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))
746 return H_UNSUPPORTED_FLAG_START;
747 if (value2 & DABRX_HYP)
749 vcpu->arch.dawr = value1;
750 vcpu->arch.dawrx = value2;
757 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
759 struct kvmppc_vcore *vcore = target->arch.vcore;
762 * We expect to have been called by the real mode handler
763 * (kvmppc_rm_h_confer()) which would have directly returned
764 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
765 * have useful work to do and should not confer) so we don't
769 spin_lock(&vcore->lock);
770 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
771 vcore->vcore_state != VCORE_INACTIVE &&
773 target = vcore->runner;
774 spin_unlock(&vcore->lock);
776 return kvm_vcpu_yield_to(target);
779 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
782 struct lppaca *lppaca;
784 spin_lock(&vcpu->arch.vpa_update_lock);
785 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
787 yield_count = be32_to_cpu(lppaca->yield_count);
788 spin_unlock(&vcpu->arch.vpa_update_lock);
792 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
794 unsigned long req = kvmppc_get_gpr(vcpu, 3);
795 unsigned long target, ret = H_SUCCESS;
797 struct kvm_vcpu *tvcpu;
800 if (req <= MAX_HCALL_OPCODE &&
801 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
808 target = kvmppc_get_gpr(vcpu, 4);
809 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
814 tvcpu->arch.prodded = 1;
816 if (tvcpu->arch.ceded)
817 kvmppc_fast_vcpu_kick_hv(tvcpu);
820 target = kvmppc_get_gpr(vcpu, 4);
823 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
828 yield_count = kvmppc_get_gpr(vcpu, 5);
829 if (kvmppc_get_yield_count(tvcpu) != yield_count)
831 kvm_arch_vcpu_yield_to(tvcpu);
834 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
835 kvmppc_get_gpr(vcpu, 5),
836 kvmppc_get_gpr(vcpu, 6));
839 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
842 idx = srcu_read_lock(&vcpu->kvm->srcu);
843 rc = kvmppc_rtas_hcall(vcpu);
844 srcu_read_unlock(&vcpu->kvm->srcu, idx);
851 /* Send the error out to userspace via KVM_RUN */
853 case H_LOGICAL_CI_LOAD:
854 ret = kvmppc_h_logical_ci_load(vcpu);
855 if (ret == H_TOO_HARD)
858 case H_LOGICAL_CI_STORE:
859 ret = kvmppc_h_logical_ci_store(vcpu);
860 if (ret == H_TOO_HARD)
864 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
865 kvmppc_get_gpr(vcpu, 5),
866 kvmppc_get_gpr(vcpu, 6),
867 kvmppc_get_gpr(vcpu, 7));
868 if (ret == H_TOO_HARD)
877 if (kvmppc_xics_enabled(vcpu)) {
878 if (xive_enabled()) {
879 ret = H_NOT_AVAILABLE;
882 ret = kvmppc_xics_hcall(vcpu, req);
887 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
888 kvmppc_get_gpr(vcpu, 5),
889 kvmppc_get_gpr(vcpu, 6));
890 if (ret == H_TOO_HARD)
893 case H_PUT_TCE_INDIRECT:
894 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
895 kvmppc_get_gpr(vcpu, 5),
896 kvmppc_get_gpr(vcpu, 6),
897 kvmppc_get_gpr(vcpu, 7));
898 if (ret == H_TOO_HARD)
902 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
903 kvmppc_get_gpr(vcpu, 5),
904 kvmppc_get_gpr(vcpu, 6),
905 kvmppc_get_gpr(vcpu, 7));
906 if (ret == H_TOO_HARD)
912 kvmppc_set_gpr(vcpu, 3, ret);
913 vcpu->arch.hcall_needed = 0;
917 static int kvmppc_hcall_impl_hv(unsigned long cmd)
925 case H_LOGICAL_CI_LOAD:
926 case H_LOGICAL_CI_STORE:
927 #ifdef CONFIG_KVM_XICS
938 /* See if it's in the real-mode table */
939 return kvmppc_hcall_impl_hv_realmode(cmd);
942 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
943 struct kvm_vcpu *vcpu)
947 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
950 * Fetch failed, so return to guest and
951 * try executing it again.
956 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
957 run->exit_reason = KVM_EXIT_DEBUG;
958 run->debug.arch.address = kvmppc_get_pc(vcpu);
961 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
966 static void do_nothing(void *x)
970 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
972 int thr, cpu, pcpu, nthreads;
976 nthreads = vcpu->kvm->arch.emul_smt_mode;
978 cpu = vcpu->vcpu_id & ~(nthreads - 1);
979 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
980 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
984 * If the vcpu is currently running on a physical cpu thread,
985 * interrupt it in order to pull it out of the guest briefly,
986 * which will update its vcore->dpdes value.
988 pcpu = READ_ONCE(v->cpu);
990 smp_call_function_single(pcpu, do_nothing, NULL, 1);
991 if (kvmppc_doorbell_pending(v))
998 * On POWER9, emulate doorbell-related instructions in order to
999 * give the guest the illusion of running on a multi-threaded core.
1000 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1003 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1007 struct kvm *kvm = vcpu->kvm;
1008 struct kvm_vcpu *tvcpu;
1010 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1011 return RESUME_GUEST;
1012 if (get_op(inst) != 31)
1013 return EMULATE_FAIL;
1015 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1016 switch (get_xop(inst)) {
1017 case OP_31_XOP_MSGSNDP:
1018 arg = kvmppc_get_gpr(vcpu, rb);
1019 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1022 if (arg >= kvm->arch.emul_smt_mode)
1024 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1027 if (!tvcpu->arch.doorbell_request) {
1028 tvcpu->arch.doorbell_request = 1;
1029 kvmppc_fast_vcpu_kick_hv(tvcpu);
1032 case OP_31_XOP_MSGCLRP:
1033 arg = kvmppc_get_gpr(vcpu, rb);
1034 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1036 vcpu->arch.vcore->dpdes = 0;
1037 vcpu->arch.doorbell_request = 0;
1039 case OP_31_XOP_MFSPR:
1040 switch (get_sprn(inst)) {
1045 arg = kvmppc_read_dpdes(vcpu);
1048 return EMULATE_FAIL;
1050 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1053 return EMULATE_FAIL;
1055 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1056 return RESUME_GUEST;
1059 /* Called with vcpu->arch.vcore->lock held */
1060 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1061 struct task_struct *tsk)
1063 int r = RESUME_HOST;
1065 vcpu->stat.sum_exits++;
1068 * This can happen if an interrupt occurs in the last stages
1069 * of guest entry or the first stages of guest exit (i.e. after
1070 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1071 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1072 * That can happen due to a bug, or due to a machine check
1073 * occurring at just the wrong time.
1075 if (vcpu->arch.shregs.msr & MSR_HV) {
1076 printk(KERN_EMERG "KVM trap in HV mode!\n");
1077 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1078 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1079 vcpu->arch.shregs.msr);
1080 kvmppc_dump_regs(vcpu);
1081 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1082 run->hw.hardware_exit_reason = vcpu->arch.trap;
1085 run->exit_reason = KVM_EXIT_UNKNOWN;
1086 run->ready_for_interrupt_injection = 1;
1087 switch (vcpu->arch.trap) {
1088 /* We're good on these - the host merely wanted to get our attention */
1089 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1090 vcpu->stat.dec_exits++;
1093 case BOOK3S_INTERRUPT_EXTERNAL:
1094 case BOOK3S_INTERRUPT_H_DOORBELL:
1095 case BOOK3S_INTERRUPT_H_VIRT:
1096 vcpu->stat.ext_intr_exits++;
1099 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1100 case BOOK3S_INTERRUPT_HMI:
1101 case BOOK3S_INTERRUPT_PERFMON:
1102 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1105 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1106 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1107 run->exit_reason = KVM_EXIT_NMI;
1108 run->hw.hardware_exit_reason = vcpu->arch.trap;
1109 /* Clear out the old NMI status from run->flags */
1110 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1111 /* Now set the NMI status */
1112 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1113 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1115 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1118 /* Print the MCE event to host console. */
1119 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1121 case BOOK3S_INTERRUPT_PROGRAM:
1125 * Normally program interrupts are delivered directly
1126 * to the guest by the hardware, but we can get here
1127 * as a result of a hypervisor emulation interrupt
1128 * (e40) getting turned into a 700 by BML RTAS.
1130 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1131 kvmppc_core_queue_program(vcpu, flags);
1135 case BOOK3S_INTERRUPT_SYSCALL:
1137 /* hcall - punt to userspace */
1140 /* hypercall with MSR_PR has already been handled in rmode,
1141 * and never reaches here.
1144 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1145 for (i = 0; i < 9; ++i)
1146 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1147 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1148 vcpu->arch.hcall_needed = 1;
1153 * We get these next two if the guest accesses a page which it thinks
1154 * it has mapped but which is not actually present, either because
1155 * it is for an emulated I/O device or because the corresonding
1156 * host page has been paged out. Any other HDSI/HISI interrupts
1157 * have been handled already.
1159 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1160 r = RESUME_PAGE_FAULT;
1162 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1163 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1164 vcpu->arch.fault_dsisr = 0;
1165 r = RESUME_PAGE_FAULT;
1168 * This occurs if the guest executes an illegal instruction.
1169 * If the guest debug is disabled, generate a program interrupt
1170 * to the guest. If guest debug is enabled, we need to check
1171 * whether the instruction is a software breakpoint instruction.
1172 * Accordingly return to Guest or Host.
1174 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1175 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1176 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1177 swab32(vcpu->arch.emul_inst) :
1178 vcpu->arch.emul_inst;
1179 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1180 /* Need vcore unlocked to call kvmppc_get_last_inst */
1181 spin_unlock(&vcpu->arch.vcore->lock);
1182 r = kvmppc_emulate_debug_inst(run, vcpu);
1183 spin_lock(&vcpu->arch.vcore->lock);
1185 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1190 * This occurs if the guest (kernel or userspace), does something that
1191 * is prohibited by HFSCR.
1192 * On POWER9, this could be a doorbell instruction that we need
1194 * Otherwise, we just generate a program interrupt to the guest.
1196 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1198 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1199 cpu_has_feature(CPU_FTR_ARCH_300)) {
1200 /* Need vcore unlocked to call kvmppc_get_last_inst */
1201 spin_unlock(&vcpu->arch.vcore->lock);
1202 r = kvmppc_emulate_doorbell_instr(vcpu);
1203 spin_lock(&vcpu->arch.vcore->lock);
1205 if (r == EMULATE_FAIL) {
1206 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1210 case BOOK3S_INTERRUPT_HV_RM_HARD:
1211 r = RESUME_PASSTHROUGH;
1214 kvmppc_dump_regs(vcpu);
1215 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1216 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1217 vcpu->arch.shregs.msr);
1218 run->hw.hardware_exit_reason = vcpu->arch.trap;
1226 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1227 struct kvm_sregs *sregs)
1231 memset(sregs, 0, sizeof(struct kvm_sregs));
1232 sregs->pvr = vcpu->arch.pvr;
1233 for (i = 0; i < vcpu->arch.slb_max; i++) {
1234 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1235 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1241 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1242 struct kvm_sregs *sregs)
1246 /* Only accept the same PVR as the host's, since we can't spoof it */
1247 if (sregs->pvr != vcpu->arch.pvr)
1251 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1252 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1253 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1254 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1258 vcpu->arch.slb_max = j;
1263 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1264 bool preserve_top32)
1266 struct kvm *kvm = vcpu->kvm;
1267 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1270 mutex_lock(&kvm->lock);
1271 spin_lock(&vc->lock);
1273 * If ILE (interrupt little-endian) has changed, update the
1274 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1276 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1277 struct kvm_vcpu *vcpu;
1280 kvm_for_each_vcpu(i, vcpu, kvm) {
1281 if (vcpu->arch.vcore != vc)
1283 if (new_lpcr & LPCR_ILE)
1284 vcpu->arch.intr_msr |= MSR_LE;
1286 vcpu->arch.intr_msr &= ~MSR_LE;
1291 * Userspace can only modify DPFD (default prefetch depth),
1292 * ILE (interrupt little-endian) and TC (translation control).
