaval.length;
1659 if (addr && (len < sizeof(struct dtl_entry) ||
1660 !vcpu->arch.vpa.next_gpa))
1662 len -= len % sizeof(struct dtl_entry);
1663 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1665 case KVM_REG_PPC_TB_OFFSET:
1667 * POWER9 DD1 has an erratum where writing TBU40 causes
1668 * the timebase to lose ticks. So we don't let the
1669 * timebase offset be changed on P9 DD1. (It is
1670 * initialized to zero.)
1672 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1674 /* round up to multiple of 2^24 */
1675 vcpu->arch.vcore->tb_offset =
1676 ALIGN(set_reg_val(id, *val), 1UL << 24);
1678 case KVM_REG_PPC_LPCR:
1679 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1681 case KVM_REG_PPC_LPCR_64:
1682 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1684 case KVM_REG_PPC_PPR:
1685 vcpu->arch.ppr = set_reg_val(id, *val);
1687 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1688 case KVM_REG_PPC_TFHAR:
1689 vcpu->arch.tfhar = set_reg_val(id, *val);
1691 case KVM_REG_PPC_TFIAR:
1692 vcpu->arch.tfiar = set_reg_val(id, *val);
1694 case KVM_REG_PPC_TEXASR:
1695 vcpu->arch.texasr = set_reg_val(id, *val);
1697 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1698 i = id - KVM_REG_PPC_TM_GPR0;
1699 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1701 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1704 i = id - KVM_REG_PPC_TM_VSR0;
1706 for (j = 0; j < TS_FPRWIDTH; j++)
1707 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1709 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1710 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1715 case KVM_REG_PPC_TM_CR:
1716 vcpu->arch.cr_tm = set_reg_val(id, *val);
1718 case KVM_REG_PPC_TM_XER:
1719 vcpu->arch.xer_tm = set_reg_val(id, *val);
1721 case KVM_REG_PPC_TM_LR:
1722 vcpu->arch.lr_tm = set_reg_val(id, *val);
1724 case KVM_REG_PPC_TM_CTR:
1725 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1727 case KVM_REG_PPC_TM_FPSCR:
1728 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1730 case KVM_REG_PPC_TM_AMR:
1731 vcpu->arch.amr_tm = set_reg_val(id, *val);
1733 case KVM_REG_PPC_TM_PPR:
1734 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1736 case KVM_REG_PPC_TM_VRSAVE:
1737 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1739 case KVM_REG_PPC_TM_VSCR:
1740 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1741 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1745 case KVM_REG_PPC_TM_DSCR:
1746 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1748 case KVM_REG_PPC_TM_TAR:
1749 vcpu->arch.tar_tm = set_reg_val(id, *val);
1752 case KVM_REG_PPC_ARCH_COMPAT:
1753 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1755 case KVM_REG_PPC_DEC_EXPIRY:
1756 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1757 vcpu->arch.vcore->tb_offset;
1768 * On POWER9, threads are independent and can be in different partitions.
1769 * Therefore we consider each thread to be a subcore.
1770 * There is a restriction that all threads have to be in the same
1771 * MMU mode (radix or HPT), unfortunately, but since we only support
1772 * HPT guests on a HPT host so far, that isn't an impediment yet.
1774 static int threads_per_vcore(struct kvm *kvm)
1776 if (kvm->arch.threads_indep)
1778 return threads_per_subcore;
1781 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1783 struct kvmppc_vcore *vcore;
1785 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1790 spin_lock_init(&vcore->lock);
1791 spin_lock_init(&vcore->stoltb_lock);
1792 init_swait_queue_head(&vcore->wq);
1793 vcore->preempt_tb = TB_NIL;
1794 vcore->lpcr = kvm->arch.lpcr;
1795 vcore->first_vcpuid = core * kvm->arch.smt_mode;
1797 INIT_LIST_HEAD(&vcore->preempt_list);
1802 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1803 static struct debugfs_timings_element {
1807 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1808 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1809 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1810 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1811 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1814 #define N_TIMINGS (ARRAY_SIZE(timings))
1816 struct debugfs_timings_state {
1817 struct kvm_vcpu *vcpu;
1818 unsigned int buflen;
1819 char buf[N_TIMINGS * 100];
1822 static int debugfs_timings_open(struct inode *inode, struct file *file)
1824 struct kvm_vcpu *vcpu = inode->i_private;
1825 struct debugfs_timings_state *p;
1827 p = kzalloc(sizeof(*p), GFP_KERNEL);
1831 kvm_get_kvm(vcpu->kvm);
1833 file->private_data = p;
1835 return nonseekable_open(inode, file);
1838 static int debugfs_timings_release(struct inode *inode, struct file *file)
1840 struct debugfs_timings_state *p = file->private_data;
1842 kvm_put_kvm(p->vcpu->kvm);
1847 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1848 size_t len, loff_t *ppos)
1850 struct debugfs_timings_state *p = file->private_data;
1851 struct kvm_vcpu *vcpu = p->vcpu;
1853 struct kvmhv_tb_accumulator tb;
1862 buf_end = s + sizeof(p->buf);
1863 for (i = 0; i < N_TIMINGS; ++i) {
1864 struct kvmhv_tb_accumulator *acc;
1866 acc = (struct kvmhv_tb_accumulator *)
1867 ((unsigned long)vcpu + timings[i].offset);
1869 for (loops = 0; loops < 1000; ++loops) {
1870 count = acc->seqcount;
1875 if (count == acc->seqcount) {
1883 snprintf(s, buf_end - s, "%s: stuck\n",
1886 snprintf(s, buf_end - s,
1887 "%s: %llu %llu %llu %llu\n",
1888 timings[i].name, count / 2,
1889 tb_to_ns(tb.tb_total),
1890 tb_to_ns(tb.tb_min),
1891 tb_to_ns(tb.tb_max));
1894 p->buflen = s - p->buf;
1898 if (pos >= p->buflen)
1900 if (len > p->buflen - pos)
1901 len = p->buflen - pos;
1902 n = copy_to_user(buf, p->buf + pos, len);
1912 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1913 size_t len, loff_t *ppos)
1918 static const struct file_operations debugfs_timings_ops = {
1919 .owner = THIS_MODULE,
1920 .open = debugfs_timings_open,
1921 .release = debugfs_timings_release,
1922 .read = debugfs_timings_read,
1923 .write = debugfs_timings_write,
1924 .llseek = generic_file_llseek,
1927 /* Create a debugfs directory for the vcpu */
1928 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1931 struct kvm *kvm = vcpu->kvm;
1933 snprintf(buf, sizeof(buf), "vcpu%u", id);
1934 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1936 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1937 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1939 vcpu->arch.debugfs_timings =
1940 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1941 vcpu, &debugfs_timings_ops);
1944 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1945 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1948 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1950 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1953 struct kvm_vcpu *vcpu;
1956 struct kvmppc_vcore *vcore;
1959 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1963 err = kvm_vcpu_init(vcpu, kvm, id);
1967 vcpu->arch.shared = &vcpu->arch.shregs;
1968 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1970 * The shared struct is never shared on HV,
1971 * so we can always use host endianness
1973 #ifdef __BIG_ENDIAN__
1974 vcpu->arch.shared_big_endian = true;
1976 vcpu->arch.shared_big_endian = false;
1979 vcpu->arch.mmcr[0] = MMCR0_FC;
1980 vcpu->arch.ctrl = CTRL_RUNLATCH;
1981 /* default to host PVR, since we can't spoof it */
1982 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1983 spin_lock_init(&vcpu->arch.vpa_update_lock);
1984 spin_lock_init(&vcpu->arch.tbacct_lock);
1985 vcpu->arch.busy_preempt = TB_NIL;
1986 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1989 * Set the default HFSCR for the guest from the host value.
1990 * This value is only used on POWER9.
1991 * On POWER9 DD1, TM doesn't work, so we make sure to
1992 * prevent the guest from using it.
1993 * On POWER9, we want to virtualize the doorbell facility, so we
1994 * turn off the HFSCR bit, which causes those instructions to trap.
