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/debug.h>
53 #include <asm/disassemble.h>
54 #include <asm/cputable.h>
55 #include <asm/cacheflush.h>
56 #include <asm/tlbflush.h>
57 #include <linux/uaccess.h>
59 #include <asm/kvm_ppc.h>
60 #include <asm/kvm_book3s.h>
61 #include <asm/mmu_context.h>
62 #include <asm/lppaca.h>
63 #include <asm/processor.h>
64 #include <asm/cputhreads.h>
66 #include <asm/hvcall.h>
67 #include <asm/switch_to.h>
69 #include <asm/dbell.h>
71 #include <asm/pnv-pci.h>
79 #define CREATE_TRACE_POINTS
82 /* #define EXIT_DEBUG */
83 /* #define EXIT_DEBUG_SIMPLE */
84 /* #define EXIT_DEBUG_INT */
86 /* Used to indicate that a guest page fault needs to be handled */
87 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
88 /* Used to indicate that a guest passthrough interrupt needs to be handled */
89 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
91 /* Used as a "null" value for timebase values */
92 #define TB_NIL (~(u64)0)
94 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
96 static int dynamic_mt_modes = 6;
97 module_param(dynamic_mt_modes, int, 0644);
98 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
99 static int target_smt_mode;
100 module_param(target_smt_mode, int, 0644);
101 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
103 static bool indep_threads_mode = true;
104 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
105 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
107 #ifdef CONFIG_KVM_XICS
108 static struct kernel_param_ops module_param_ops = {
109 .set = param_set_int,
110 .get = param_get_int,
113 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
114 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
116 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
117 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
120 /* If set, the threads on each CPU core have to be in the same MMU mode */
121 static bool no_mixing_hpt_and_radix;
123 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
124 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
126 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
130 struct kvm_vcpu *vcpu;
132 while (++i < MAX_SMT_THREADS) {
133 vcpu = READ_ONCE(vc->runnable_threads[i]);
142 /* Used to traverse the list of runnable threads for a given vcore */
143 #define for_each_runnable_thread(i, vcpu, vc) \
144 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
146 static bool kvmppc_ipi_thread(int cpu)
148 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
150 /* On POWER9 we can use msgsnd to IPI any cpu */
151 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
152 msg |= get_hard_smp_processor_id(cpu);
154 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
158 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
159 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
161 if (cpu_first_thread_sibling(cpu) ==
162 cpu_first_thread_sibling(smp_processor_id())) {
163 msg |= cpu_thread_in_core(cpu);
165 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
172 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
173 if (cpu >= 0 && cpu < nr_cpu_ids) {
174 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
178 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
186 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
189 struct swait_queue_head *wqp;
191 wqp = kvm_arch_vcpu_wq(vcpu);
192 if (swq_has_sleeper(wqp)) {
194 ++vcpu->stat.halt_wakeup;
197 cpu = READ_ONCE(vcpu->arch.thread_cpu);
198 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
201 /* CPU points to the first thread of the core */
203 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
204 smp_send_reschedule(cpu);
208 * We use the vcpu_load/put functions to measure stolen time.
209 * Stolen time is counted as time when either the vcpu is able to
210 * run as part of a virtual core, but the task running the vcore
211 * is preempted or sleeping, or when the vcpu needs something done
212 * in the kernel by the task running the vcpu, but that task is
213 * preempted or sleeping. Those two things have to be counted
214 * separately, since one of the vcpu tasks will take on the job
215 * of running the core, and the other vcpu tasks in the vcore will
216 * sleep waiting for it to do that, but that sleep shouldn't count
219 * Hence we accumulate stolen time when the vcpu can run as part of
220 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
221 * needs its task to do other things in the kernel (for example,
222 * service a page fault) in busy_stolen. We don't accumulate
223 * stolen time for a vcore when it is inactive, or for a vcpu
224 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
225 * a misnomer; it means that the vcpu task is not executing in
226 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
227 * the kernel. We don't have any way of dividing up that time
228 * between time that the vcpu is genuinely stopped, time that
229 * the task is actively working on behalf of the vcpu, and time
230 * that the task is preempted, so we don't count any of it as
233 * Updates to busy_stolen are protected by arch.tbacct_lock;
234 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
235 * lock. The stolen times are measured in units of timebase ticks.
236 * (Note that the != TB_NIL checks below are purely defensive;
237 * they should never fail.)
240 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
244 spin_lock_irqsave(&vc->stoltb_lock, flags);
245 vc->preempt_tb = mftb();
246 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
249 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
253 spin_lock_irqsave(&vc->stoltb_lock, flags);
254 if (vc->preempt_tb != TB_NIL) {
255 vc->stolen_tb += mftb() - vc->preempt_tb;
256 vc->preempt_tb = TB_NIL;
258 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
261 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
263 struct kvmppc_vcore *vc = vcpu->arch.vcore;
267 * We can test vc->runner without taking the vcore lock,
268 * because only this task ever sets vc->runner to this
269 * vcpu, and once it is set to this vcpu, only this task
270 * ever sets it to NULL.