1293 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1295 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1296 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1299 * On POWER9, allow userspace to enable large decrementer for the
1300 * guest, whether or not the host has it enabled.
1302 if (cpu_has_feature(CPU_FTR_ARCH_300))
1305 /* Broken 32-bit version of LPCR must not clear top bits */
1308 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1309 spin_unlock(&vc->lock);
1310 mutex_unlock(&kvm->lock);
1313 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1314 union kvmppc_one_reg *val)
1320 case KVM_REG_PPC_DEBUG_INST:
1321 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1323 case KVM_REG_PPC_HIOR:
1324 *val = get_reg_val(id, 0);
1326 case KVM_REG_PPC_DABR:
1327 *val = get_reg_val(id, vcpu->arch.dabr);
1329 case KVM_REG_PPC_DABRX:
1330 *val = get_reg_val(id, vcpu->arch.dabrx);
1332 case KVM_REG_PPC_DSCR:
1333 *val = get_reg_val(id, vcpu->arch.dscr);
1335 case KVM_REG_PPC_PURR:
1336 *val = get_reg_val(id, vcpu->arch.purr);
1338 case KVM_REG_PPC_SPURR:
1339 *val = get_reg_val(id, vcpu->arch.spurr);
1341 case KVM_REG_PPC_AMR:
1342 *val = get_reg_val(id, vcpu->arch.amr);
1344 case KVM_REG_PPC_UAMOR:
1345 *val = get_reg_val(id, vcpu->arch.uamor);
1347 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1348 i = id - KVM_REG_PPC_MMCR0;
1349 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1351 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1352 i = id - KVM_REG_PPC_PMC1;
1353 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1355 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1356 i = id - KVM_REG_PPC_SPMC1;
1357 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1359 case KVM_REG_PPC_SIAR:
1360 *val = get_reg_val(id, vcpu->arch.siar);
1362 case KVM_REG_PPC_SDAR:
1363 *val = get_reg_val(id, vcpu->arch.sdar);
1365 case KVM_REG_PPC_SIER:
1366 *val = get_reg_val(id, vcpu->arch.sier);
1368 case KVM_REG_PPC_IAMR:
1369 *val = get_reg_val(id, vcpu->arch.iamr);
1371 case KVM_REG_PPC_PSPB:
1372 *val = get_reg_val(id, vcpu->arch.pspb);
1374 case KVM_REG_PPC_DPDES:
1375 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1377 case KVM_REG_PPC_VTB:
1378 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1380 case KVM_REG_PPC_DAWR:
1381 *val = get_reg_val(id, vcpu->arch.dawr);
1383 case KVM_REG_PPC_DAWRX:
1384 *val = get_reg_val(id, vcpu->arch.dawrx);
1386 case KVM_REG_PPC_CIABR:
1387 *val = get_reg_val(id, vcpu->arch.ciabr);
1389 case KVM_REG_PPC_CSIGR:
1390 *val = get_reg_val(id, vcpu->arch.csigr);
1392 case KVM_REG_PPC_TACR:
1393 *val = get_reg_val(id, vcpu->arch.tacr);
1395 case KVM_REG_PPC_TCSCR:
1396 *val = get_reg_val(id, vcpu->arch.tcscr);
1398 case KVM_REG_PPC_PID:
1399 *val = get_reg_val(id, vcpu->arch.pid);
1401 case KVM_REG_PPC_ACOP:
1402 *val = get_reg_val(id, vcpu->arch.acop);
1404 case KVM_REG_PPC_WORT:
1405 *val = get_reg_val(id, vcpu->arch.wort);
1407 case KVM_REG_PPC_TIDR:
1408 *val = get_reg_val(id, vcpu->arch.tid);
1410 case KVM_REG_PPC_PSSCR:
1411 *val = get_reg_val(id, vcpu->arch.psscr);
1413 case KVM_REG_PPC_VPA_ADDR:
1414 spin_lock(&vcpu->arch.vpa_update_lock);
1415 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1416 spin_unlock(&vcpu->arch.vpa_update_lock);
1418 case KVM_REG_PPC_VPA_SLB:
1419 spin_lock(&vcpu->arch.vpa_update_lock);
1420 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1421 val->vpaval.length = vcpu->arch.slb_shadow.len;
1422 spin_unlock(&vcpu->arch.vpa_update_lock);
1424 case KVM_REG_PPC_VPA_DTL:
1425 spin_lock(&vcpu->arch.vpa_update_lock);
1426 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1427 val->vpaval.length = vcpu->arch.dtl.len;
1428 spin_unlock(&vcpu->arch.vpa_update_lock);
1430 case KVM_REG_PPC_TB_OFFSET:
1431 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1433 case KVM_REG_PPC_LPCR:
1434 case KVM_REG_PPC_LPCR_64:
1435 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1437 case KVM_REG_PPC_PPR:
1438 *val = get_reg_val(id, vcpu->arch.ppr);
1440 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1441 case KVM_REG_PPC_TFHAR:
1442 *val = get_reg_val(id, vcpu->arch.tfhar);
1444 case KVM_REG_PPC_TFIAR:
1445 *val = get_reg_val(id, vcpu->arch.tfiar);
1447 case KVM_REG_PPC_TEXASR:
1448 *val = get_reg_val(id, vcpu->arch.texasr);
1450 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1451 i = id - KVM_REG_PPC_TM_GPR0;
1452 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1454 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1457 i = id - KVM_REG_PPC_TM_VSR0;
1459 for (j = 0; j < TS_FPRWIDTH; j++)
1460 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1462 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1463 val->vval = vcpu->arch.vr_tm.vr[i-32];
1469 case KVM_REG_PPC_TM_CR:
1470 *val = get_reg_val(id, vcpu->arch.cr_tm);
1472 case KVM_REG_PPC_TM_XER:
1473 *val = get_reg_val(id, vcpu->arch.xer_tm);
1475 case KVM_REG_PPC_TM_LR:
1476 *val = get_reg_val(id, vcpu->arch.lr_tm);
1478 case KVM_REG_PPC_TM_CTR:
1479 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1481 case KVM_REG_PPC_TM_FPSCR:
1482 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1484 case KVM_REG_PPC_TM_AMR:
1485 *val = get_reg_val(id, vcpu->arch.amr_tm);
1487 case KVM_REG_PPC_TM_PPR:
1488 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1490 case KVM_REG_PPC_TM_VRSAVE:
1491 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1493 case KVM_REG_PPC_TM_VSCR:
1494 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1495 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1499 case KVM_REG_PPC_TM_DSCR:
1500 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1502 case KVM_REG_PPC_TM_TAR:
1503 *val = get_reg_val(id, vcpu->arch.tar_tm);
1506 case KVM_REG_PPC_ARCH_COMPAT:
1507 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1509 case KVM_REG_PPC_DEC_EXPIRY:
1510 *val = get_reg_val(id, vcpu->arch.dec_expires +
1511 vcpu->arch.vcore->tb_offset);
1521 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1522 union kvmppc_one_reg *val)
1526 unsigned long addr, len;
1529 case KVM_REG_PPC_HIOR:
1530 /* Only allow this to be set to zero */
1531 if (set_reg_val(id, *val))
1534 case KVM_REG_PPC_DABR:
1535 vcpu->arch.dabr = set_reg_val(id, *val);
1537 case KVM_REG_PPC_DABRX:
1538 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1540 case KVM_REG_PPC_DSCR:
1541 vcpu->arch.dscr = set_reg_val(id, *val);
1543 case KVM_REG_PPC_PURR:
1544 vcpu->arch.purr = set_reg_val(id, *val);
1546 case KVM_REG_PPC_SPURR:
1547 vcpu->arch.spurr = set_reg_val(id, *val);
1549 case KVM_REG_PPC_AMR:
1550 vcpu->arch.amr = set_reg_val(id, *val);
1552 case KVM_REG_PPC_UAMOR:
1553 vcpu->arch.uamor = set_reg_val(id, *val);
1555 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1556 i = id - KVM_REG_PPC_MMCR0;
1557 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1559 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1560 i = id - KVM_REG_PPC_PMC1;
1561 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1563 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1564 i = id - KVM_REG_PPC_SPMC1;
1565 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1567 case KVM_REG_PPC_SIAR:
1568 vcpu->arch.siar = set_reg_val(id, *val);
1570 case KVM_REG_PPC_SDAR:
1571 vcpu->arch.sdar = set_reg_val(id, *val);
1573 case KVM_REG_PPC_SIER:
1574 vcpu->arch.sier = set_reg_val(id, *val);
1576 case KVM_REG_PPC_IAMR:
1577 vcpu->arch.iamr = set_reg_val(id, *val);
1579 case KVM_REG_PPC_PSPB:
1580 vcpu->arch.pspb = set_reg_val(id, *val);
1582 case KVM_REG_PPC_DPDES:
1583 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1585 case KVM_REG_PPC_VTB:
1586 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1588 case KVM_REG_PPC_DAWR:
1589 vcpu->arch.dawr = set_reg_val(id, *val);
1591 case KVM_REG_PPC_DAWRX:
1592 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1594 case KVM_REG_PPC_CIABR:
1595 vcpu->arch.ciabr = set_reg_val(id, *val);
1596 /* Don't allow setting breakpoints in hypervisor code */
1597 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1598 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1600 case KVM_REG_PPC_CSIGR:
1601 vcpu->arch.csigr = set_reg_val(id, *val);
1603 case KVM_REG_PPC_TACR:
1604 vcpu->arch.tacr = set_reg_val(id, *val);
1606 case KVM_REG_PPC_TCSCR:
1607 vcpu->arch.tcscr = set_reg_val(id, *val);
1609 case KVM_REG_PPC_PID:
1610 vcpu->arch.pid = set_reg_val(id, *val);
1612 case KVM_REG_PPC_ACOP:
1613 vcpu->arch.acop = set_reg_val(id, *val);
1615 case KVM_REG_PPC_WORT:
1616 vcpu->arch.wort = set_reg_val(id, *val);
1618 case KVM_REG_PPC_TIDR:
1619 vcpu->arch.tid = set_reg_val(id, *val);
1621 case KVM_REG_PPC_PSSCR:
1622 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1624 case KVM_REG_PPC_VPA_ADDR:
1625 addr = set_reg_val(id, *val);
1627 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1628 vcpu->arch.dtl.next_gpa))
1630 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1632 case KVM_REG_PPC_VPA_SLB:
1633 addr = val->vpaval.addr;
1634 len = val->vpaval.length;
1636 if (addr && !vcpu->arch.vpa.next_gpa)
1638 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1640 case KVM_REG_PPC_VPA_DTL:
1641 addr = val->vpaval.addr;
1642 len = val->vpaval.length;
1644 if (addr && (len < sizeof(struct dtl_entry) ||
1645 !vcpu->arch.vpa.next_gpa))
1647 len -= len % sizeof(struct dtl_entry);
1648 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1650 case KVM_REG_PPC_TB_OFFSET:
1652 * POWER9 DD1 has an erratum where writing TBU40 causes
1653 * the timebase to lose ticks. So we don't let the
1654 * timebase offset be changed on P9 DD1. (It is
1655 * initialized to zero.)