1996 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1997 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
1998 vcpu->arch.hfscr |= HFSCR_TM;
1999 else if (!cpu_has_feature(CPU_FTR_TM_COMP))
2000 vcpu->arch.hfscr &= ~HFSCR_TM;
2001 if (cpu_has_feature(CPU_FTR_ARCH_300))
2002 vcpu->arch.hfscr &= ~HFSCR_MSGP;
2004 kvmppc_mmu_book3s_hv_init(vcpu);
2006 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2008 init_waitqueue_head(&vcpu->arch.cpu_run);
2010 mutex_lock(&kvm->lock);
2013 core = id / kvm->arch.smt_mode;
2014 if (core < KVM_MAX_VCORES) {
2015 vcore = kvm->arch.vcores[core];
2018 vcore = kvmppc_vcore_create(kvm, core);
2019 kvm->arch.vcores[core] = vcore;
2020 kvm->arch.online_vcores++;
2023 mutex_unlock(&kvm->lock);
2028 spin_lock(&vcore->lock);
2029 ++vcore->num_threads;
2030 spin_unlock(&vcore->lock);
2031 vcpu->arch.vcore = vcore;
2032 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2033 vcpu->arch.thread_cpu = -1;
2034 vcpu->arch.prev_cpu = -1;
2036 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2037 kvmppc_sanity_check(vcpu);
2039 debugfs_vcpu_init(vcpu, id);
2044 kmem_cache_free(kvm_vcpu_cache, vcpu);
2046 return ERR_PTR(err);
2049 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2050 unsigned long flags)
2057 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2059 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2061 * On POWER8 (or POWER7), the threading mode is "strict",
2062 * so we pack smt_mode vcpus per vcore.
2064 if (smt_mode > threads_per_subcore)
2068 * On POWER9, the threading mode is "loose",
2069 * so each vcpu gets its own vcore.
2074 mutex_lock(&kvm->lock);
2076 if (!kvm->arch.online_vcores) {
2077 kvm->arch.smt_mode = smt_mode;
2078 kvm->arch.emul_smt_mode = esmt;
2081 mutex_unlock(&kvm->lock);
2086 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2088 if (vpa->pinned_addr)
2089 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2093 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2095 spin_lock(&vcpu->arch.vpa_update_lock);
2096 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2097 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2098 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2099 spin_unlock(&vcpu->arch.vpa_update_lock);
2100 kvm_vcpu_uninit(vcpu);
2101 kmem_cache_free(kvm_vcpu_cache, vcpu);
2104 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2106 /* Indicate we want to get back into the guest */
2110 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2112 unsigned long dec_nsec, now;
2115 if (now > vcpu->arch.dec_expires) {
2116 /* decrementer has already gone negative */
2117 kvmppc_core_queue_dec(vcpu);
2118 kvmppc_core_prepare_to_enter(vcpu);
2121 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2123 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2124 vcpu->arch.timer_running = 1;
2127 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2129 vcpu->arch.ceded = 0;
2130 if (vcpu->arch.timer_running) {
2131 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2132 vcpu->arch.timer_running = 0;
2136 extern int __kvmppc_vcore_entry(void);
2138 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2139 struct kvm_vcpu *vcpu)
2143 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2145 spin_lock_irq(&vcpu->arch.tbacct_lock);
2147 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2148 vcpu->arch.stolen_logged;
2149 vcpu->arch.busy_preempt = now;
2150 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2151 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2153 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2156 static int kvmppc_grab_hwthread(int cpu)
2158 struct paca_struct *tpaca;
2159 long timeout = 10000;
2161 tpaca = paca_ptrs[cpu];
2163 /* Ensure the thread won't go into the kernel if it wakes */
2164 tpaca->kvm_hstate.kvm_vcpu = NULL;
2165 tpaca->kvm_hstate.kvm_vcore = NULL;
2166 tpaca->kvm_hstate.napping = 0;
2168 tpaca->kvm_hstate.hwthread_req = 1;
2171 * If the thread is already executing in the kernel (e.g. handling
2172 * a stray interrupt), wait for it to get back to nap mode.
2173 * The smp_mb() is to ensure that our setting of hwthread_req
2174 * is visible before we look at hwthread_state, so if this
2175 * races with the code at system_reset_pSeries and the thread
2176 * misses our setting of hwthread_req, we are sure to see its
2177 * setting of hwthread_state, and vice versa.
2180 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2181 if (--timeout <= 0) {
2182 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2190 static void kvmppc_release_hwthread(int cpu)
2192 struct paca_struct *tpaca;
2194 tpaca = paca_ptrs[cpu];
2195 tpaca->kvm_hstate.hwthread_req = 0;
2196 tpaca->kvm_hstate.kvm_vcpu = NULL;
2197 tpaca->kvm_hstate.kvm_vcore = NULL;
2198 tpaca->kvm_hstate.kvm_split_mode = NULL;
2201 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2205 cpu = cpu_first_thread_sibling(cpu);
2206 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2208 * Make sure setting of bit in need_tlb_flush precedes
2209 * testing of cpu_in_guest bits. The matching barrier on
2210 * the other side is the first smp_mb() in kvmppc_run_core().
2213 for (i = 0; i < threads_per_core; ++i)
2214 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2215 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2218 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2220 struct kvm *kvm = vcpu->kvm;
2223 * With radix, the guest can do TLB invalidations itself,
2224 * and it could choose to use the local form (tlbiel) if
2225 * it is invalidating a translation that has only ever been
2226 * used on one vcpu. However, that doesn't mean it has
2227 * only ever been used on one physical cpu, since vcpus
2228 * can move around between pcpus. To cope with this, when
2229 * a vcpu moves from one pcpu to another, we need to tell
2230 * any vcpus running on the same core as this vcpu previously
2231 * ran to flush the TLB. The TLB is shared between threads,
2232 * so we use a single bit in .need_tlb_flush for all 4 threads.
2234 if (vcpu->arch.prev_cpu != pcpu) {
2235 if (vcpu->arch.prev_cpu >= 0 &&
2236 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2237 cpu_first_thread_sibling(pcpu))
2238 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2239 vcpu->arch.prev_cpu = pcpu;
2243 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2246 struct paca_struct *tpaca;
2247 struct kvm *kvm = vc->kvm;
2251 if (vcpu->arch.timer_running) {
2252 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2253 vcpu->arch.timer_running = 0;
2255 cpu += vcpu->arch.ptid;
2256 vcpu->cpu = vc->pcpu;
2257 vcpu->arch.thread_cpu = cpu;
2258 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2260 tpaca = paca_ptrs[cpu];
2261 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2262 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2263 tpaca->kvm_hstate.fake_suspend = 0;
2264 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2266 tpaca->kvm_hstate.kvm_vcore = vc;
2267 if (cpu != smp_processor_id())
2268 kvmppc_ipi_thread(cpu);
2271 static void kvmppc_wait_for_nap(int n_threads)
2273 int cpu = smp_processor_id();
2278 for (loops = 0; loops < 1000000; ++loops) {
2280 * Check if all threads are finished.
2281 * We set the vcore pointer when starting a thread
2282 * and the thread clears it when finished, so we look
2283 * for any threads that still have a non-NULL vcore ptr.
2285 for (i = 1; i < n_threads; ++i)
2286 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2288 if (i == n_threads) {
2295 for (i = 1; i < n_threads; ++i)
2296 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2297 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2301 * Check that we are on thread 0 and that any other threads in
2302 * this core are off-line. Then grab the threads so they can't
2305 static int on_primary_thread(void)
2307 int cpu = smp_processor_id();
2310 /* Are we on a primary subcore? */
2311 if (cpu_thread_in_subcore(cpu))
2315 while (++thr < threads_per_subcore)
2316 if (cpu_online(cpu + thr))
2319 /* Grab all hw threads so they can't go into the kernel */
2320 for (thr = 1; thr < threads_per_subcore; ++thr) {
2321 if (kvmppc_grab_hwthread(cpu + thr)) {
2322 /* Couldn't grab one; let the others go */
2324 kvmppc_release_hwthread(cpu + thr);
2325 } while (--thr > 0);
2333 * A list of virtual cores for each physical CPU.
2334 * These are vcores that could run but their runner VCPU tasks are
2335 * (or may be) preempted.
2337 struct preempted_vcore_list {
2338 struct list_head list;
2342 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2344 static void init_vcore_lists(void)
2348 for_each_possible_cpu(cpu) {
2349 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2350 spin_lock_init(&lp->lock);
2351 INIT_LIST_HEAD(&lp->list);
2355 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2357 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2359 vc->vcore_state = VCORE_PREEMPT;
2360 vc->pcpu = smp_processor_id();
2361 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2362 spin_lock(&lp->lock);
2363 list_add_tail(&vc->preempt_list, &lp->list);
2364 spin_unlock(&lp->lock);
2367 /* Start accumulating stolen time */
2368 kvmppc_core_start_stolen(vc);
2371 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2373 struct preempted_vcore_list *lp;
2375 kvmppc_core_end_stolen(vc);
2376 if (!list_empty(&vc->preempt_list)) {
2377 lp = &per_cpu(preempted_vcores, vc->pcpu);
2378 spin_lock(&lp->lock);
2379 list_del_init(&vc->preempt_list);
2380 spin_unlock(&lp->lock);
2382 vc->vcore_state = VCORE_INACTIVE;
2386 * This stores information about the virtual cores currently
2387 * assigned to a physical core.