272 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
273 kvmppc_core_end_stolen(vc);
275 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
276 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
277 vcpu->arch.busy_preempt != TB_NIL) {
278 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
279 vcpu->arch.busy_preempt = TB_NIL;
281 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
284 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
286 struct kvmppc_vcore *vc = vcpu->arch.vcore;
289 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
290 kvmppc_core_start_stolen(vc);
292 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
293 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
294 vcpu->arch.busy_preempt = mftb();
295 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
298 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
301 * Check for illegal transactional state bit combination
302 * and if we find it, force the TS field to a safe state.
304 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
306 vcpu->arch.shregs.msr = msr;
307 kvmppc_end_cede(vcpu);
310 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
312 vcpu->arch.pvr = pvr;
315 /* Dummy value used in computing PCR value below */
316 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
318 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
320 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
321 struct kvmppc_vcore *vc = vcpu->arch.vcore;
323 /* We can (emulate) our own architecture version and anything older */
324 if (cpu_has_feature(CPU_FTR_ARCH_300))
325 host_pcr_bit = PCR_ARCH_300;
326 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
327 host_pcr_bit = PCR_ARCH_207;
328 else if (cpu_has_feature(CPU_FTR_ARCH_206))
329 host_pcr_bit = PCR_ARCH_206;
331 host_pcr_bit = PCR_ARCH_205;
333 /* Determine lowest PCR bit needed to run guest in given PVR level */
334 guest_pcr_bit = host_pcr_bit;
336 switch (arch_compat) {
338 guest_pcr_bit = PCR_ARCH_205;
342 guest_pcr_bit = PCR_ARCH_206;
345 guest_pcr_bit = PCR_ARCH_207;
348 guest_pcr_bit = PCR_ARCH_300;
355 /* Check requested PCR bits don't exceed our capabilities */
356 if (guest_pcr_bit > host_pcr_bit)
359 spin_lock(&vc->lock);
360 vc->arch_compat = arch_compat;
361 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
362 vc->pcr = host_pcr_bit - guest_pcr_bit;
363 spin_unlock(&vc->lock);
368 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
372 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
373 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
374 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
375 for (r = 0; r < 16; ++r)
376 pr_err("r%2d = %.16lx r%d = %.16lx\n",
377 r, kvmppc_get_gpr(vcpu, r),
378 r+16, kvmppc_get_gpr(vcpu, r+16));
379 pr_err("ctr = %.16lx lr = %.16lx\n",
380 vcpu->arch.ctr, vcpu->arch.lr);
381 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
382 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
383 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
384 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
385 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
386 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
387 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
388 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
389 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
390 pr_err("fault dar = %.16lx dsisr = %.8x\n",
391 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
392 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
393 for (r = 0; r < vcpu->arch.slb_max; ++r)
394 pr_err(" ESID = %.16llx VSID = %.16llx\n",
395 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
396 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
397 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
398 vcpu->arch.last_inst);
401 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
403 struct kvm_vcpu *ret;
405 mutex_lock(&kvm->lock);
406 ret = kvm_get_vcpu_by_id(kvm, id);
407 mutex_unlock(&kvm->lock);
411 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
413 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
414 vpa->yield_count = cpu_to_be32(1);
417 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
418 unsigned long addr, unsigned long len)
420 /* check address is cacheline aligned */
421 if (addr & (L1_CACHE_BYTES - 1))
423 spin_lock(&vcpu->arch.vpa_update_lock);
424 if (v->next_gpa != addr || v->len != len) {
426 v->len = addr ? len : 0;
427 v->update_pending = 1;
429 spin_unlock(&vcpu->arch.vpa_update_lock);
433 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
442 static int vpa_is_registered(struct kvmppc_vpa *vpap)
444 if (vpap->update_pending)
445 return vpap->next_gpa != 0;
446 return vpap->pinned_addr != NULL;
449 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
451 unsigned long vcpuid, unsigned long vpa)
453 struct kvm *kvm = vcpu->kvm;
454 unsigned long len, nb;
456 struct kvm_vcpu *tvcpu;
459 struct kvmppc_vpa *vpap;
461 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
465 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
466 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
467 subfunc == H_VPA_REG_SLB) {
468 /* Registering new area - address must be cache-line aligned */
469 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
472 /* convert logical addr to kernel addr and read length */
473 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
476 if (subfunc == H_VPA_REG_VPA)
477 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
479 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
480 kvmppc_unpin_guest_page(kvm, va, vpa, false);
483 if (len > nb || len < sizeof(struct reg_vpa))
492 spin_lock(&tvcpu->arch.vpa_update_lock);
495 case H_VPA_REG_VPA: /* register VPA */
497 * The size of our lppaca is 1kB because of the way we align
498 * it for the guest to avoid crossing a 4kB boundary. We only
499 * use 640 bytes of the structure though, so we should accept
500 * clients that set a size of 640.