1657 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1659 /* round up to multiple of 2^24 */
1660 vcpu->arch.vcore->tb_offset =
1661 ALIGN(set_reg_val(id, *val), 1UL << 24);
1663 case KVM_REG_PPC_LPCR:
1664 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1666 case KVM_REG_PPC_LPCR_64:
1667 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1669 case KVM_REG_PPC_PPR:
1670 vcpu->arch.ppr = set_reg_val(id, *val);
1672 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1673 case KVM_REG_PPC_TFHAR:
1674 vcpu->arch.tfhar = set_reg_val(id, *val);
1676 case KVM_REG_PPC_TFIAR:
1677 vcpu->arch.tfiar = set_reg_val(id, *val);
1679 case KVM_REG_PPC_TEXASR:
1680 vcpu->arch.texasr = set_reg_val(id, *val);
1682 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1683 i = id - KVM_REG_PPC_TM_GPR0;
1684 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1686 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1689 i = id - KVM_REG_PPC_TM_VSR0;
1691 for (j = 0; j < TS_FPRWIDTH; j++)
1692 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1694 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1695 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1700 case KVM_REG_PPC_TM_CR:
1701 vcpu->arch.cr_tm = set_reg_val(id, *val);
1703 case KVM_REG_PPC_TM_XER:
1704 vcpu->arch.xer_tm = set_reg_val(id, *val);
1706 case KVM_REG_PPC_TM_LR:
1707 vcpu->arch.lr_tm = set_reg_val(id, *val);
1709 case KVM_REG_PPC_TM_CTR:
1710 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1712 case KVM_REG_PPC_TM_FPSCR:
1713 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1715 case KVM_REG_PPC_TM_AMR:
1716 vcpu->arch.amr_tm = set_reg_val(id, *val);
1718 case KVM_REG_PPC_TM_PPR:
1719 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1721 case KVM_REG_PPC_TM_VRSAVE:
1722 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1724 case KVM_REG_PPC_TM_VSCR:
1725 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1726 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1730 case KVM_REG_PPC_TM_DSCR:
1731 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1733 case KVM_REG_PPC_TM_TAR:
1734 vcpu->arch.tar_tm = set_reg_val(id, *val);
1737 case KVM_REG_PPC_ARCH_COMPAT:
1738 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1740 case KVM_REG_PPC_DEC_EXPIRY:
1741 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1742 vcpu->arch.vcore->tb_offset;
1753 * On POWER9, threads are independent and can be in different partitions.
1754 * Therefore we consider each thread to be a subcore.
1755 * There is a restriction that all threads have to be in the same
1756 * MMU mode (radix or HPT), unfortunately, but since we only support
1757 * HPT guests on a HPT host so far, that isn't an impediment yet.
1759 static int threads_per_vcore(struct kvm *kvm)
1761 if (kvm->arch.threads_indep)
1763 return threads_per_subcore;
1766 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1768 struct kvmppc_vcore *vcore;
1770 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1775 spin_lock_init(&vcore->lock);
1776 spin_lock_init(&vcore->stoltb_lock);
1777 init_swait_queue_head(&vcore->wq);
1778 vcore->preempt_tb = TB_NIL;
1779 vcore->lpcr = kvm->arch.lpcr;
1780 vcore->first_vcpuid = core * kvm->arch.smt_mode;
1782 INIT_LIST_HEAD(&vcore->preempt_list);
1787 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1788 static struct debugfs_timings_element {
1792 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1793 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1794 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1795 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1796 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1799 #define N_TIMINGS (ARRAY_SIZE(timings))
1801 struct debugfs_timings_state {
1802 struct kvm_vcpu *vcpu;
1803 unsigned int buflen;
1804 char buf[N_TIMINGS * 100];
1807 static int debugfs_timings_open(struct inode *inode, struct file *file)
1809 struct kvm_vcpu *vcpu = inode->i_private;
1810 struct debugfs_timings_state *p;
1812 p = kzalloc(sizeof(*p), GFP_KERNEL);
1816 kvm_get_kvm(vcpu->kvm);
1818 file->private_data = p;
1820 return nonseekable_open(inode, file);
1823 static int debugfs_timings_release(struct inode *inode, struct file *file)
1825 struct debugfs_timings_state *p = file->private_data;
1827 kvm_put_kvm(p->vcpu->kvm);
1832 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1833 size_t len, loff_t *ppos)
1835 struct debugfs_timings_state *p = file->private_data;
1836 struct kvm_vcpu *vcpu = p->vcpu;
1838 struct kvmhv_tb_accumulator tb;
1847 buf_end = s + sizeof(p->buf);
1848 for (i = 0; i < N_TIMINGS; ++i) {
1849 struct kvmhv_tb_accumulator *acc;
1851 acc = (struct kvmhv_tb_accumulator *)
1852 ((unsigned long)vcpu + timings[i].offset);
1854 for (loops = 0; loops < 1000; ++loops) {
1855 count = acc->seqcount;
1860 if (count == acc->seqcount) {
1868 snprintf(s, buf_end - s, "%s: stuck\n",
1871 snprintf(s, buf_end - s,
1872 "%s: %llu %llu %llu %llu\n",
1873 timings[i].name, count / 2,
1874 tb_to_ns(tb.tb_total),
1875 tb_to_ns(tb.tb_min),
1876 tb_to_ns(tb.tb_max));
1879 p->buflen = s - p->buf;
1883 if (pos >= p->buflen)
1885 if (len > p->buflen - pos)
1886 len = p->buflen - pos;
1887 n = copy_to_user(buf, p->buf + pos, len);
1897 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1898 size_t len, loff_t *ppos)
1903 static const struct file_operations debugfs_timings_ops = {
1904 .owner = THIS_MODULE,
1905 .open = debugfs_timings_open,
1906 .release = debugfs_timings_release,
1907 .read = debugfs_timings_read,
1908 .write = debugfs_timings_write,
1909 .llseek = generic_file_llseek,
1912 /* Create a debugfs directory for the vcpu */
1913 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1916 struct kvm *kvm = vcpu->kvm;
1918 snprintf(buf, sizeof(buf), "vcpu%u", id);
1919 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1921 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1922 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1924 vcpu->arch.debugfs_timings =
1925 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1926 vcpu, &debugfs_timings_ops);
1929 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1930 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1933 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1935 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1938 struct kvm_vcpu *vcpu;
1941 struct kvmppc_vcore *vcore;
1944 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1948 err = kvm_vcpu_init(vcpu, kvm, id);
1952 vcpu->arch.shared = &vcpu->arch.shregs;
1953 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1955 * The shared struct is never shared on HV,
1956 * so we can always use host endianness
1958 #ifdef __BIG_ENDIAN__
1959 vcpu->arch.shared_big_endian = true;
1961 vcpu->arch.shared_big_endian = false;
1964 vcpu->arch.mmcr[0] = MMCR0_FC;
1965 vcpu->arch.ctrl = CTRL_RUNLATCH;
1966 /* default to host PVR, since we can't spoof it */
1967 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1968 spin_lock_init(&vcpu->arch.vpa_update_lock);
1969 spin_lock_init(&vcpu->arch.tbacct_lock);
1970 vcpu->arch.busy_preempt = TB_NIL;
1971 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1974 * Set the default HFSCR for the guest from the host value.
1975 * This value is only used on POWER9.
1976 * On POWER9 DD1, TM doesn't work, so we make sure to
1977 * prevent the guest from using it.
1978 * On POWER9, we want to virtualize the doorbell facility, so we
1979 * turn off the HFSCR bit, which causes those instructions to trap.
1981 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1982 if (!cpu_has_feature(CPU_FTR_TM))
1983 vcpu->arch.hfscr &= ~HFSCR_TM;
1984 if (cpu_has_feature(CPU_FTR_ARCH_300))
1985 vcpu->arch.hfscr &= ~HFSCR_MSGP;
1987 kvmppc_mmu_book3s_hv_init(vcpu);
1989 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1991 init_waitqueue_head(&vcpu->arch.cpu_run);
1993 mutex_lock(&kvm->lock);
1996 core = id / kvm->arch.smt_mode;
1997 if (core < KVM_MAX_VCORES) {
1998 vcore = kvm->arch.vcores[core];
2001 vcore = kvmppc_vcore_create(kvm, core);
2002 kvm->arch.vcores[core] = vcore;
2003 kvm->arch.online_vcores++;
2006 mutex_unlock(&kvm->lock);
2011 spin_lock(&vcore->lock);
2012 ++vcore->num_threads;
2013 spin_unlock(&vcore->lock);
2014 vcpu->arch.vcore = vcore;
2015 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2016 vcpu->arch.thread_cpu = -1;
2017 vcpu->arch.prev_cpu = -1;
2019 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2020 kvmppc_sanity_check(vcpu);
2022 debugfs_vcpu_init(vcpu, id);
2027 kmem_cache_free(kvm_vcpu_cache, vcpu);
2029 return ERR_PTR(err);
2032 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2033 unsigned long flags)
2040 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2042 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2044 * On POWER8 (or POWER7), the threading mode is "strict",
2045 * so we pack smt_mode vcpus per vcore.
2047 if (smt_mode > threads_per_subcore)
2051 * On POWER9, the threading mode is "loose",
2052 * so each vcpu gets its own vcore.
2057 mutex_lock(&kvm->lock);
2059 if (!kvm->arch.online_vcores) {
2060 kvm->arch.smt_mode = smt_mode;
2061 kvm->arch.emul_smt_mode = esmt;
2064 mutex_unlock(&kvm->lock);
2069 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2071 if (vpa->pinned_addr)
2072 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2076 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2078 spin_lock(&vcpu->arch.vpa_update_lock);
2079 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2080 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2081 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2082 spin_unlock(&vcpu->arch.vpa_update_lock);
2083 kvm_vcpu_uninit(vcpu);
2084 kmem_cache_free(kvm_vcpu_cache, vcpu);
2087 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2089 /* Indicate we want to get back into the guest */
2093 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2095 unsigned long dec_nsec, now;
2098 if (now > vcpu->arch.dec_expires) {
2099 /* decrementer has already gone negative */
2100 kvmppc_core_queue_dec(vcpu);
2101 kvmppc_core_prepare_to_enter(vcpu);
2104 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2106 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2107 vcpu->arch.timer_running = 1;
2110 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2112 vcpu->arch.ceded = 0;
2113 if (vcpu->arch.timer_running) {
2114 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2115 vcpu->arch.timer_running = 0;
2119 extern int __kvmppc_vcore_entry(void);
2121 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2122 struct kvm_vcpu *vcpu)
2126 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2128 spin_lock_irq(&vcpu->arch.tbacct_lock);
2130 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2131 vcpu->arch.stolen_logged;
2132 vcpu->arch.busy_preempt = now;
2133 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2134 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2136 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2139 static int kvmppc_grab_hwthread(int cpu)
2141 struct paca_struct *tpaca;
2142 long timeout = 10000;
2144 tpaca = paca_ptrs[cpu];
2146 /* Ensure the thread won't go into the kernel if it wakes */
2147 tpaca->kvm_hstate.kvm_vcpu = NULL;
2148 tpaca->kvm_hstate.kvm_vcore = NULL;
2149 tpaca->kvm_hstate.napping = 0;
2151 tpaca->kvm_hstate.hwthread_req = 1;
2154 * If the thread is already executing in the kernel (e.g. handling
2155 * a stray interrupt), wait for it to get back to nap mode.