2391 int max_subcore_threads;
2393 int subcore_threads[MAX_SUBCORES];
2394 struct kvmppc_vcore *vc[MAX_SUBCORES];
2398 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2399 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2401 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2403 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2405 memset(cip, 0, sizeof(*cip));
2406 cip->n_subcores = 1;
2407 cip->max_subcore_threads = vc->num_threads;
2408 cip->total_threads = vc->num_threads;
2409 cip->subcore_threads[0] = vc->num_threads;
2413 static bool subcore_config_ok(int n_subcores, int n_threads)
2416 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2417 * split-core mode, with one thread per subcore.
2419 if (cpu_has_feature(CPU_FTR_ARCH_300))
2420 return n_subcores <= 4 && n_threads == 1;
2422 /* On POWER8, can only dynamically split if unsplit to begin with */
2423 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2425 if (n_subcores > MAX_SUBCORES)
2427 if (n_subcores > 1) {
2428 if (!(dynamic_mt_modes & 2))
2430 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2434 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2437 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2439 vc->entry_exit_map = 0;
2441 vc->napping_threads = 0;
2442 vc->conferring_threads = 0;
2445 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2447 int n_threads = vc->num_threads;
2450 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2453 /* Some POWER9 chips require all threads to be in the same MMU mode */
2454 if (no_mixing_hpt_and_radix &&
2455 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2458 if (n_threads < cip->max_subcore_threads)
2459 n_threads = cip->max_subcore_threads;
2460 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2462 cip->max_subcore_threads = n_threads;
2464 sub = cip->n_subcores;
2466 cip->total_threads += vc->num_threads;
2467 cip->subcore_threads[sub] = vc->num_threads;
2469 init_vcore_to_run(vc);
2470 list_del_init(&vc->preempt_list);
2476 * Work out whether it is possible to piggyback the execution of
2477 * vcore *pvc onto the execution of the other vcores described in *cip.
2479 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2482 if (cip->total_threads + pvc->num_threads > target_threads)
2485 return can_dynamic_split(pvc, cip);
2488 static void prepare_threads(struct kvmppc_vcore *vc)
2491 struct kvm_vcpu *vcpu;
2493 for_each_runnable_thread(i, vcpu, vc) {
2494 if (signal_pending(vcpu->arch.run_task))
2495 vcpu->arch.ret = -EINTR;
2496 else if (vcpu->arch.vpa.update_pending ||
2497 vcpu->arch.slb_shadow.update_pending ||
2498 vcpu->arch.dtl.update_pending)
2499 vcpu->arch.ret = RESUME_GUEST;
2502 kvmppc_remove_runnable(vc, vcpu);
2503 wake_up(&vcpu->arch.cpu_run);
2507 static void collect_piggybacks(struct core_info *cip, int target_threads)
2509 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2510 struct kvmppc_vcore *pvc, *vcnext;
2512 spin_lock(&lp->lock);
2513 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2514 if (!spin_trylock(&pvc->lock))
2516 prepare_threads(pvc);
2517 if (!pvc->n_runnable) {
2518 list_del_init(&pvc->preempt_list);
2519 if (pvc->runner == NULL) {
2520 pvc->vcore_state = VCORE_INACTIVE;
2521 kvmppc_core_end_stolen(pvc);
2523 spin_unlock(&pvc->lock);
2526 if (!can_piggyback(pvc, cip, target_threads)) {
2527 spin_unlock(&pvc->lock);
2530 kvmppc_core_end_stolen(pvc);
2531 pvc->vcore_state = VCORE_PIGGYBACK;
2532 if (cip->total_threads >= target_threads)
2535 spin_unlock(&lp->lock);
2538 static bool recheck_signals(struct core_info *cip)
2541 struct kvm_vcpu *vcpu;
2543 for (sub = 0; sub < cip->n_subcores; ++sub)
2544 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2545 if (signal_pending(vcpu->arch.run_task))
2550 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2552 int still_running = 0, i;
2555 struct kvm_vcpu *vcpu;
2557 spin_lock(&vc->lock);
2559 for_each_runnable_thread(i, vcpu, vc) {
2560 /* cancel pending dec exception if dec is positive */
2561 if (now < vcpu->arch.dec_expires &&
2562 kvmppc_core_pending_dec(vcpu))
2563 kvmppc_core_dequeue_dec(vcpu);
2565 trace_kvm_guest_exit(vcpu);
2568 if (vcpu->arch.trap)
2569 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2570 vcpu->arch.run_task);
2572 vcpu->arch.ret = ret;
2573 vcpu->arch.trap = 0;
2575 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2576 if (vcpu->arch.pending_exceptions)
2577 kvmppc_core_prepare_to_enter(vcpu);
2578 if (vcpu->arch.ceded)
2579 kvmppc_set_timer(vcpu);
2583 kvmppc_remove_runnable(vc, vcpu);
2584 wake_up(&vcpu->arch.cpu_run);
2588 if (still_running > 0) {
2589 kvmppc_vcore_preempt(vc);
2590 } else if (vc->runner) {
2591 vc->vcore_state = VCORE_PREEMPT;
2592 kvmppc_core_start_stolen(vc);
2594 vc->vcore_state = VCORE_INACTIVE;
2596 if (vc->n_runnable > 0 && vc->runner == NULL) {
2597 /* make sure there's a candidate runner awake */
2599 vcpu = next_runnable_thread(vc, &i);
2600 wake_up(&vcpu->arch.cpu_run);
2603 spin_unlock(&vc->lock);
2607 * Clear core from the list of active host cores as we are about to
2608 * enter the guest. Only do this if it is the primary thread of the
2609 * core (not if a subcore) that is entering the guest.
2611 static inline int kvmppc_clear_host_core(unsigned int cpu)
2615 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2618 * Memory barrier can be omitted here as we will do a smp_wmb()
2619 * later in kvmppc_start_thread and we need ensure that state is
2620 * visible to other CPUs only after we enter guest.
2622 core = cpu >> threads_shift;
2623 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2628 * Advertise this core as an active host core since we exited the guest
2629 * Only need to do this if it is the primary thread of the core that is
2632 static inline int kvmppc_set_host_core(unsigned int cpu)
2636 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2640 * Memory barrier can be omitted here because we do a spin_unlock
2641 * immediately after this which provides the memory barrier.
2643 core = cpu >> threads_shift;
2644 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2648 static void set_irq_happened(int trap)
2651 case BOOK3S_INTERRUPT_EXTERNAL:
2652 local_paca->irq_happened |= PACA_IRQ_EE;
2654 case BOOK3S_INTERRUPT_H_DOORBELL:
2655 local_paca->irq_happened |= PACA_IRQ_DBELL;
2657 case BOOK3S_INTERRUPT_HMI:
2658 local_paca->irq_happened |= PACA_IRQ_HMI;
2660 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2661 replay_system_reset();
2667 * Run a set of guest threads on a physical core.
2668 * Called with vc->lock held.
2670 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2672 struct kvm_vcpu *vcpu;
2675 struct core_info core_info;
2676 struct kvmppc_vcore *pvc;
2677 struct kvm_split_mode split_info, *sip;
2678 int split, subcore_size, active;
2681 unsigned long cmd_bit, stat_bit;
2684 int controlled_threads;
2690 * Remove from the list any threads that have a signal pending
2691 * or need a VPA update done
2693 prepare_threads(vc);
2695 /* if the runner is no longer runnable, let the caller pick a new one */
2696 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2702 init_vcore_to_run(vc);
2703 vc->preempt_tb = TB_NIL;
2706 * Number of threads that we will be controlling: the same as
2707 * the number of threads per subcore, except on POWER9,
2708 * where it's 1 because the threads are (mostly) independent.
2710 controlled_threads = threads_per_vcore(vc->kvm);
2713 * Make sure we are running on primary threads, and that secondary
2714 * threads are offline. Also check if the number of threads in this
2715 * guest are greater than the current system threads per guest.
2716 * On POWER9, we need to be not in independent-threads mode if
2717 * this is a HPT guest on a radix host machine where the
2718 * CPU threads may not be in different MMU modes.