502 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
503 if (len < sizeof(struct lppaca))
505 vpap = &tvcpu->arch.vpa;
509 case H_VPA_REG_DTL: /* register DTL */
510 if (len < sizeof(struct dtl_entry))
512 len -= len % sizeof(struct dtl_entry);
514 /* Check that they have previously registered a VPA */
516 if (!vpa_is_registered(&tvcpu->arch.vpa))
519 vpap = &tvcpu->arch.dtl;
523 case H_VPA_REG_SLB: /* register SLB shadow buffer */
524 /* Check that they have previously registered a VPA */
526 if (!vpa_is_registered(&tvcpu->arch.vpa))
529 vpap = &tvcpu->arch.slb_shadow;
533 case H_VPA_DEREG_VPA: /* deregister VPA */
534 /* Check they don't still have a DTL or SLB buf registered */
536 if (vpa_is_registered(&tvcpu->arch.dtl) ||
537 vpa_is_registered(&tvcpu->arch.slb_shadow))
540 vpap = &tvcpu->arch.vpa;
544 case H_VPA_DEREG_DTL: /* deregister DTL */
545 vpap = &tvcpu->arch.dtl;
549 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
550 vpap = &tvcpu->arch.slb_shadow;
556 vpap->next_gpa = vpa;
558 vpap->update_pending = 1;
561 spin_unlock(&tvcpu->arch.vpa_update_lock);
566 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
568 struct kvm *kvm = vcpu->kvm;
574 * We need to pin the page pointed to by vpap->next_gpa,
575 * but we can't call kvmppc_pin_guest_page under the lock
576 * as it does get_user_pages() and down_read(). So we
577 * have to drop the lock, pin the page, then get the lock
578 * again and check that a new area didn't get registered
582 gpa = vpap->next_gpa;
583 spin_unlock(&vcpu->arch.vpa_update_lock);
587 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
588 spin_lock(&vcpu->arch.vpa_update_lock);
589 if (gpa == vpap->next_gpa)
591 /* sigh... unpin that one and try again */
593 kvmppc_unpin_guest_page(kvm, va, gpa, false);
596 vpap->update_pending = 0;
597 if (va && nb < vpap->len) {
599 * If it's now too short, it must be that userspace
600 * has changed the mappings underlying guest memory,
601 * so unregister the region.
603 kvmppc_unpin_guest_page(kvm, va, gpa, false);
606 if (vpap->pinned_addr)
607 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
610 vpap->pinned_addr = va;
613 vpap->pinned_end = va + vpap->len;
616 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
618 if (!(vcpu->arch.vpa.update_pending ||
619 vcpu->arch.slb_shadow.update_pending ||
620 vcpu->arch.dtl.update_pending))
623 spin_lock(&vcpu->arch.vpa_update_lock);
624 if (vcpu->arch.vpa.update_pending) {
625 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
626 if (vcpu->arch.vpa.pinned_addr)
627 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
629 if (vcpu->arch.dtl.update_pending) {
630 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
631 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
632 vcpu->arch.dtl_index = 0;
634 if (vcpu->arch.slb_shadow.update_pending)
635 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
636 spin_unlock(&vcpu->arch.vpa_update_lock);
640 * Return the accumulated stolen time for the vcore up until `now'.
641 * The caller should hold the vcore lock.
643 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
648 spin_lock_irqsave(&vc->stoltb_lock, flags);
650 if (vc->vcore_state != VCORE_INACTIVE &&
651 vc->preempt_tb != TB_NIL)
652 p += now - vc->preempt_tb;
653 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
657 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
658 struct kvmppc_vcore *vc)
660 struct dtl_entry *dt;
662 unsigned long stolen;
663 unsigned long core_stolen;
667 dt = vcpu->arch.dtl_ptr;
668 vpa = vcpu->arch.vpa.pinned_addr;
670 core_stolen = vcore_stolen_time(vc, now);
671 stolen = core_stolen - vcpu->arch.stolen_logged;
672 vcpu->arch.stolen_logged = core_stolen;
673 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
674 stolen += vcpu->arch.busy_stolen;
675 vcpu->arch.busy_stolen = 0;
676 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
679 memset(dt, 0, sizeof(struct dtl_entry));
680 dt->dispatch_reason = 7;
681 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
682 dt->timebase = cpu_to_be64(now + vc->tb_offset);
683 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
684 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
685 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
687 if (dt == vcpu->arch.dtl.pinned_end)
688 dt = vcpu->arch.dtl.pinned_addr;
689 vcpu->arch.dtl_ptr = dt;
690 /* order writing *dt vs. writing vpa->dtl_idx */
692 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
693 vcpu->arch.dtl.dirty = true;
696 /* See if there is a doorbell interrupt pending for a vcpu */
697 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
700 struct kvmppc_vcore *vc;
702 if (vcpu->arch.doorbell_request)
705 * Ensure that the read of vcore->dpdes comes after the read
706 * of vcpu->doorbell_request. This barrier matches the
707 * lwsync in book3s_hv_rmhandlers.S just before the
708 * fast_guest_return label.