2156 * The smp_mb() is to ensure that our setting of hwthread_req
2157 * is visible before we look at hwthread_state, so if this
2158 * races with the code at system_reset_pSeries and the thread
2159 * misses our setting of hwthread_req, we are sure to see its
2160 * setting of hwthread_state, and vice versa.
2163 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2164 if (--timeout <= 0) {
2165 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2173 static void kvmppc_release_hwthread(int cpu)
2175 struct paca_struct *tpaca;
2177 tpaca = paca_ptrs[cpu];
2178 tpaca->kvm_hstate.hwthread_req = 0;
2179 tpaca->kvm_hstate.kvm_vcpu = NULL;
2180 tpaca->kvm_hstate.kvm_vcore = NULL;
2181 tpaca->kvm_hstate.kvm_split_mode = NULL;
2184 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2188 cpu = cpu_first_thread_sibling(cpu);
2189 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2191 * Make sure setting of bit in need_tlb_flush precedes
2192 * testing of cpu_in_guest bits. The matching barrier on
2193 * the other side is the first smp_mb() in kvmppc_run_core().
2196 for (i = 0; i < threads_per_core; ++i)
2197 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2198 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2201 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2203 struct kvm *kvm = vcpu->kvm;
2206 * With radix, the guest can do TLB invalidations itself,
2207 * and it could choose to use the local form (tlbiel) if
2208 * it is invalidating a translation that has only ever been
2209 * used on one vcpu. However, that doesn't mean it has
2210 * only ever been used on one physical cpu, since vcpus
2211 * can move around between pcpus. To cope with this, when
2212 * a vcpu moves from one pcpu to another, we need to tell
2213 * any vcpus running on the same core as this vcpu previously
2214 * ran to flush the TLB. The TLB is shared between threads,
2215 * so we use a single bit in .need_tlb_flush for all 4 threads.
2217 if (vcpu->arch.prev_cpu != pcpu) {
2218 if (vcpu->arch.prev_cpu >= 0 &&
2219 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2220 cpu_first_thread_sibling(pcpu))
2221 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2222 vcpu->arch.prev_cpu = pcpu;
2226 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2229 struct paca_struct *tpaca;
2230 struct kvm *kvm = vc->kvm;
2234 if (vcpu->arch.timer_running) {
2235 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2236 vcpu->arch.timer_running = 0;
2238 cpu += vcpu->arch.ptid;
2239 vcpu->cpu = vc->pcpu;
2240 vcpu->arch.thread_cpu = cpu;
2241 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2243 tpaca = paca_ptrs[cpu];
2244 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2245 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2246 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2248 tpaca->kvm_hstate.kvm_vcore = vc;
2249 if (cpu != smp_processor_id())
2250 kvmppc_ipi_thread(cpu);
2253 static void kvmppc_wait_for_nap(int n_threads)
2255 int cpu = smp_processor_id();
2260 for (loops = 0; loops < 1000000; ++loops) {
2262 * Check if all threads are finished.
2263 * We set the vcore pointer when starting a thread
2264 * and the thread clears it when finished, so we look
2265 * for any threads that still have a non-NULL vcore ptr.
2267 for (i = 1; i < n_threads; ++i)
2268 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2270 if (i == n_threads) {
2277 for (i = 1; i < n_threads; ++i)
2278 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2279 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2283 * Check that we are on thread 0 and that any other threads in
2284 * this core are off-line. Then grab the threads so they can't
2287 static int on_primary_thread(void)
2289 int cpu = smp_processor_id();
2292 /* Are we on a primary subcore? */
2293 if (cpu_thread_in_subcore(cpu))
2297 while (++thr < threads_per_subcore)
2298 if (cpu_online(cpu + thr))
2301 /* Grab all hw threads so they can't go into the kernel */
2302 for (thr = 1; thr < threads_per_subcore; ++thr) {
2303 if (kvmppc_grab_hwthread(cpu + thr)) {
2304 /* Couldn't grab one; let the others go */
2306 kvmppc_release_hwthread(cpu + thr);
2307 } while (--thr > 0);
2315 * A list of virtual cores for each physical CPU.
2316 * These are vcores that could run but their runner VCPU tasks are
2317 * (or may be) preempted.
2319 struct preempted_vcore_list {
2320 struct list_head list;
2324 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2326 static void init_vcore_lists(void)
2330 for_each_possible_cpu(cpu) {
2331 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2332 spin_lock_init(&lp->lock);
2333 INIT_LIST_HEAD(&lp->list);
2337 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2339 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2341 vc->vcore_state = VCORE_PREEMPT;
2342 vc->pcpu = smp_processor_id();
2343 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2344 spin_lock(&lp->lock);
2345 list_add_tail(&vc->preempt_list, &lp->list);
2346 spin_unlock(&lp->lock);
2349 /* Start accumulating stolen time */
2350 kvmppc_core_start_stolen(vc);
2353 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2355 struct preempted_vcore_list *lp;
2357 kvmppc_core_end_stolen(vc);
2358 if (!list_empty(&vc->preempt_list)) {
2359 lp = &per_cpu(preempted_vcores, vc->pcpu);
2360 spin_lock(&lp->lock);
2361 list_del_init(&vc->preempt_list);
2362 spin_unlock(&lp->lock);
2364 vc->vcore_state = VCORE_INACTIVE;
2368 * This stores information about the virtual cores currently
2369 * assigned to a physical core.
2373 int max_subcore_threads;
2375 int subcore_threads[MAX_SUBCORES];
2376 struct kvmppc_vcore *vc[MAX_SUBCORES];
2380 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2381 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2383 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2385 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2387 memset(cip, 0, sizeof(*cip));
2388 cip->n_subcores = 1;
2389 cip->max_subcore_threads = vc->num_threads;
2390 cip->total_threads = vc->num_threads;
2391 cip->subcore_threads[0] = vc->num_threads;
2395 static bool subcore_config_ok(int n_subcores, int n_threads)
2398 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2399 * split-core mode, with one thread per subcore.
2401 if (cpu_has_feature(CPU_FTR_ARCH_300))
2402 return n_subcores <= 4 && n_threads == 1;
2404 /* On POWER8, can only dynamically split if unsplit to begin with */
2405 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2407 if (n_subcores > MAX_SUBCORES)
2409 if (n_subcores > 1) {
2410 if (!(dynamic_mt_modes & 2))
2412 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2416 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2419 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2421 vc->entry_exit_map = 0;
2423 vc->napping_threads = 0;
2424 vc->conferring_threads = 0;
2427 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2429 int n_threads = vc->num_threads;
2432 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2435 /* Some POWER9 chips require all threads to be in the same MMU mode */
2436 if (no_mixing_hpt_and_radix &&
2437 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2440 if (n_threads < cip->max_subcore_threads)
2441 n_threads = cip->max_subcore_threads;
2442 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2444 cip->max_subcore_threads = n_threads;
2446 sub = cip->n_subcores;
2448 cip->total_threads += vc->num_threads;
2449 cip->subcore_threads[sub] = vc->num_threads;
2451 init_vcore_to_run(vc);
2452 list_del_init(&vc->preempt_list);
2458 * Work out whether it is possible to piggyback the execution of
2459 * vcore *pvc onto the execution of the other vcores described in *cip.
2461 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2464 if (cip->total_threads + pvc->num_threads > target_threads)
2467 return can_dynamic_split(pvc, cip);
2470 static void prepare_threads(struct kvmppc_vcore *vc)
2473 struct kvm_vcpu *vcpu;
2475 for_each_runnable_thread(i, vcpu, vc) {
2476 if (signal_pending(vcpu->arch.run_task))
2477 vcpu->arch.ret = -EINTR;
2478 else if (vcpu->arch.vpa.update_pending ||
2479 vcpu->arch.slb_shadow.update_pending ||
2480 vcpu->arch.dtl.update_pending)
2481 vcpu->arch.ret = RESUME_GUEST;
2484 kvmppc_remove_runnable(vc, vcpu);
2485 wake_up(&vcpu->arch.cpu_run);
2489 static void collect_piggybacks(struct core_info *cip, int target_threads)
2491 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2492 struct kvmppc_vcore *pvc, *vcnext;
2494 spin_lock(&lp->lock);
2495 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2496 if (!spin_trylock(&pvc->lock))
2498 prepare_threads(pvc);
2499 if (!pvc->n_runnable) {
2500 list_del_init(&pvc->preempt_list);
2501 if (pvc->runner == NULL) {
2502 pvc->vcore_state = VCORE_INACTIVE;
2503 kvmppc_core_end_stolen(pvc);
2505 spin_unlock(&pvc->lock);
2508 if (!can_piggyback(pvc, cip, target_threads)) {
2509 spin_unlock(&pvc->lock);
2512 kvmppc_core_end_stolen(pvc);
2513 pvc->vcore_state = VCORE_PIGGYBACK;
2514 if (cip->total_threads >= target_threads)
2517 spin_unlock(&lp->lock);
2520 static bool recheck_signals(struct core_info *cip)
2523 struct kvm_vcpu *vcpu;
2525 for (sub = 0; sub < cip->n_subcores; ++sub)
2526 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2527 if (signal_pending(vcpu->arch.run_task))
2532 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2534 int still_running = 0, i;
2537 struct kvm_vcpu *vcpu;
2539 spin_lock(&vc->lock);
2541 for_each_runnable_thread(i, vcpu, vc) {
2542 /* cancel pending dec exception if dec is positive */
2543 if (now < vcpu->arch.dec_expires &&
2544 kvmppc_core_pending_dec(vcpu))
2545 kvmppc_core_dequeue_dec(vcpu);
2547 trace_kvm_guest_exit(vcpu);
2550 if (vcpu->arch.trap)
2551 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2552 vcpu->arch.run_task);
2554 vcpu->arch.ret = ret;
2555 vcpu->arch.trap = 0;
2557 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2558 if (vcpu->arch.pending_exceptions)
2559 kvmppc_core_prepare_to_enter(vcpu);
2560 if (vcpu->arch.ceded)
2561 kvmppc_set_timer(vcpu);
2565 kvmppc_remove_runnable(vc, vcpu);
2566 wake_up(&vcpu->arch.cpu_run);
2570 if (still_running > 0) {
2571 kvmppc_vcore_preempt(vc);
2572 } else if (vc->runner) {
2573 vc->vcore_state = VCORE_PREEMPT;
2574 kvmppc_core_start_stolen(vc);
2576 vc->vcore_state = VCORE_INACTIVE;
2578 if (vc->n_runnable > 0 && vc->runner == NULL) {
2579 /* make sure there's a candidate runner awake */
2581 vcpu = next_runnable_thread(vc, &i);
2582 wake_up(&vcpu->arch.cpu_run);
2585 spin_unlock(&vc->lock);
2589 * Clear core from the list of active host cores as we are about to
2590 * enter the guest. Only do this if it is the primary thread of the
2591 * core (not if a subcore) that is entering the guest.