2720 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
2721 !kvm_is_radix(vc->kvm);
2722 if (((controlled_threads > 1) &&
2723 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2724 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2725 for_each_runnable_thread(i, vcpu, vc) {
2726 vcpu->arch.ret = -EBUSY;
2727 kvmppc_remove_runnable(vc, vcpu);
2728 wake_up(&vcpu->arch.cpu_run);
2734 * See if we could run any other vcores on the physical core
2735 * along with this one.
2737 init_core_info(&core_info, vc);
2738 pcpu = smp_processor_id();
2739 target_threads = controlled_threads;
2740 if (target_smt_mode && target_smt_mode < target_threads)
2741 target_threads = target_smt_mode;
2742 if (vc->num_threads < target_threads)
2743 collect_piggybacks(&core_info, target_threads);
2746 * On radix, arrange for TLB flushing if necessary.
2747 * This has to be done before disabling interrupts since
2748 * it uses smp_call_function().
2750 pcpu = smp_processor_id();
2751 if (kvm_is_radix(vc->kvm)) {
2752 for (sub = 0; sub < core_info.n_subcores; ++sub)
2753 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2754 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2758 * Hard-disable interrupts, and check resched flag and signals.
2759 * If we need to reschedule or deliver a signal, clean up
2760 * and return without going into the guest(s).
2761 * If the mmu_ready flag has been cleared, don't go into the
2762 * guest because that means a HPT resize operation is in progress.
2764 local_irq_disable();
2766 if (lazy_irq_pending() || need_resched() ||
2767 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
2769 vc->vcore_state = VCORE_INACTIVE;
2770 /* Unlock all except the primary vcore */
2771 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2772 pvc = core_info.vc[sub];
2773 /* Put back on to the preempted vcores list */
2774 kvmppc_vcore_preempt(pvc);
2775 spin_unlock(&pvc->lock);
2777 for (i = 0; i < controlled_threads; ++i)
2778 kvmppc_release_hwthread(pcpu + i);
2782 kvmppc_clear_host_core(pcpu);
2784 /* Decide on micro-threading (split-core) mode */
2785 subcore_size = threads_per_subcore;
2786 cmd_bit = stat_bit = 0;
2787 split = core_info.n_subcores;
2789 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2790 && !cpu_has_feature(CPU_FTR_ARCH_300);
2792 if (split > 1 || hpt_on_radix) {
2794 memset(&split_info, 0, sizeof(split_info));
2795 for (sub = 0; sub < core_info.n_subcores; ++sub)
2796 split_info.vc[sub] = core_info.vc[sub];
2799 if (split == 2 && (dynamic_mt_modes & 2)) {
2800 cmd_bit = HID0_POWER8_1TO2LPAR;
2801 stat_bit = HID0_POWER8_2LPARMODE;
2804 cmd_bit = HID0_POWER8_1TO4LPAR;
2805 stat_bit = HID0_POWER8_4LPARMODE;
2807 subcore_size = MAX_SMT_THREADS / split;
2808 split_info.rpr = mfspr(SPRN_RPR);
2809 split_info.pmmar = mfspr(SPRN_PMMAR);
2810 split_info.ldbar = mfspr(SPRN_LDBAR);
2811 split_info.subcore_size = subcore_size;
2813 split_info.subcore_size = 1;
2815 /* Use the split_info for LPCR/LPIDR changes */
2816 split_info.lpcr_req = vc->lpcr;
2817 split_info.lpidr_req = vc->kvm->arch.lpid;
2818 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2819 split_info.do_set = 1;
2823 /* order writes to split_info before kvm_split_mode pointer */
2827 for (thr = 0; thr < controlled_threads; ++thr) {
2828 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2830 paca->kvm_hstate.tid = thr;
2831 paca->kvm_hstate.napping = 0;
2832 paca->kvm_hstate.kvm_split_mode = sip;
2835 /* Initiate micro-threading (split-core) on POWER8 if required */
2837 unsigned long hid0 = mfspr(SPRN_HID0);
2839 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2841 mtspr(SPRN_HID0, hid0);
2844 hid0 = mfspr(SPRN_HID0);
2845 if (hid0 & stat_bit)
2851 /* Start all the threads */
2853 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2854 thr = is_power8 ? subcore_thread_map[sub] : sub;
2857 pvc = core_info.vc[sub];
2858 pvc->pcpu = pcpu + thr;
2859 for_each_runnable_thread(i, vcpu, pvc) {
2860 kvmppc_start_thread(vcpu, pvc);
2861 kvmppc_create_dtl_entry(vcpu, pvc);
2862 trace_kvm_guest_enter(vcpu);
2863 if (!vcpu->arch.ptid)
2865 active |= 1 << (thr + vcpu->arch.ptid);
2868 * We need to start the first thread of each subcore
2869 * even if it doesn't have a vcpu.
2872 kvmppc_start_thread(NULL, pvc);
2876 * Ensure that split_info.do_nap is set after setting
2877 * the vcore pointer in the PACA of the secondaries.
2882 * When doing micro-threading, poke the inactive threads as well.
2883 * This gets them to the nap instruction after kvm_do_nap,
2884 * which reduces the time taken to unsplit later.
2885 * For POWER9 HPT guest on radix host, we need all the secondary
2886 * threads woken up so they can do the LPCR/LPIDR change.
2888 if (cmd_bit || hpt_on_radix) {
2889 split_info.do_nap = 1; /* ask secondaries to nap when done */
2890 for (thr = 1; thr < threads_per_subcore; ++thr)
2891 if (!(active & (1 << thr)))
2892 kvmppc_ipi_thread(pcpu + thr);
2895 vc->vcore_state = VCORE_RUNNING;
2898 trace_kvmppc_run_core(vc, 0);
2900 for (sub = 0; sub < core_info.n_subcores; ++sub)
2901 spin_unlock(&core_info.vc[sub]->lock);
2904 * Interrupts will be enabled once we get into the guest,
2905 * so tell lockdep that we're about to enable interrupts.
2907 trace_hardirqs_on();
2911 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2913 trap = __kvmppc_vcore_entry();
2915 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2919 trace_hardirqs_off();
2920 set_irq_happened(trap);
2922 spin_lock(&vc->lock);
2923 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2924 vc->vcore_state = VCORE_EXITING;
2926 /* wait for secondary threads to finish writing their state to memory */
2927 kvmppc_wait_for_nap(controlled_threads);
2929 /* Return to whole-core mode if we split the core earlier */
2931 unsigned long hid0 = mfspr(SPRN_HID0);
2932 unsigned long loops = 0;
2934 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2935 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2937 mtspr(SPRN_HID0, hid0);
2940 hid0 = mfspr(SPRN_HID0);
2941 if (!(hid0 & stat_bit))
2946 } else if (hpt_on_radix) {
2947 /* Wait for all threads to have seen final sync */
2948 for (thr = 1; thr < controlled_threads; ++thr) {
2949 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2951 while (paca->kvm_hstate.kvm_split_mode) {
2958 split_info.do_nap = 0;
2960 kvmppc_set_host_core(pcpu);
2964 /* Let secondaries go back to the offline loop */
2965 for (i = 0; i < controlled_threads; ++i) {
2966 kvmppc_release_hwthread(pcpu + i);
2967 if (sip && sip->napped[i])
2968 kvmppc_ipi_thread(pcpu + i);
2969 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2972 spin_unlock(&vc->lock);
2974 /* make sure updates to secondary vcpu structs are visible now */
2979 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2980 pvc = core_info.vc[sub];
2981 post_guest_process(pvc, pvc == vc);
2984 spin_lock(&vc->lock);
2987 vc->vcore_state = VCORE_INACTIVE;
2988 trace_kvmppc_run_core(vc, 1);
2992 * Wait for some other vcpu thread to execute us, and
2993 * wake us up when we need to handle something in the host.
2995 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2996 struct kvm_vcpu *vcpu, int wait_state)
3000 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3001 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3002 spin_unlock(&vc->lock);
3004 spin_lock(&vc->lock);
3006 finish_wait(&vcpu->arch.cpu_run, &wait);
3009 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3012 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3013 vc->halt_poll_ns = 10000;
3015 vc->halt_poll_ns *= halt_poll_ns_grow;
3018 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3020 if (halt_poll_ns_shrink == 0)
3021 vc->halt_poll_ns = 0;
3023 vc->halt_poll_ns /= halt_poll_ns_shrink;
3026 #ifdef CONFIG_KVM_XICS
3027 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3029 if (!xive_enabled())
3031 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3032 vcpu->arch.xive_saved_state.cppr;
3035 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3039 #endif /* CONFIG_KVM_XICS */
3041 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3043 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3044 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3051 * Check to see if any of the runnable vcpus on the vcore have pending
3052 * exceptions or are no longer ceded
3054 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3056 struct kvm_vcpu *vcpu;
3059 for_each_runnable_thread(i, vcpu, vc) {
3060 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3068 * All the vcpus in this vcore are idle, so wait for a decrementer
3069 * or external interrupt to one of the vcpus. vc->lock is held.