2593 static inline int kvmppc_clear_host_core(unsigned int cpu)
2597 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2600 * Memory barrier can be omitted here as we will do a smp_wmb()
2601 * later in kvmppc_start_thread and we need ensure that state is
2602 * visible to other CPUs only after we enter guest.
2604 core = cpu >> threads_shift;
2605 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2610 * Advertise this core as an active host core since we exited the guest
2611 * Only need to do this if it is the primary thread of the core that is
2614 static inline int kvmppc_set_host_core(unsigned int cpu)
2618 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2622 * Memory barrier can be omitted here because we do a spin_unlock
2623 * immediately after this which provides the memory barrier.
2625 core = cpu >> threads_shift;
2626 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2630 static void set_irq_happened(int trap)
2633 case BOOK3S_INTERRUPT_EXTERNAL:
2634 local_paca->irq_happened |= PACA_IRQ_EE;
2636 case BOOK3S_INTERRUPT_H_DOORBELL:
2637 local_paca->irq_happened |= PACA_IRQ_DBELL;
2639 case BOOK3S_INTERRUPT_HMI:
2640 local_paca->irq_happened |= PACA_IRQ_HMI;
2642 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2643 replay_system_reset();
2649 * Run a set of guest threads on a physical core.
2650 * Called with vc->lock held.
2652 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2654 struct kvm_vcpu *vcpu;
2657 struct core_info core_info;
2658 struct kvmppc_vcore *pvc;
2659 struct kvm_split_mode split_info, *sip;
2660 int split, subcore_size, active;
2663 unsigned long cmd_bit, stat_bit;
2666 int controlled_threads;
2672 * Remove from the list any threads that have a signal pending
2673 * or need a VPA update done
2675 prepare_threads(vc);
2677 /* if the runner is no longer runnable, let the caller pick a new one */
2678 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2684 init_vcore_to_run(vc);
2685 vc->preempt_tb = TB_NIL;
2688 * Number of threads that we will be controlling: the same as
2689 * the number of threads per subcore, except on POWER9,
2690 * where it's 1 because the threads are (mostly) independent.
2692 controlled_threads = threads_per_vcore(vc->kvm);
2695 * Make sure we are running on primary threads, and that secondary
2696 * threads are offline. Also check if the number of threads in this
2697 * guest are greater than the current system threads per guest.
2698 * On POWER9, we need to be not in independent-threads mode if
2699 * this is a HPT guest on a radix host machine where the
2700 * CPU threads may not be in different MMU modes.
2702 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
2703 !kvm_is_radix(vc->kvm);
2704 if (((controlled_threads > 1) &&
2705 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2706 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2707 for_each_runnable_thread(i, vcpu, vc) {
2708 vcpu->arch.ret = -EBUSY;
2709 kvmppc_remove_runnable(vc, vcpu);
2710 wake_up(&vcpu->arch.cpu_run);
2716 * See if we could run any other vcores on the physical core
2717 * along with this one.
2719 init_core_info(&core_info, vc);
2720 pcpu = smp_processor_id();
2721 target_threads = controlled_threads;
2722 if (target_smt_mode && target_smt_mode < target_threads)
2723 target_threads = target_smt_mode;
2724 if (vc->num_threads < target_threads)
2725 collect_piggybacks(&core_info, target_threads);
2728 * On radix, arrange for TLB flushing if necessary.
2729 * This has to be done before disabling interrupts since
2730 * it uses smp_call_function().
2732 pcpu = smp_processor_id();
2733 if (kvm_is_radix(vc->kvm)) {
2734 for (sub = 0; sub < core_info.n_subcores; ++sub)
2735 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2736 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2740 * Hard-disable interrupts, and check resched flag and signals.
2741 * If we need to reschedule or deliver a signal, clean up
2742 * and return without going into the guest(s).
2743 * If the mmu_ready flag has been cleared, don't go into the
2744 * guest because that means a HPT resize operation is in progress.
2746 local_irq_disable();
2748 if (lazy_irq_pending() || need_resched() ||
2749 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
2751 vc->vcore_state = VCORE_INACTIVE;
2752 /* Unlock all except the primary vcore */
2753 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2754 pvc = core_info.vc[sub];
2755 /* Put back on to the preempted vcores list */
2756 kvmppc_vcore_preempt(pvc);
2757 spin_unlock(&pvc->lock);
2759 for (i = 0; i < controlled_threads; ++i)
2760 kvmppc_release_hwthread(pcpu + i);
2764 kvmppc_clear_host_core(pcpu);
2766 /* Decide on micro-threading (split-core) mode */
2767 subcore_size = threads_per_subcore;
2768 cmd_bit = stat_bit = 0;
2769 split = core_info.n_subcores;
2771 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2772 && !cpu_has_feature(CPU_FTR_ARCH_300);
2774 if (split > 1 || hpt_on_radix) {
2776 memset(&split_info, 0, sizeof(split_info));
2777 for (sub = 0; sub < core_info.n_subcores; ++sub)
2778 split_info.vc[sub] = core_info.vc[sub];
2781 if (split == 2 && (dynamic_mt_modes & 2)) {
2782 cmd_bit = HID0_POWER8_1TO2LPAR;
2783 stat_bit = HID0_POWER8_2LPARMODE;
2786 cmd_bit = HID0_POWER8_1TO4LPAR;
2787 stat_bit = HID0_POWER8_4LPARMODE;
2789 subcore_size = MAX_SMT_THREADS / split;
2790 split_info.rpr = mfspr(SPRN_RPR);
2791 split_info.pmmar = mfspr(SPRN_PMMAR);
2792 split_info.ldbar = mfspr(SPRN_LDBAR);
2793 split_info.subcore_size = subcore_size;
2795 split_info.subcore_size = 1;
2797 /* Use the split_info for LPCR/LPIDR changes */
2798 split_info.lpcr_req = vc->lpcr;
2799 split_info.lpidr_req = vc->kvm->arch.lpid;
2800 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2801 split_info.do_set = 1;
2805 /* order writes to split_info before kvm_split_mode pointer */
2809 for (thr = 0; thr < controlled_threads; ++thr) {
2810 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2812 paca->kvm_hstate.tid = thr;
2813 paca->kvm_hstate.napping = 0;
2814 paca->kvm_hstate.kvm_split_mode = sip;
2817 /* Initiate micro-threading (split-core) on POWER8 if required */
2819 unsigned long hid0 = mfspr(SPRN_HID0);
2821 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2823 mtspr(SPRN_HID0, hid0);
2826 hid0 = mfspr(SPRN_HID0);
2827 if (hid0 & stat_bit)
2833 /* Start all the threads */
2835 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2836 thr = is_power8 ? subcore_thread_map[sub] : sub;
2839 pvc = core_info.vc[sub];
2840 pvc->pcpu = pcpu + thr;
2841 for_each_runnable_thread(i, vcpu, pvc) {
2842 kvmppc_start_thread(vcpu, pvc);
2843 kvmppc_create_dtl_entry(vcpu, pvc);
2844 trace_kvm_guest_enter(vcpu);
2845 if (!vcpu->arch.ptid)
2847 active |= 1 << (thr + vcpu->arch.ptid);
2850 * We need to start the first thread of each subcore
2851 * even if it doesn't have a vcpu.
2854 kvmppc_start_thread(NULL, pvc);
2858 * Ensure that split_info.do_nap is set after setting
2859 * the vcore pointer in the PACA of the secondaries.
2864 * When doing micro-threading, poke the inactive threads as well.
2865 * This gets them to the nap instruction after kvm_do_nap,
2866 * which reduces the time taken to unsplit later.
2867 * For POWER9 HPT guest on radix host, we need all the secondary
2868 * threads woken up so they can do the LPCR/LPIDR change.
2870 if (cmd_bit || hpt_on_radix) {
2871 split_info.do_nap = 1; /* ask secondaries to nap when done */
2872 for (thr = 1; thr < threads_per_subcore; ++thr)
2873 if (!(active & (1 << thr)))
2874 kvmppc_ipi_thread(pcpu + thr);
2877 vc->vcore_state = VCORE_RUNNING;
2880 trace_kvmppc_run_core(vc, 0);
2882 for (sub = 0; sub < core_info.n_subcores; ++sub)
2883 spin_unlock(&core_info.vc[sub]->lock);
2886 * Interrupts will be enabled once we get into the guest,
2887 * so tell lockdep that we're about to enable interrupts.
2889 trace_hardirqs_on();
2893 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2895 trap = __kvmppc_vcore_entry();
2897 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2901 trace_hardirqs_off();
2902 set_irq_happened(trap);
2904 spin_lock(&vc->lock);
2905 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2906 vc->vcore_state = VCORE_EXITING;
2908 /* wait for secondary threads to finish writing their state to memory */
2909 kvmppc_wait_for_nap(controlled_threads);
2911 /* Return to whole-core mode if we split the core earlier */
2913 unsigned long hid0 = mfspr(SPRN_HID0);
2914 unsigned long loops = 0;
2916 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2917 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2919 mtspr(SPRN_HID0, hid0);
2922 hid0 = mfspr(SPRN_HID0);
2923 if (!(hid0 & stat_bit))
2928 } else if (hpt_on_radix) {
2929 /* Wait for all threads to have seen final sync */
2930 for (thr = 1; thr < controlled_threads; ++thr) {
2931 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2933 while (paca->kvm_hstate.kvm_split_mode) {
2940 split_info.do_nap = 0;
2942 kvmppc_set_host_core(pcpu);
2946 /* Let secondaries go back to the offline loop */
2947 for (i = 0; i < controlled_threads; ++i) {
2948 kvmppc_release_hwthread(pcpu + i);
2949 if (sip && sip->napped[i])
2950 kvmppc_ipi_thread(pcpu + i);
2951 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2954 spin_unlock(&vc->lock);
2956 /* make sure updates to secondary vcpu structs are visible now */
2961 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2962 pvc = core_info.vc[sub];
2963 post_guest_process(pvc, pvc == vc);
2966 spin_lock(&vc->lock);
2969 vc->vcore_state = VCORE_INACTIVE;
2970 trace_kvmppc_run_core(vc, 1);
2974 * Wait for some other vcpu thread to execute us, and
2975 * wake us up when we need to handle something in the host.
2977 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2978 struct kvm_vcpu *vcpu, int wait_state)
2982 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2983 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2984 spin_unlock(&vc->lock);
2986 spin_lock(&vc->lock);
2988 finish_wait(&vcpu->arch.cpu_run, &wait);
2991 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2994 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2995 vc->halt_poll_ns = 10000;
2997 vc->halt_poll_ns *= halt_poll_ns_grow;
3000 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3002 if (halt_poll_ns_shrink == 0)
3003 vc->halt_poll_ns = 0;
3005 vc->halt_poll_ns /= halt_poll_ns_shrink;
3008 #ifdef CONFIG_KVM_XICS
3009 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3011 if (!xive_enabled())
3013 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3014 vcpu->arch.xive_saved_state.cppr;
3017 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3021 #endif /* CONFIG_KVM_XICS */
3023 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3025 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3026 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3033 * Check to see if any of the runnable vcpus on the vcore have pending
3034 * exceptions or are no longer ceded
3036 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3038 struct kvm_vcpu *vcpu;
3041 for_each_runnable_thread(i, vcpu, vc) {
3042 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3050 * All the vcpus in this vcore are idle, so wait for a decrementer
3051 * or external interrupt to one of the vcpus. vc->lock is held.