3071 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3073 ktime_t cur, start_poll, start_wait;
3076 DECLARE_SWAITQUEUE(wait);
3078 /* Poll for pending exceptions and ceded state */
3079 cur = start_poll = ktime_get();
3080 if (vc->halt_poll_ns) {
3081 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3082 ++vc->runner->stat.halt_attempted_poll;
3084 vc->vcore_state = VCORE_POLLING;
3085 spin_unlock(&vc->lock);
3088 if (kvmppc_vcore_check_block(vc)) {
3093 } while (single_task_running() && ktime_before(cur, stop));
3095 spin_lock(&vc->lock);
3096 vc->vcore_state = VCORE_INACTIVE;
3099 ++vc->runner->stat.halt_successful_poll;
3104 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3106 if (kvmppc_vcore_check_block(vc)) {
3107 finish_swait(&vc->wq, &wait);
3109 /* If we polled, count this as a successful poll */
3110 if (vc->halt_poll_ns)
3111 ++vc->runner->stat.halt_successful_poll;
3115 start_wait = ktime_get();
3117 vc->vcore_state = VCORE_SLEEPING;
3118 trace_kvmppc_vcore_blocked(vc, 0);
3119 spin_unlock(&vc->lock);
3121 finish_swait(&vc->wq, &wait);
3122 spin_lock(&vc->lock);
3123 vc->vcore_state = VCORE_INACTIVE;
3124 trace_kvmppc_vcore_blocked(vc, 1);
3125 ++vc->runner->stat.halt_successful_wait;
3130 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3132 /* Attribute wait time */
3134 vc->runner->stat.halt_wait_ns +=
3135 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3136 /* Attribute failed poll time */
3137 if (vc->halt_poll_ns)
3138 vc->runner->stat.halt_poll_fail_ns +=
3139 ktime_to_ns(start_wait) -
3140 ktime_to_ns(start_poll);
3142 /* Attribute successful poll time */
3143 if (vc->halt_poll_ns)
3144 vc->runner->stat.halt_poll_success_ns +=
3146 ktime_to_ns(start_poll);
3149 /* Adjust poll time */
3151 if (block_ns <= vc->halt_poll_ns)
3153 /* We slept and blocked for longer than the max halt time */
3154 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3155 shrink_halt_poll_ns(vc);
3156 /* We slept and our poll time is too small */
3157 else if (vc->halt_poll_ns < halt_poll_ns &&
3158 block_ns < halt_poll_ns)
3159 grow_halt_poll_ns(vc);
3160 if (vc->halt_poll_ns > halt_poll_ns)
3161 vc->halt_poll_ns = halt_poll_ns;
3163 vc->halt_poll_ns = 0;
3165 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3168 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3171 struct kvm *kvm = vcpu->kvm;
3173 mutex_lock(&kvm->lock);
3174 if (!kvm->arch.mmu_ready) {
3175 if (!kvm_is_radix(kvm))
3176 r = kvmppc_hv_setup_htab_rma(vcpu);
3178 if (cpu_has_feature(CPU_FTR_ARCH_300))
3179 kvmppc_setup_partition_table(kvm);
3180 kvm->arch.mmu_ready = 1;
3183 mutex_unlock(&kvm->lock);
3187 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3190 struct kvmppc_vcore *vc;
3193 trace_kvmppc_run_vcpu_enter(vcpu);
3195 kvm_run->exit_reason = 0;
3196 vcpu->arch.ret = RESUME_GUEST;
3197 vcpu->arch.trap = 0;
3198 kvmppc_update_vpas(vcpu);
3201 * Synchronize with other threads in this virtual core
3203 vc = vcpu->arch.vcore;
3204 spin_lock(&vc->lock);
3205 vcpu->arch.ceded = 0;
3206 vcpu->arch.run_task = current;
3207 vcpu->arch.kvm_run = kvm_run;
3208 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3209 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3210 vcpu->arch.busy_preempt = TB_NIL;
3211 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3215 * This happens the first time this is called for a vcpu.
3216 * If the vcore is already running, we may be able to start
3217 * this thread straight away and have it join in.
3219 if (!signal_pending(current)) {
3220 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3221 vc->vcore_state == VCORE_RUNNING) &&
3222 !VCORE_IS_EXITING(vc)) {
3223 kvmppc_create_dtl_entry(vcpu, vc);
3224 kvmppc_start_thread(vcpu, vc);
3225 trace_kvm_guest_enter(vcpu);
3226 } else if (vc->vcore_state == VCORE_SLEEPING) {
3232 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3233 !signal_pending(current)) {
3234 /* See if the MMU is ready to go */
3235 if (!vcpu->kvm->arch.mmu_ready) {
3236 spin_unlock(&vc->lock);
3237 r = kvmhv_setup_mmu(vcpu);
3238 spin_lock(&vc->lock);
3240 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3241 kvm_run->fail_entry.
3242 hardware_entry_failure_reason = 0;
3248 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3249 kvmppc_vcore_end_preempt(vc);
3251 if (vc->vcore_state != VCORE_INACTIVE) {
3252 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3255 for_each_runnable_thread(i, v, vc) {
3256 kvmppc_core_prepare_to_enter(v);
3257 if (signal_pending(v->arch.run_task)) {
3258 kvmppc_remove_runnable(vc, v);
3259 v->stat.signal_exits++;
3260 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3261 v->arch.ret = -EINTR;
3262 wake_up(&v->arch.cpu_run);
3265 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3268 for_each_runnable_thread(i, v, vc) {
3269 if (!kvmppc_vcpu_woken(v))
3270 n_ceded += v->arch.ceded;
3275 if (n_ceded == vc->n_runnable) {
3276 kvmppc_vcore_blocked(vc);
3277 } else if (need_resched()) {
3278 kvmppc_vcore_preempt(vc);
3279 /* Let something else run */
3280 cond_resched_lock(&vc->lock);
3281 if (vc->vcore_state == VCORE_PREEMPT)
3282 kvmppc_vcore_end_preempt(vc);
3284 kvmppc_run_core(vc);
3289 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3290 (vc->vcore_state == VCORE_RUNNING ||
3291 vc->vcore_state == VCORE_EXITING ||
3292 vc->vcore_state == VCORE_PIGGYBACK))
3293 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3295 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3296 kvmppc_vcore_end_preempt(vc);
3298 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3299 kvmppc_remove_runnable(vc, vcpu);
3300 vcpu->stat.signal_exits++;
3301 kvm_run->exit_reason = KVM_EXIT_INTR;
3302 vcpu->arch.ret = -EINTR;
3305 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3306 /* Wake up some vcpu to run the core */
3308 v = next_runnable_thread(vc, &i);
3309 wake_up(&v->arch.cpu_run);
3312 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3313 spin_unlock(&vc->lock);
3314 return vcpu->arch.ret;
3317 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3321 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3322 unsigned long user_tar = 0;
3323 unsigned int user_vrsave;
3326 if (!vcpu->arch.sane) {
3327 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3332 * Don't allow entry with a suspended transaction, because
3333 * the guest entry/exit code will lose it.
3334 * If the guest has TM enabled, save away their TM-related SPRs
3335 * (they will get restored by the TM unavailable interrupt).
3337 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3338 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3339 (current->thread.regs->msr & MSR_TM)) {
3340 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3341 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3342 run->fail_entry.hardware_entry_failure_reason = 0;
3345 /* Enable TM so we can read the TM SPRs */
3346 mtmsr(mfmsr() | MSR_TM);
3347 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3348 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3349 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3350 current->thread.regs->msr &= ~MSR_TM;
3354 kvmppc_core_prepare_to_enter(vcpu);
3356 /* No need to go into the guest when all we'll do is come back out */
3357 if (signal_pending(current)) {
3358 run->exit_reason = KVM_EXIT_INTR;
3363 atomic_inc(&kvm->arch.vcpus_running);
3364 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3367 flush_all_to_thread(current);
3369 /* Save userspace EBB and other register values */
3370 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3371 ebb_regs[0] = mfspr(SPRN_EBBHR);
3372 ebb_regs[1] = mfspr(SPRN_EBBRR);
3373 ebb_regs[2] = mfspr(SPRN_BESCR);
3374 user_tar = mfspr(SPRN_TAR);
3376 user_vrsave = mfspr(SPRN_VRSAVE);
3378 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3379 vcpu->arch.pgdir = current->mm->pgd;
3380 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3383 r = kvmppc_run_vcpu(run, vcpu);
3385 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3386 !(vcpu->arch.shregs.msr & MSR_PR)) {
3387 trace_kvm_hcall_enter(vcpu);
3388 r = kvmppc_pseries_do_hcall(vcpu);
3389 trace_kvm_hcall_exit(vcpu, r);
3390 kvmppc_core_prepare_to_enter(vcpu);
3391 } else if (r == RESUME_PAGE_FAULT) {
3392 srcu_idx = srcu_read_lock(&kvm->srcu);
3393 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3394 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3395 srcu_read_unlock(&kvm->srcu, srcu_idx);
3396 } else if (r == RESUME_PASSTHROUGH) {
3397 if (WARN_ON(xive_enabled()))
3400 r = kvmppc_xics_rm_complete(vcpu, 0);
3402 } while (is_kvmppc_resume_guest(r));
3404 /* Restore userspace EBB and other register values */
3405 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3406 mtspr(SPRN_EBBHR, ebb_regs[0]);
3407 mtspr(SPRN_EBBRR, ebb_regs[1]);
3408 mtspr(SPRN_BESCR, ebb_regs[2]);
3409 mtspr(SPRN_TAR, user_tar);
3410 mtspr(SPRN_FSCR, current->thread.fscr);
3412 mtspr(SPRN_VRSAVE, user_vrsave);
3414 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3415 atomic_dec(&kvm->arch.vcpus_running);
3419 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3420 int shift, int sllp)
3422 (*sps)->page_shift = shift;
3423 (*sps)->slb_enc = sllp;
3424 (*sps)->enc[0].page_shift = shift;
3425 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3427 * Add 16MB MPSS support (may get filtered out by userspace)
3430 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3432 (*sps)->enc[1].page_shift = 24;
3433 (*sps)->enc[1].pte_enc = penc;
3439 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3440 struct kvm_ppc_smmu_info *info)
3442 struct kvm_ppc_one_seg_page_size *sps;
3445 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3446 * POWER7 doesn't support keys for instruction accesses,
3447 * POWER8 and POWER9 do.
3449 info->data_keys = 32;
3450 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3452 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3453 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3454 info->slb_size = 32;
3456 /* We only support these sizes for now, and no muti-size segments */
3457 sps = &info->sps[0];
3458 kvmppc_add_seg_page_size(&sps, 12, 0);
3459 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3460 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3466 * Get (and clear) the dirty memory log for a memory slot.
3468 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3469 struct kvm_dirty_log *log)
3471 struct kvm_memslots *slots;
3472 struct kvm_memory_slot *memslot;
3475 unsigned long *buf, *p;
3476 struct kvm_vcpu *vcpu;
3478 mutex_lock(&kvm->slots_lock);
3481 if (log->slot >= KVM_USER_MEM_SLOTS)
3484 slots = kvm_memslots(kvm);
3485 memslot = id_to_memslot(slots, log->slot);
3487 if (!memslot->dirty_bitmap)
3491 * Use second half of bitmap area because both HPT and radix
3492 * accumulate bits in the first half.
3494 n = kvm_dirty_bitmap_bytes(memslot);
3495 buf = memslot->dirty_bitmap + n / sizeof(long);
3498 if (kvm_is_radix(kvm))
3499 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3501 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3506 * We accumulate dirty bits in the first half of the
3507 * memslot's dirty_bitmap area, for when pages are paged
3508 * out or modified by the host directly. Pick up these
3509 * bits and add them to the map.
3511 p = memslot->dirty_bitmap;
3512 for (i = 0; i < n / sizeof(long); ++i)
3513 buf[i] |= xchg(&p[i], 0);
3515 /* Harvest dirty bits from VPA and DTL updates */
3516 /* Note: we never modify the SLB shadow buffer areas */
3517 kvm_for_each_vcpu(i, vcpu, kvm) {
3518 spin_lock(&vcpu->arch.vpa_update_lock);
3519 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3520 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3521 spin_unlock(&vcpu->arch.vpa_update_lock);
3525 if (copy_to_user(log->dirty_bitmap, buf, n))
3530 mutex_unlock(&kvm->slots_lock);
3534 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3535 struct kvm_memory_slot *dont)
3537 if (!dont || free->arch.rmap != dont->arch.rmap) {
3538 vfree(free->arch.rmap);
3539 free->arch.rmap = NULL;
3543 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3544 unsigned long npages)
3546 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3547 if (!slot->arch.rmap)
3553 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3554 struct kvm_memory_slot *memslot,
3555 const struct kvm_userspace_memory_region *mem)
3560 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3561 const struct kvm_userspace_memory_region *mem,
3562 const struct kvm_memory_slot *old,
3563 const struct kvm_memory_slot *new)
3565 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3568 * If we are making a new memslot, it might make
3569 * some address that was previously cached as emulated
3570 * MMIO be no longer emulated MMIO, so invalidate
3571 * all the caches of emulated MMIO translations.
3574 atomic64_inc(&kvm->arch.mmio_update);
3578 * Update LPCR values in kvm->arch and in vcores.
3579 * Caller must hold kvm->lock.
3581 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3586 if ((kvm->arch.lpcr & mask) == lpcr)
3589 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3591 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3592 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3595 spin_lock(&vc->lock);
3596 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3597 spin_unlock(&vc->lock);
3598 if (++cores_done >= kvm->arch.online_vcores)
3603 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3608 void kvmppc_setup_partition_table(struct kvm *kvm)
3610 unsigned long dw0, dw1;
3612 if (!kvm_is_radix(kvm)) {
3613 /* PS field - page size for VRMA */
3614 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3615 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3616 /* HTABSIZE and HTABORG fields */
3617 dw0 |= kvm->arch.sdr1;
3619 /* Second dword as set by userspace */
3620 dw1 = kvm->arch.process_table;
3622 dw0 = PATB_HR | radix__get_tree_size() |
3623 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3624 dw1 = PATB_GR | kvm->arch.process_table;
3627 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3631 * Set up HPT (hashed page table) and RMA (real-mode area).
3632 * Must be called with kvm->lock held.
3634 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3637 struct kvm *kvm = vcpu->kvm;
3639 struct kvm_memory_slot *memslot;
3640 struct vm_area_struct *vma;
3641 unsigned long lpcr = 0, senc;
3642 unsigned long psize, porder;
3645 /* Allocate hashed page table (if not done already) and reset it */
3646 if (!kvm->arch.hpt.virt) {
3647 int order = KVM_DEFAULT_HPT_ORDER;
3648 struct kvm_hpt_info info;
3650 err = kvmppc_allocate_hpt(&info, order);
3651 /* If we get here, it means userspace didn't specify a
3652 * size explicitly. So, try successively smaller
3653 * sizes if the default failed. */
3654 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3655 err = kvmppc_allocate_hpt(&info, order);
3658 pr_err("KVM: Couldn't alloc HPT\n");
3662 kvmppc_set_hpt(kvm, &info);
3665 /* Look up the memslot for guest physical address 0 */
3666 srcu_idx = srcu_read_lock(&kvm->srcu);
3667 memslot = gfn_to_memslot(kvm, 0);
3669 /* We must have some memory at 0 by now */
3671 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3674 /* Look up the VMA for the start of this memory slot */
3675 hva = memslot->userspace_addr;
3676 down_read(¤t->mm->mmap_sem);
3677 vma = find_vma(current->mm, hva);
3678 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3681 psize = vma_kernel_pagesize(vma);
3682 porder = __ilog2(psize);
3684 up_read(¤t->mm->mmap_sem);
3686 /* We can handle 4k, 64k or 16M pages in the VRMA */
3688 if (!(psize == 0x1000 || psize == 0x10000 ||
3689 psize == 0x1000000))
3692 senc = slb_pgsize_encoding(psize);
3693 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3694 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3695 /* Create HPTEs in the hash page table for the VRMA */
3696 kvmppc_map_vrma(vcpu, memslot, porder);
3698 /* Update VRMASD field in the LPCR */
3699 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3700 /* the -4 is to account for senc values starting at 0x10 */
3701 lpcr = senc << (LPCR_VRMASD_SH - 4);
3702 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3705 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3709 srcu_read_unlock(&kvm->srcu, srcu_idx);
3714 up_read(¤t->mm->mmap_sem);
3718 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3719 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3721 kvmppc_free_radix(kvm);
3722 kvmppc_update_lpcr(kvm, LPCR_VPM1,
3723 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3724 kvmppc_rmap_reset(kvm);
3725 kvm->arch.radix = 0;
3726 kvm->arch.process_table = 0;
3730 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3731 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3735 err = kvmppc_init_vm_radix(kvm);
3739 kvmppc_free_hpt(&kvm->arch.hpt);
3740 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3741 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3742 kvm->arch.radix = 1;
3746 #ifdef CONFIG_KVM_XICS
3748 * Allocate a per-core structure for managing state about which cores are
3749 * running in the host versus the guest and for exchanging data between
3750 * real mode KVM and CPU running in the host.