3053 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3055 ktime_t cur, start_poll, start_wait;
3058 DECLARE_SWAITQUEUE(wait);
3060 /* Poll for pending exceptions and ceded state */
3061 cur = start_poll = ktime_get();
3062 if (vc->halt_poll_ns) {
3063 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3064 ++vc->runner->stat.halt_attempted_poll;
3066 vc->vcore_state = VCORE_POLLING;
3067 spin_unlock(&vc->lock);
3070 if (kvmppc_vcore_check_block(vc)) {
3075 } while (single_task_running() && ktime_before(cur, stop));
3077 spin_lock(&vc->lock);
3078 vc->vcore_state = VCORE_INACTIVE;
3081 ++vc->runner->stat.halt_successful_poll;
3086 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3088 if (kvmppc_vcore_check_block(vc)) {
3089 finish_swait(&vc->wq, &wait);
3091 /* If we polled, count this as a successful poll */
3092 if (vc->halt_poll_ns)
3093 ++vc->runner->stat.halt_successful_poll;
3097 start_wait = ktime_get();
3099 vc->vcore_state = VCORE_SLEEPING;
3100 trace_kvmppc_vcore_blocked(vc, 0);
3101 spin_unlock(&vc->lock);
3103 finish_swait(&vc->wq, &wait);
3104 spin_lock(&vc->lock);
3105 vc->vcore_state = VCORE_INACTIVE;
3106 trace_kvmppc_vcore_blocked(vc, 1);
3107 ++vc->runner->stat.halt_successful_wait;
3112 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3114 /* Attribute wait time */
3116 vc->runner->stat.halt_wait_ns +=
3117 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3118 /* Attribute failed poll time */
3119 if (vc->halt_poll_ns)
3120 vc->runner->stat.halt_poll_fail_ns +=
3121 ktime_to_ns(start_wait) -
3122 ktime_to_ns(start_poll);
3124 /* Attribute successful poll time */
3125 if (vc->halt_poll_ns)
3126 vc->runner->stat.halt_poll_success_ns +=
3128 ktime_to_ns(start_poll);
3131 /* Adjust poll time */
3133 if (block_ns <= vc->halt_poll_ns)
3135 /* We slept and blocked for longer than the max halt time */
3136 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3137 shrink_halt_poll_ns(vc);
3138 /* We slept and our poll time is too small */
3139 else if (vc->halt_poll_ns < halt_poll_ns &&
3140 block_ns < halt_poll_ns)
3141 grow_halt_poll_ns(vc);
3142 if (vc->halt_poll_ns > halt_poll_ns)
3143 vc->halt_poll_ns = halt_poll_ns;
3145 vc->halt_poll_ns = 0;
3147 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3150 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3153 struct kvm *kvm = vcpu->kvm;
3155 mutex_lock(&kvm->lock);
3156 if (!kvm->arch.mmu_ready) {
3157 if (!kvm_is_radix(kvm))
3158 r = kvmppc_hv_setup_htab_rma(vcpu);
3160 if (cpu_has_feature(CPU_FTR_ARCH_300))
3161 kvmppc_setup_partition_table(kvm);
3162 kvm->arch.mmu_ready = 1;
3165 mutex_unlock(&kvm->lock);
3169 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3172 struct kvmppc_vcore *vc;
3175 trace_kvmppc_run_vcpu_enter(vcpu);
3177 kvm_run->exit_reason = 0;
3178 vcpu->arch.ret = RESUME_GUEST;
3179 vcpu->arch.trap = 0;
3180 kvmppc_update_vpas(vcpu);
3183 * Synchronize with other threads in this virtual core
3185 vc = vcpu->arch.vcore;
3186 spin_lock(&vc->lock);
3187 vcpu->arch.ceded = 0;
3188 vcpu->arch.run_task = current;
3189 vcpu->arch.kvm_run = kvm_run;
3190 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3191 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3192 vcpu->arch.busy_preempt = TB_NIL;
3193 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3197 * This happens the first time this is called for a vcpu.
3198 * If the vcore is already running, we may be able to start
3199 * this thread straight away and have it join in.
3201 if (!signal_pending(current)) {
3202 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3203 vc->vcore_state == VCORE_RUNNING) &&
3204 !VCORE_IS_EXITING(vc)) {
3205 kvmppc_create_dtl_entry(vcpu, vc);
3206 kvmppc_start_thread(vcpu, vc);
3207 trace_kvm_guest_enter(vcpu);
3208 } else if (vc->vcore_state == VCORE_SLEEPING) {
3214 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3215 !signal_pending(current)) {
3216 /* See if the MMU is ready to go */
3217 if (!vcpu->kvm->arch.mmu_ready) {
3218 spin_unlock(&vc->lock);
3219 r = kvmhv_setup_mmu(vcpu);
3220 spin_lock(&vc->lock);
3222 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3223 kvm_run->fail_entry.
3224 hardware_entry_failure_reason = 0;
3230 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3231 kvmppc_vcore_end_preempt(vc);
3233 if (vc->vcore_state != VCORE_INACTIVE) {
3234 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3237 for_each_runnable_thread(i, v, vc) {
3238 kvmppc_core_prepare_to_enter(v);
3239 if (signal_pending(v->arch.run_task)) {
3240 kvmppc_remove_runnable(vc, v);
3241 v->stat.signal_exits++;
3242 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3243 v->arch.ret = -EINTR;
3244 wake_up(&v->arch.cpu_run);
3247 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3250 for_each_runnable_thread(i, v, vc) {
3251 if (!kvmppc_vcpu_woken(v))
3252 n_ceded += v->arch.ceded;
3257 if (n_ceded == vc->n_runnable) {
3258 kvmppc_vcore_blocked(vc);
3259 } else if (need_resched()) {
3260 kvmppc_vcore_preempt(vc);
3261 /* Let something else run */
3262 cond_resched_lock(&vc->lock);
3263 if (vc->vcore_state == VCORE_PREEMPT)
3264 kvmppc_vcore_end_preempt(vc);
3266 kvmppc_run_core(vc);
3271 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3272 (vc->vcore_state == VCORE_RUNNING ||
3273 vc->vcore_state == VCORE_EXITING ||
3274 vc->vcore_state == VCORE_PIGGYBACK))
3275 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3277 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3278 kvmppc_vcore_end_preempt(vc);
3280 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3281 kvmppc_remove_runnable(vc, vcpu);
3282 vcpu->stat.signal_exits++;
3283 kvm_run->exit_reason = KVM_EXIT_INTR;
3284 vcpu->arch.ret = -EINTR;
3287 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3288 /* Wake up some vcpu to run the core */
3290 v = next_runnable_thread(vc, &i);
3291 wake_up(&v->arch.cpu_run);
3294 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3295 spin_unlock(&vc->lock);
3296 return vcpu->arch.ret;
3299 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3303 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3304 unsigned long user_tar = 0;
3305 unsigned int user_vrsave;
3308 if (!vcpu->arch.sane) {
3309 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3314 * Don't allow entry with a suspended transaction, because
3315 * the guest entry/exit code will lose it.
3316 * If the guest has TM enabled, save away their TM-related SPRs
3317 * (they will get restored by the TM unavailable interrupt).
3319 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3320 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3321 (current->thread.regs->msr & MSR_TM)) {
3322 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3323 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3324 run->fail_entry.hardware_entry_failure_reason = 0;
3327 /* Enable TM so we can read the TM SPRs */
3328 mtmsr(mfmsr() | MSR_TM);
3329 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3330 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3331 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3332 current->thread.regs->msr &= ~MSR_TM;
3336 kvmppc_core_prepare_to_enter(vcpu);
3338 /* No need to go into the guest when all we'll do is come back out */
3339 if (signal_pending(current)) {
3340 run->exit_reason = KVM_EXIT_INTR;
3345 atomic_inc(&kvm->arch.vcpus_running);
3346 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3349 flush_all_to_thread(current);
3351 /* Save userspace EBB and other register values */
3352 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3353 ebb_regs[0] = mfspr(SPRN_EBBHR);
3354 ebb_regs[1] = mfspr(SPRN_EBBRR);
3355 ebb_regs[2] = mfspr(SPRN_BESCR);
3356 user_tar = mfspr(SPRN_TAR);
3358 user_vrsave = mfspr(SPRN_VRSAVE);
3360 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3361 vcpu->arch.pgdir = current->mm->pgd;
3362 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3365 r = kvmppc_run_vcpu(run, vcpu);
3367 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3368 !(vcpu->arch.shregs.msr & MSR_PR)) {
3369 trace_kvm_hcall_enter(vcpu);
3370 r = kvmppc_pseries_do_hcall(vcpu);
3371 trace_kvm_hcall_exit(vcpu, r);
3372 kvmppc_core_prepare_to_enter(vcpu);
3373 } else if (r == RESUME_PAGE_FAULT) {
3374 srcu_idx = srcu_read_lock(&kvm->srcu);
3375 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3376 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3377 srcu_read_unlock(&kvm->srcu, srcu_idx);
3378 } else if (r == RESUME_PASSTHROUGH) {
3379 if (WARN_ON(xive_enabled()))
3382 r = kvmppc_xics_rm_complete(vcpu, 0);
3384 } while (is_kvmppc_resume_guest(r));
3386 /* Restore userspace EBB and other register values */
3387 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3388 mtspr(SPRN_EBBHR, ebb_regs[0]);
3389 mtspr(SPRN_EBBRR, ebb_regs[1]);
3390 mtspr(SPRN_BESCR, ebb_regs[2]);
3391 mtspr(SPRN_TAR, user_tar);
3392 mtspr(SPRN_FSCR, current->thread.fscr);
3394 mtspr(SPRN_VRSAVE, user_vrsave);
3396 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3397 atomic_dec(&kvm->arch.vcpus_running);
3401 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3402 int shift, int sllp)
3404 (*sps)->page_shift = shift;
3405 (*sps)->slb_enc = sllp;
3406 (*sps)->enc[0].page_shift = shift;
3407 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3409 * Add 16MB MPSS support (may get filtered out by userspace)
3412 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3414 (*sps)->enc[1].page_shift = 24;
3415 (*sps)->enc[1].pte_enc = penc;
3421 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3422 struct kvm_ppc_smmu_info *info)
3424 struct kvm_ppc_one_seg_page_size *sps;
3427 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3428 * POWER7 doesn't support keys for instruction accesses,
3429 * POWER8 and POWER9 do.
3431 info->data_keys = 32;
3432 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3434 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3435 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3436 info->slb_size = 32;
3438 /* We only support these sizes for now, and no muti-size segments */
3439 sps = &info->sps[0];
3440 kvmppc_add_seg_page_size(&sps, 12, 0);
3441 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3442 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3448 * Get (and clear) the dirty memory log for a memory slot.
3450 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3451 struct kvm_dirty_log *log)
3453 struct kvm_memslots *slots;
3454 struct kvm_memory_slot *memslot;
3457 unsigned long *buf, *p;
3458 struct kvm_vcpu *vcpu;
3460 mutex_lock(&kvm->slots_lock);
3463 if (log->slot >= KVM_USER_MEM_SLOTS)
3466 slots = kvm_memslots(kvm);
3467 memslot = id_to_memslot(slots, log->slot);
3469 if (!memslot->dirty_bitmap)
3473 * Use second half of bitmap area because both HPT and radix
3474 * accumulate bits in the first half.