3751 * This is only done for the first VM.
3752 * The allocated structure stays even if all VMs have stopped.
3753 * It is only freed when the kvm-hv module is unloaded.
3754 * It's OK for this routine to fail, we just don't support host
3755 * core operations like redirecting H_IPI wakeups.
3757 void kvmppc_alloc_host_rm_ops(void)
3759 struct kvmppc_host_rm_ops *ops;
3760 unsigned long l_ops;
3764 /* Not the first time here ? */
3765 if (kvmppc_host_rm_ops_hv != NULL)
3768 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3772 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3773 ops->rm_core = kzalloc(size, GFP_KERNEL);
3775 if (!ops->rm_core) {
3782 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3783 if (!cpu_online(cpu))
3786 core = cpu >> threads_shift;
3787 ops->rm_core[core].rm_state.in_host = 1;
3790 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3793 * Make the contents of the kvmppc_host_rm_ops structure visible
3794 * to other CPUs before we assign it to the global variable.
3795 * Do an atomic assignment (no locks used here), but if someone
3796 * beats us to it, just free our copy and return.
3799 l_ops = (unsigned long) ops;
3801 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3803 kfree(ops->rm_core);
3808 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3809 "ppc/kvm_book3s:prepare",
3810 kvmppc_set_host_core,
3811 kvmppc_clear_host_core);
3815 void kvmppc_free_host_rm_ops(void)
3817 if (kvmppc_host_rm_ops_hv) {
3818 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3819 kfree(kvmppc_host_rm_ops_hv->rm_core);
3820 kfree(kvmppc_host_rm_ops_hv);
3821 kvmppc_host_rm_ops_hv = NULL;
3826 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3828 unsigned long lpcr, lpid;
3832 /* Allocate the guest's logical partition ID */
3834 lpid = kvmppc_alloc_lpid();
3837 kvm->arch.lpid = lpid;
3839 kvmppc_alloc_host_rm_ops();
3842 * Since we don't flush the TLB when tearing down a VM,
3843 * and this lpid might have previously been used,
3844 * make sure we flush on each core before running the new VM.
3845 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3846 * does this flush for us.
3848 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3849 cpumask_setall(&kvm->arch.need_tlb_flush);
3851 /* Start out with the default set of hcalls enabled */
3852 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3853 sizeof(kvm->arch.enabled_hcalls));
3855 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3856 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3858 /* Init LPCR for virtual RMA mode */
3859 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3860 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3861 lpcr &= LPCR_PECE | LPCR_LPES;
3862 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3863 LPCR_VPM0 | LPCR_VPM1;
3864 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3865 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3866 /* On POWER8 turn on online bit to enable PURR/SPURR */
3867 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3870 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3871 * Set HVICE bit to enable hypervisor virtualization interrupts.
3872 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3873 * be unnecessary but better safe than sorry in case we re-enable
3874 * EE in HV mode with this LPCR still set)
3876 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3878 lpcr |= LPCR_HVICE | LPCR_HEIC;
3881 * If xive is enabled, we route 0x500 interrupts directly
3889 * If the host uses radix, the guest starts out as radix.
3891 if (radix_enabled()) {
3892 kvm->arch.radix = 1;
3893 kvm->arch.mmu_ready = 1;
3895 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3896 ret = kvmppc_init_vm_radix(kvm);
3898 kvmppc_free_lpid(kvm->arch.lpid);
3901 kvmppc_setup_partition_table(kvm);
3904 kvm->arch.lpcr = lpcr;
3906 /* Initialization for future HPT resizes */
3907 kvm->arch.resize_hpt = NULL;
3910 * Work out how many sets the TLB has, for the use of
3911 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3913 if (radix_enabled())
3914 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
3915 else if (cpu_has_feature(CPU_FTR_ARCH_300))
3916 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
3917 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3918 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
3920 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
3923 * Track that we now have a HV mode VM active. This blocks secondary
3924 * CPU threads from coming online.
3925 * On POWER9, we only need to do this if the "indep_threads_mode"
3926 * module parameter has been set to N.
3928 if (cpu_has_feature(CPU_FTR_ARCH_300))
3929 kvm->arch.threads_indep = indep_threads_mode;
3930 if (!kvm->arch.threads_indep)
3931 kvm_hv_vm_activated();
3934 * Initialize smt_mode depending on processor.
3935 * POWER8 and earlier have to use "strict" threading, where
3936 * all vCPUs in a vcore have to run on the same (sub)core,
3937 * whereas on POWER9 the threads can each run a different
3940 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3941 kvm->arch.smt_mode = threads_per_subcore;
3943 kvm->arch.smt_mode = 1;
3944 kvm->arch.emul_smt_mode = 1;
3947 * Create a debugfs directory for the VM
3949 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3950 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3951 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3952 kvmppc_mmu_debugfs_init(kvm);
3957 static void kvmppc_free_vcores(struct kvm *kvm)
3961 for (i = 0; i < KVM_MAX_VCORES; ++i)
3962 kfree(kvm->arch.vcores[i]);
3963 kvm->arch.online_vcores = 0;
3966 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3968 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3970 if (!kvm->arch.threads_indep)
3971 kvm_hv_vm_deactivated();
3973 kvmppc_free_vcores(kvm);
3975 kvmppc_free_lpid(kvm->arch.lpid);
3977 if (kvm_is_radix(kvm))
3978 kvmppc_free_radix(kvm);
3980 kvmppc_free_hpt(&kvm->arch.hpt);
3982 kvmppc_free_pimap(kvm);
3985 /* We don't need to emulate any privileged instructions or dcbz */
3986 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3987 unsigned int inst, int *advance)
3989 return EMULATE_FAIL;
3992 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3995 return EMULATE_FAIL;
3998 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4001 return EMULATE_FAIL;
4004 static int kvmppc_core_check_processor_compat_hv(void)
4006 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
4007 !cpu_has_feature(CPU_FTR_ARCH_206))
4013 #ifdef CONFIG_KVM_XICS
4015 void kvmppc_free_pimap(struct kvm *kvm)
4017 kfree(kvm->arch.pimap);
4020 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4022 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4025 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4027 struct irq_desc *desc;
4028 struct kvmppc_irq_map *irq_map;
4029 struct kvmppc_passthru_irqmap *pimap;
4030 struct irq_chip *chip;
4033 if (!kvm_irq_bypass)
4036 desc = irq_to_desc(host_irq);
4040 mutex_lock(&kvm->lock);
4042 pimap = kvm->arch.pimap;
4043 if (pimap == NULL) {
4044 /* First call, allocate structure to hold IRQ map */
4045 pimap = kvmppc_alloc_pimap();
4046 if (pimap == NULL) {
4047 mutex_unlock(&kvm->lock);
4050 kvm->arch.pimap = pimap;
4054 * For now, we only support interrupts for which the EOI operation
4055 * is an OPAL call followed by a write to XIRR, since that's
4056 * what our real-mode EOI code does, or a XIVE interrupt
4058 chip = irq_data_get_irq_chip(&desc->irq_data);
4059 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4060 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4061 host_irq, guest_gsi);
4062 mutex_unlock(&kvm->lock);
4067 * See if we already have an entry for this guest IRQ number.
4068 * If it's mapped to a hardware IRQ number, that's an error,
4069 * otherwise re-use this entry.
4071 for (i = 0; i < pimap->n_mapped; i++) {
4072 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4073 if (pimap->mapped[i].r_hwirq) {
4074 mutex_unlock(&kvm->lock);
4081 if (i == KVMPPC_PIRQ_MAPPED) {
4082 mutex_unlock(&kvm->lock);
4083 return -EAGAIN; /* table is full */
4086 irq_map = &pimap->mapped[i];
4088 irq_map->v_hwirq = guest_gsi;
4089 irq_map->desc = desc;
4092 * Order the above two stores before the next to serialize with
4093 * the KVM real mode handler.
4096 irq_map->r_hwirq = desc->irq_data.hwirq;
4098 if (i == pimap->n_mapped)
4102 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4104 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4106 irq_map->r_hwirq = 0;
4108 mutex_unlock(&kvm->lock);
4113 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4115 struct irq_desc *desc;
4116 struct kvmppc_passthru_irqmap *pimap;
4119 if (!kvm_irq_bypass)
4122 desc = irq_to_desc(host_irq);
4126 mutex_lock(&kvm->lock);
4127 if (!kvm->arch.pimap)
4130 pimap = kvm->arch.pimap;
4132 for (i = 0; i < pimap->n_mapped; i++) {
4133 if (guest_gsi == pimap->mapped[i].v_hwirq)
4137 if (i == pimap->n_mapped) {
4138 mutex_unlock(&kvm->lock);
4143 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4145 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4147 /* invalidate the entry (what do do on error from the above ?) */
4148 pimap->mapped[i].r_hwirq = 0;
4151 * We don't free this structure even when the count goes to
4152 * zero. The structure is freed when we destroy the VM.
4155 mutex_unlock(&kvm->lock);
4159 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4160 struct irq_bypass_producer *prod)
4163 struct kvm_kernel_irqfd *irqfd =
4164 container_of(cons, struct kvm_kernel_irqfd, consumer);
4166 irqfd->producer = prod;
4168 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4170 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4171 prod->irq, irqfd->gsi, ret);
4176 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4177 struct irq_bypass_producer *prod)
4180 struct kvm_kernel_irqfd *irqfd =
4181 container_of(cons, struct kvm_kernel_irqfd, consumer);
4183 irqfd->producer = NULL;
4186 * When producer of consumer is unregistered, we change back to
4187 * default external interrupt handling mode - KVM real mode
4188 * will switch back to host.
4190 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4192 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4193 prod->irq, irqfd->gsi, ret);
4197 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4198 unsigned int ioctl, unsigned long arg)
4200 struct kvm *kvm __maybe_unused = filp->private_data;
4201 void __user *argp = (void __user *)arg;
4206 case KVM_PPC_ALLOCATE_HTAB: {
4210 if (get_user(htab_order, (u32 __user *)argp))
4212 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4219 case KVM_PPC_GET_HTAB_FD: {
4220 struct kvm_get_htab_fd ghf;
4223 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4225 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4229 case KVM_PPC_RESIZE_HPT_PREPARE: {
4230 struct kvm_ppc_resize_hpt rhpt;
4233 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4236 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4240 case KVM_PPC_RESIZE_HPT_COMMIT: {
4241 struct kvm_ppc_resize_hpt rhpt;
4244 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4247 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4259 * List of hcall numbers to enable by default.
4260 * For compatibility with old userspace, we enable by default
4261 * all hcalls that were implemented before the hcall-enabling
4262 * facility was added. Note this list should not include H_RTAS.
4264 static unsigned int default_hcall_list[] = {
4278 #ifdef CONFIG_KVM_XICS
4289 static void init_default_hcalls(void)
4294 for (i = 0; default_hcall_list[i]; ++i) {
4295 hcall = default_hcall_list[i];
4296 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4297 __set_bit(hcall / 4, default_enabled_hcalls);
4301 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4307 /* If not on a POWER9, reject it */
4308 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4311 /* If any unknown flags set, reject it */
4312 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4315 /* GR (guest radix) bit in process_table field must match */
4316 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4317 if (!!(cfg->process_table & PATB_GR) != radix)
4320 /* Process table size field must be reasonable, i.e. <= 24 */
4321 if ((cfg->process_table & PRTS_MASK) > 24)
4324 /* We can change a guest to/from radix now, if the host is radix */
4325 if (radix && !radix_enabled())
4328 mutex_lock(&kvm->lock);
4329 if (radix != kvm_is_radix(kvm)) {
4330 if (kvm->arch.mmu_ready) {
4331 kvm->arch.mmu_ready = 0;
4332 /* order mmu_ready vs. vcpus_running */
4334 if (atomic_read(&kvm->arch.vcpus_running)) {
4335 kvm->arch.mmu_ready = 1;
4341 err = kvmppc_switch_mmu_to_radix(kvm);
4343 err = kvmppc_switch_mmu_to_hpt(kvm);
4348 kvm->arch.process_table = cfg->process_table;
4349 kvmppc_setup_partition_table(kvm);
4351 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4352 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4356 mutex_unlock(&kvm->lock);
4360 static struct kvmppc_ops kvm_ops_hv = {
4361 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4362 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4363 .get_one_reg = kvmppc_get_one_reg_hv,
4364 .set_one_reg = kvmppc_set_one_reg_hv,
4365 .vcpu_load = kvmppc_core_vcpu_load_hv,
4366 .vcpu_put = kvmppc_core_vcpu_put_hv,
4367 .set_msr = kvmppc_set_msr_hv,
4368 .vcpu_run = kvmppc_vcpu_run_hv,
4369 .vcpu_create = kvmppc_core_vcpu_create_hv,
4370 .vcpu_free = kvmppc_core_vcpu_free_hv,
4371 .check_requests = kvmppc_core_check_requests_hv,
4372 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4373 .flush_memslot = kvmppc_core_flush_memslot_hv,
4374 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4375 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4376 .unmap_hva = kvm_unmap_hva_hv,
4377 .unmap_hva_range = kvm_unmap_hva_range_hv,
4378 .age_hva = kvm_age_hva_hv,
4379 .test_age_hva = kvm_test_age_hva_hv,
4380 .set_spte_hva = kvm_set_spte_hva_hv,
4381 .mmu_destroy = kvmppc_mmu_destroy_hv,
4382 .free_memslot = kvmppc_core_free_memslot_hv,
4383 .create_memslot = kvmppc_core_create_memslot_hv,
4384 .init_vm = kvmppc_core_init_vm_hv,
4385 .destroy_vm = kvmppc_core_destroy_vm_hv,
4386 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4387 .emulate_op = kvmppc_core_emulate_op_hv,
4388 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4389 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4390 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4391 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4392 .hcall_implemented = kvmppc_hcall_impl_hv,
4393 #ifdef CONFIG_KVM_XICS
4394 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4395 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4397 .configure_mmu = kvmhv_configure_mmu,
4398 .get_rmmu_info = kvmhv_get_rmmu_info,
4399 .set_smt_mode = kvmhv_set_smt_mode,
4402 static int kvm_init_subcore_bitmap(void)
4405 int nr_cores = cpu_nr_cores();
4406 struct sibling_subcore_state *sibling_subcore_state;
4408 for (i = 0; i < nr_cores; i++) {
4409 int first_cpu = i * threads_per_core;
4410 int node = cpu_to_node(first_cpu);
4412 /* Ignore if it is already allocated. */
4413 if (paca_ptrs[first_cpu]->sibling_subcore_state)
4416 sibling_subcore_state =
4417 kmalloc_node(sizeof(struct sibling_subcore_state),
4419 if (!sibling_subcore_state)
4422 memset(sibling_subcore_state, 0,
4423 sizeof(struct sibling_subcore_state));
4425 for (j = 0; j < threads_per_core; j++) {
4426 int cpu = first_cpu + j;
4428 paca_ptrs[cpu]->sibling_subcore_state =
4429 sibling_subcore_state;
4435 static int kvmppc_radix_possible(void)
4437 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4440 static int kvmppc_book3s_init_hv(void)
4444 * FIXME!! Do we need to check on all cpus ?
4446 r = kvmppc_core_check_processor_compat_hv();
4450 r = kvm_init_subcore_bitmap();
4455 * We need a way of accessing the XICS interrupt controller,
4456 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
4457 * indirectly, via OPAL.
4460 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4461 struct device_node *np;
4463 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4465 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4471 kvm_ops_hv.owner = THIS_MODULE;
4472 kvmppc_hv_ops = &kvm_ops_hv;
4474 init_default_hcalls();
4478 r = kvmppc_mmu_hv_init();
4482 if (kvmppc_radix_possible())
4483 r = kvmppc_radix_init();
4486 * POWER9 chips before version 2.02 can't have some threads in
4487 * HPT mode and some in radix mode on the same core.
4489 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4490 unsigned int pvr = mfspr(SPRN_PVR);
4491 if ((pvr >> 16) == PVR_POWER9 &&
4492 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
4493 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
4494 no_mixing_hpt_and_radix = true;
4500 static void kvmppc_book3s_exit_hv(void)
4502 kvmppc_free_host_rm_ops();
4503 if (kvmppc_radix_possible())
4504 kvmppc_radix_exit();
4505 kvmppc_hv_ops = NULL;
4508 module_init(kvmppc_book3s_init_hv);
4509 module_exit(kvmppc_book3s_exit_hv);
4510 MODULE_LICENSE("GPL");
4511 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4512 MODULE_ALIAS("devname:kvm");