3476 n = kvm_dirty_bitmap_bytes(memslot);
3477 buf = memslot->dirty_bitmap + n / sizeof(long);
3480 if (kvm_is_radix(kvm))
3481 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3483 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3488 * We accumulate dirty bits in the first half of the
3489 * memslot's dirty_bitmap area, for when pages are paged
3490 * out or modified by the host directly. Pick up these
3491 * bits and add them to the map.
3493 p = memslot->dirty_bitmap;
3494 for (i = 0; i < n / sizeof(long); ++i)
3495 buf[i] |= xchg(&p[i], 0);
3497 /* Harvest dirty bits from VPA and DTL updates */
3498 /* Note: we never modify the SLB shadow buffer areas */
3499 kvm_for_each_vcpu(i, vcpu, kvm) {
3500 spin_lock(&vcpu->arch.vpa_update_lock);
3501 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3502 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3503 spin_unlock(&vcpu->arch.vpa_update_lock);
3507 if (copy_to_user(log->dirty_bitmap, buf, n))
3512 mutex_unlock(&kvm->slots_lock);
3516 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3517 struct kvm_memory_slot *dont)
3519 if (!dont || free->arch.rmap != dont->arch.rmap) {
3520 vfree(free->arch.rmap);
3521 free->arch.rmap = NULL;
3525 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3526 unsigned long npages)
3528 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3529 if (!slot->arch.rmap)
3535 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3536 struct kvm_memory_slot *memslot,
3537 const struct kvm_userspace_memory_region *mem)
3542 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3543 const struct kvm_userspace_memory_region *mem,
3544 const struct kvm_memory_slot *old,
3545 const struct kvm_memory_slot *new)
3547 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3550 * If we are making a new memslot, it might make
3551 * some address that was previously cached as emulated
3552 * MMIO be no longer emulated MMIO, so invalidate
3553 * all the caches of emulated MMIO translations.
3556 atomic64_inc(&kvm->arch.mmio_update);
3560 * Update LPCR values in kvm->arch and in vcores.
3561 * Caller must hold kvm->lock.
3563 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3568 if ((kvm->arch.lpcr & mask) == lpcr)
3571 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3573 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3574 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3577 spin_lock(&vc->lock);
3578 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3579 spin_unlock(&vc->lock);
3580 if (++cores_done >= kvm->arch.online_vcores)
3585 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3590 void kvmppc_setup_partition_table(struct kvm *kvm)
3592 unsigned long dw0, dw1;
3594 if (!kvm_is_radix(kvm)) {
3595 /* PS field - page size for VRMA */
3596 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3597 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3598 /* HTABSIZE and HTABORG fields */
3599 dw0 |= kvm->arch.sdr1;
3601 /* Second dword as set by userspace */
3602 dw1 = kvm->arch.process_table;
3604 dw0 = PATB_HR | radix__get_tree_size() |
3605 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3606 dw1 = PATB_GR | kvm->arch.process_table;
3609 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3613 * Set up HPT (hashed page table) and RMA (real-mode area).
3614 * Must be called with kvm->lock held.
3616 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3619 struct kvm *kvm = vcpu->kvm;
3621 struct kvm_memory_slot *memslot;
3622 struct vm_area_struct *vma;
3623 unsigned long lpcr = 0, senc;
3624 unsigned long psize, porder;
3627 /* Allocate hashed page table (if not done already) and reset it */
3628 if (!kvm->arch.hpt.virt) {
3629 int order = KVM_DEFAULT_HPT_ORDER;
3630 struct kvm_hpt_info info;
3632 err = kvmppc_allocate_hpt(&info, order);
3633 /* If we get here, it means userspace didn't specify a
3634 * size explicitly. So, try successively smaller
3635 * sizes if the default failed. */
3636 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3637 err = kvmppc_allocate_hpt(&info, order);
3640 pr_err("KVM: Couldn't alloc HPT\n");
3644 kvmppc_set_hpt(kvm, &info);
3647 /* Look up the memslot for guest physical address 0 */
3648 srcu_idx = srcu_read_lock(&kvm->srcu);
3649 memslot = gfn_to_memslot(kvm, 0);
3651 /* We must have some memory at 0 by now */
3653 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3656 /* Look up the VMA for the start of this memory slot */
3657 hva = memslot->userspace_addr;
3658 down_read(¤t->mm->mmap_sem);
3659 vma = find_vma(current->mm, hva);
3660 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3663 psize = vma_kernel_pagesize(vma);
3664 porder = __ilog2(psize);
3666 up_read(¤t->mm->mmap_sem);
3668 /* We can handle 4k, 64k or 16M pages in the VRMA */
3670 if (!(psize == 0x1000 || psize == 0x10000 ||
3671 psize == 0x1000000))
3674 senc = slb_pgsize_encoding(psize);
3675 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3676 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3677 /* Create HPTEs in the hash page table for the VRMA */
3678 kvmppc_map_vrma(vcpu, memslot, porder);
3680 /* Update VRMASD field in the LPCR */
3681 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3682 /* the -4 is to account for senc values starting at 0x10 */
3683 lpcr = senc << (LPCR_VRMASD_SH - 4);
3684 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3687 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3691 srcu_read_unlock(&kvm->srcu, srcu_idx);
3696 up_read(¤t->mm->mmap_sem);
3700 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3701 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3703 kvmppc_free_radix(kvm);
3704 kvmppc_update_lpcr(kvm, LPCR_VPM1,
3705 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3706 kvmppc_rmap_reset(kvm);
3707 kvm->arch.radix = 0;
3708 kvm->arch.process_table = 0;
3712 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3713 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3717 err = kvmppc_init_vm_radix(kvm);
3721 kvmppc_free_hpt(&kvm->arch.hpt);
3722 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3723 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3724 kvm->arch.radix = 1;
3728 #ifdef CONFIG_KVM_XICS
3730 * Allocate a per-core structure for managing state about which cores are
3731 * running in the host versus the guest and for exchanging data between
3732 * real mode KVM and CPU running in the host.
3733 * This is only done for the first VM.
3734 * The allocated structure stays even if all VMs have stopped.
3735 * It is only freed when the kvm-hv module is unloaded.
3736 * It's OK for this routine to fail, we just don't support host
3737 * core operations like redirecting H_IPI wakeups.
3739 void kvmppc_alloc_host_rm_ops(void)
3741 struct kvmppc_host_rm_ops *ops;
3742 unsigned long l_ops;
3746 /* Not the first time here ? */
3747 if (kvmppc_host_rm_ops_hv != NULL)
3750 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3754 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3755 ops->rm_core = kzalloc(size, GFP_KERNEL);
3757 if (!ops->rm_core) {
3764 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3765 if (!cpu_online(cpu))
3768 core = cpu >> threads_shift;
3769 ops->rm_core[core].rm_state.in_host = 1;
3772 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3775 * Make the contents of the kvmppc_host_rm_ops structure visible
3776 * to other CPUs before we assign it to the global variable.
3777 * Do an atomic assignment (no locks used here), but if someone
3778 * beats us to it, just free our copy and return.
3781 l_ops = (unsigned long) ops;
3783 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3785 kfree(ops->rm_core);
3790 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3791 "ppc/kvm_book3s:prepare",
3792 kvmppc_set_host_core,
3793 kvmppc_clear_host_core);
3797 void kvmppc_free_host_rm_ops(void)
3799 if (kvmppc_host_rm_ops_hv) {
3800 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3801 kfree(kvmppc_host_rm_ops_hv->rm_core);
3802 kfree(kvmppc_host_rm_ops_hv);
3803 kvmppc_host_rm_ops_hv = NULL;
3808 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3810 unsigned long lpcr, lpid;
3814 /* Allocate the guest's logical partition ID */
3816 lpid = kvmppc_alloc_lpid();
3819 kvm->arch.lpid = lpid;
3821 kvmppc_alloc_host_rm_ops();
3824 * Since we don't flush the TLB when tearing down a VM,
3825 * and this lpid might have previously been used,
3826 * make sure we flush on each core before running the new VM.
3827 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3828 * does this flush for us.
3830 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3831 cpumask_setall(&kvm->arch.need_tlb_flush);
3833 /* Start out with the default set of hcalls enabled */
3834 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3835 sizeof(kvm->arch.enabled_hcalls));
3837 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3838 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3840 /* Init LPCR for virtual RMA mode */
3841 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3842 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3843 lpcr &= LPCR_PECE | LPCR_LPES;
3844 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3845 LPCR_VPM0 | LPCR_VPM1;
3846 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3847 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3848 /* On POWER8 turn on online bit to enable PURR/SPURR */
3849 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3852 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3853 * Set HVICE bit to enable hypervisor virtualization interrupts.
3854 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3855 * be unnecessary but better safe than sorry in case we re-enable
3856 * EE in HV mode with this LPCR still set)
3858 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3860 lpcr |= LPCR_HVICE | LPCR_HEIC;
3863 * If xive is enabled, we route 0x500 interrupts directly
3871 * If the host uses radix, the guest starts out as radix.
3873 if (radix_enabled()) {
3874 kvm->arch.radix = 1;
3875 kvm->arch.mmu_ready = 1;
3877 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3878 ret = kvmppc_init_vm_radix(kvm);
3880 kvmppc_free_lpid(kvm->arch.lpid);
3883 kvmppc_setup_partition_table(kvm);
3886 kvm->arch.lpcr = lpcr;
3888 /* Initialization for future HPT resizes */
3889 kvm->arch.resize_hpt = NULL;
3892 * Work out how many sets the TLB has, for the use of
3893 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3895 if (radix_enabled())
3896 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
3897 else if (cpu_has_feature(CPU_FTR_ARCH_300))
3898 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
3899 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3900 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
3902 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
3905 * Track that we now have a HV mode VM active. This blocks secondary
3906 * CPU threads from coming online.
3907 * On POWER9, we only need to do this if the "indep_threads_mode"
3908 * module parameter has been set to N.
3910 if (cpu_has_feature(CPU_FTR_ARCH_300))
3911 kvm->arch.threads_indep = indep_threads_mode;
3912 if (!kvm->arch.threads_indep)
3913 kvm_hv_vm_activated();
3916 * Initialize smt_mode depending on processor.
3917 * POWER8 and earlier have to use "strict" threading, where
3918 * all vCPUs in a vcore have to run on the same (sub)core,
3919 * whereas on POWER9 the threads can each run a different
3922 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3923 kvm->arch.smt_mode = threads_per_subcore;
3925 kvm->arch.smt_mode = 1;
3926 kvm->arch.emul_smt_mode = 1;
3929 * Create a debugfs directory for the VM
3931 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3932 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3933 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3934 kvmppc_mmu_debugfs_init(kvm);
3939 static void kvmppc_free_vcores(struct kvm *kvm)
3943 for (i = 0; i < KVM_MAX_VCORES; ++i)
3944 kfree(kvm->arch.vcores[i]);
3945 kvm->arch.online_vcores = 0;
3948 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3950 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3952 if (!kvm->arch.threads_indep)
3953 kvm_hv_vm_deactivated();
3955 kvmppc_free_vcores(kvm);
3957 kvmppc_free_lpid(kvm->arch.lpid);
3959 if (kvm_is_radix(kvm))
3960 kvmppc_free_radix(kvm);
3962 kvmppc_free_hpt(&kvm->arch.hpt);
3964 kvmppc_free_pimap(kvm);
3967 /* We don't need to emulate any privileged instructions or dcbz */
3968 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3969 unsigned int inst, int *advance)
3971 return EMULATE_FAIL;
3974 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3977 return EMULATE_FAIL;
3980 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3983 return EMULATE_FAIL;
3986 static int kvmppc_core_check_processor_compat_hv(void)
3988 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3989 !cpu_has_feature(CPU_FTR_ARCH_206))
3995 #ifdef CONFIG_KVM_XICS
3997 void kvmppc_free_pimap(struct kvm *kvm)
3999 kfree(kvm->arch.pimap);
4002 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4004 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4007 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4009 struct irq_desc *desc;
4010 struct kvmppc_irq_map *irq_map;
4011 struct kvmppc_passthru_irqmap *pimap;
4012 struct irq_chip *chip;
4015 if (!kvm_irq_bypass)
4018 desc = irq_to_desc(host_irq);
4022 mutex_lock(&kvm->lock);
4024 pimap = kvm->arch.pimap;
4025 if (pimap == NULL) {
4026 /* First call, allocate structure to hold IRQ map */
4027 pimap = kvmppc_alloc_pimap();
4028 if (pimap == NULL) {
4029 mutex_unlock(&kvm->lock);
4032 kvm->arch.pimap = pimap;
4036 * For now, we only support interrupts for which the EOI operation
4037 * is an OPAL call followed by a write to XIRR, since that's
4038 * what our real-mode EOI code does, or a XIVE interrupt
4040 chip = irq_data_get_irq_chip(&desc->irq_data);
4041 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4042 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4043 host_irq, guest_gsi);
4044 mutex_unlock(&kvm->lock);
4049 * See if we already have an entry for this guest IRQ number.
4050 * If it's mapped to a hardware IRQ number, that's an error,
4051 * otherwise re-use this entry.
4053 for (i = 0; i < pimap->n_mapped; i++) {
4054 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4055 if (pimap->mapped[i].r_hwirq) {
4056 mutex_unlock(&kvm->lock);
4063 if (i == KVMPPC_PIRQ_MAPPED) {
4064 mutex_unlock(&kvm->lock);
4065 return -EAGAIN; /* table is full */
4068 irq_map = &pimap->mapped[i];
4070 irq_map->v_hwirq = guest_gsi;
4071 irq_map->desc = desc;
4074 * Order the above two stores before the next to serialize with
4075 * the KVM real mode handler.
4078 irq_map->r_hwirq = desc->irq_data.hwirq;
4080 if (i == pimap->n_mapped)
4084 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4086 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4088 irq_map->r_hwirq = 0;
4090 mutex_unlock(&kvm->lock);
4095 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4097 struct irq_desc *desc;
4098 struct kvmppc_passthru_irqmap *pimap;
4101 if (!kvm_irq_bypass)
4104 desc = irq_to_desc(host_irq);
4108 mutex_lock(&kvm->lock);
4109 if (!kvm->arch.pimap)
4112 pimap = kvm->arch.pimap;
4114 for (i = 0; i < pimap->n_mapped; i++) {
4115 if (guest_gsi == pimap->mapped[i].v_hwirq)
4119 if (i == pimap->n_mapped) {
4120 mutex_unlock(&kvm->lock);
4125 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4127 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4129 /* invalidate the entry (what do do on error from the above ?) */
4130 pimap->mapped[i].r_hwirq = 0;
4133 * We don't free this structure even when the count goes to
4134 * zero. The structure is freed when we destroy the VM.
4137 mutex_unlock(&kvm->lock);
4141 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4142 struct irq_bypass_producer *prod)
4145 struct kvm_kernel_irqfd *irqfd =
4146 container_of(cons, struct kvm_kernel_irqfd, consumer);
4148 irqfd->producer = prod;
4150 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4152 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4153 prod->irq, irqfd->gsi, ret);
4158 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4159 struct irq_bypass_producer *prod)
4162 struct kvm_kernel_irqfd *irqfd =
4163 container_of(cons, struct kvm_kernel_irqfd, consumer);
4165 irqfd->producer = NULL;
4168 * When producer of consumer is unregistered, we change back to
4169 * default external interrupt handling mode - KVM real mode
4170 * will switch back to host.
4172 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4174 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4175 prod->irq, irqfd->gsi, ret);
4179 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4180 unsigned int ioctl, unsigned long arg)
4182 struct kvm *kvm __maybe_unused = filp->private_data;
4183 void __user *argp = (void __user *)arg;
4188 case KVM_PPC_ALLOCATE_HTAB: {
4192 if (get_user(htab_order, (u32 __user *)argp))
4194 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4201 case KVM_PPC_GET_HTAB_FD: {
4202 struct kvm_get_htab_fd ghf;
4205 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4207 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4211 case KVM_PPC_RESIZE_HPT_PREPARE: {
4212 struct kvm_ppc_resize_hpt rhpt;
4215 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4218 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4222 case KVM_PPC_RESIZE_HPT_COMMIT: {
4223 struct kvm_ppc_resize_hpt rhpt;
4226 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4229 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4241 * List of hcall numbers to enable by default.
4242 * For compatibility with old userspace, we enable by default
4243 * all hcalls that were implemented before the hcall-enabling
4244 * facility was added. Note this list should not include H_RTAS.
4246 static unsigned int default_hcall_list[] = {
4260 #ifdef CONFIG_KVM_XICS
4271 static void init_default_hcalls(void)
4276 for (i = 0; default_hcall_list[i]; ++i) {
4277 hcall = default_hcall_list[i];
4278 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4279 __set_bit(hcall / 4, default_enabled_hcalls);
4283 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4289 /* If not on a POWER9, reject it */
4290 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4293 /* If any unknown flags set, reject it */
4294 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4297 /* GR (guest radix) bit in process_table field must match */
4298 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4299 if (!!(cfg->process_table & PATB_GR) != radix)
4302 /* Process table size field must be reasonable, i.e. <= 24 */
4303 if ((cfg->process_table & PRTS_MASK) > 24)
4306 /* We can change a guest to/from radix now, if the host is radix */
4307 if (radix && !radix_enabled())
4310 mutex_lock(&kvm->lock);
4311 if (radix != kvm_is_radix(kvm)) {
4312 if (kvm->arch.mmu_ready) {
4313 kvm->arch.mmu_ready = 0;
4314 /* order mmu_ready vs. vcpus_running */
4316 if (atomic_read(&kvm->arch.vcpus_running)) {
4317 kvm->arch.mmu_ready = 1;
4323 err = kvmppc_switch_mmu_to_radix(kvm);
4325 err = kvmppc_switch_mmu_to_hpt(kvm);
4330 kvm->arch.process_table = cfg->process_table;
4331 kvmppc_setup_partition_table(kvm);
4333 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4334 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4338 mutex_unlock(&kvm->lock);
4342 static struct kvmppc_ops kvm_ops_hv = {
4343 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4344 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4345 .get_one_reg = kvmppc_get_one_reg_hv,
4346 .set_one_reg = kvmppc_set_one_reg_hv,
4347 .vcpu_load = kvmppc_core_vcpu_load_hv,
4348 .vcpu_put = kvmppc_core_vcpu_put_hv,
4349 .set_msr = kvmppc_set_msr_hv,
4350 .vcpu_run = kvmppc_vcpu_run_hv,
4351 .vcpu_create = kvmppc_core_vcpu_create_hv,
4352 .vcpu_free = kvmppc_core_vcpu_free_hv,
4353 .check_requests = kvmppc_core_check_requests_hv,
4354 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4355 .flush_memslot = kvmppc_core_flush_memslot_hv,
4356 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4357 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4358 .unmap_hva = kvm_unmap_hva_hv,
4359 .unmap_hva_range = kvm_unmap_hva_range_hv,
4360 .age_hva = kvm_age_hva_hv,
4361 .test_age_hva = kvm_test_age_hva_hv,
4362 .set_spte_hva = kvm_set_spte_hva_hv,
4363 .mmu_destroy = kvmppc_mmu_destroy_hv,
4364 .free_memslot = kvmppc_core_free_memslot_hv,
4365 .create_memslot = kvmppc_core_create_memslot_hv,
4366 .init_vm = kvmppc_core_init_vm_hv,
4367 .destroy_vm = kvmppc_core_destroy_vm_hv,
4368 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4369 .emulate_op = kvmppc_core_emulate_op_hv,
4370 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4371 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4372 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4373 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4374 .hcall_implemented = kvmppc_hcall_impl_hv,
4375 #ifdef CONFIG_KVM_XICS
4376 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4377 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4379 .configure_mmu = kvmhv_configure_mmu,
4380 .get_rmmu_info = kvmhv_get_rmmu_info,
4381 .set_smt_mode = kvmhv_set_smt_mode,
4384 static int kvm_init_subcore_bitmap(void)
4387 int nr_cores = cpu_nr_cores();
4388 struct sibling_subcore_state *sibling_subcore_state;
4390 for (i = 0; i < nr_cores; i++) {
4391 int first_cpu = i * threads_per_core;
4392 int node = cpu_to_node(first_cpu);
4394 /* Ignore if it is already allocated. */
4395 if (paca_ptrs[first_cpu]->sibling_subcore_state)
4398 sibling_subcore_state =
4399 kmalloc_node(sizeof(struct sibling_subcore_state),
4401 if (!sibling_subcore_state)
4404 memset(sibling_subcore_state, 0,
4405 sizeof(struct sibling_subcore_state));
4407 for (j = 0; j < threads_per_core; j++) {
4408 int cpu = first_cpu + j;
4410 paca_ptrs[cpu]->sibling_subcore_state =
4411 sibling_subcore_state;
4417 static int kvmppc_radix_possible(void)
4419 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4422 static int kvmppc_book3s_init_hv(void)
4426 * FIXME!! Do we need to check on all cpus ?
4428 r = kvmppc_core_check_processor_compat_hv();
4432 r = kvm_init_subcore_bitmap();
4437 * We need a way of accessing the XICS interrupt controller,
4438 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
4439 * indirectly, via OPAL.
4442 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4443 struct device_node *np;
4445 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4447 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4453 kvm_ops_hv.owner = THIS_MODULE;
4454 kvmppc_hv_ops = &kvm_ops_hv;
4456 init_default_hcalls();
4460 r = kvmppc_mmu_hv_init();
4464 if (kvmppc_radix_possible())
4465 r = kvmppc_radix_init();
4468 * POWER9 chips before version 2.02 can't have some threads in
4469 * HPT mode and some in radix mode on the same core.
4471 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4472 unsigned int pvr = mfspr(SPRN_PVR);
4473 if ((pvr >> 16) == PVR_POWER9 &&
4474 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
4475 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
4476 no_mixing_hpt_and_radix = true;
4482 static void kvmppc_book3s_exit_hv(void)
4484 kvmppc_free_host_rm_ops();
4485 if (kvmppc_radix_possible())
4486 kvmppc_radix_exit();
4487 kvmppc_hv_ops = NULL;
4490 module_init(kvmppc_book3s_init_hv);
4491 module_exit(kvmppc_book3s_exit_hv);
4492 MODULE_LICENSE("GPL");
4493 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4494 MODULE_ALIAS("devname:kvm");