711 vc = vcpu->arch.vcore;
712 thr = vcpu->vcpu_id - vc->first_vcpuid;
713 return !!(vc->dpdes & (1 << thr));
716 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
718 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
720 if ((!vcpu->arch.vcore->arch_compat) &&
721 cpu_has_feature(CPU_FTR_ARCH_207S))
726 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
727 unsigned long resource, unsigned long value1,
728 unsigned long value2)
731 case H_SET_MODE_RESOURCE_SET_CIABR:
732 if (!kvmppc_power8_compatible(vcpu))
737 return H_UNSUPPORTED_FLAG_START;
738 /* Guests can't breakpoint the hypervisor */
739 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
741 vcpu->arch.ciabr = value1;
743 case H_SET_MODE_RESOURCE_SET_DAWR:
744 if (!kvmppc_power8_compatible(vcpu))
746 if (!ppc_breakpoint_available())
749 return H_UNSUPPORTED_FLAG_START;
750 if (value2 & DABRX_HYP)
752 vcpu->arch.dawr = value1;
753 vcpu->arch.dawrx = value2;
760 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
762 struct kvmppc_vcore *vcore = target->arch.vcore;
765 * We expect to have been called by the real mode handler
766 * (kvmppc_rm_h_confer()) which would have directly returned
767 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
768 * have useful work to do and should not confer) so we don't
772 spin_lock(&vcore->lock);
773 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
774 vcore->vcore_state != VCORE_INACTIVE &&
776 target = vcore->runner;
777 spin_unlock(&vcore->lock);
779 return kvm_vcpu_yield_to(target);
782 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
785 struct lppaca *lppaca;
787 spin_lock(&vcpu->arch.vpa_update_lock);
788 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
790 yield_count = be32_to_cpu(lppaca->yield_count);
791 spin_unlock(&vcpu->arch.vpa_update_lock);
795 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
797 unsigned long req = kvmppc_get_gpr(vcpu, 3);
798 unsigned long target, ret = H_SUCCESS;
800 struct kvm_vcpu *tvcpu;
803 if (req <= MAX_HCALL_OPCODE &&
804 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
811 target = kvmppc_get_gpr(vcpu, 4);
812 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
817 tvcpu->arch.prodded = 1;
819 if (tvcpu->arch.ceded)
820 kvmppc_fast_vcpu_kick_hv(tvcpu);
823 target = kvmppc_get_gpr(vcpu, 4);
826 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
831 yield_count = kvmppc_get_gpr(vcpu, 5);
832 if (kvmppc_get_yield_count(tvcpu) != yield_count)
834 kvm_arch_vcpu_yield_to(tvcpu);
837 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
838 kvmppc_get_gpr(vcpu, 5),
839 kvmppc_get_gpr(vcpu, 6));
842 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
845 idx = srcu_read_lock(&vcpu->kvm->srcu);
846 rc = kvmppc_rtas_hcall(vcpu);
847 srcu_read_unlock(&vcpu->kvm->srcu, idx);
854 /* Send the error out to userspace via KVM_RUN */
856 case H_LOGICAL_CI_LOAD:
857 ret = kvmppc_h_logical_ci_load(vcpu);
858 if (ret == H_TOO_HARD)
861 case H_LOGICAL_CI_STORE:
862 ret = kvmppc_h_logical_ci_store(vcpu);
863 if (ret == H_TOO_HARD)
867 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
868 kvmppc_get_gpr(vcpu, 5),
869 kvmppc_get_gpr(vcpu, 6),
870 kvmppc_get_gpr(vcpu, 7));
871 if (ret == H_TOO_HARD)
880 if (kvmppc_xics_enabled(vcpu)) {
881 if (xive_enabled()) {
882 ret = H_NOT_AVAILABLE;
885 ret = kvmppc_xics_hcall(vcpu, req);
890 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
891 kvmppc_get_gpr(vcpu, 5),
892 kvmppc_get_gpr(vcpu, 6));
893 if (ret == H_TOO_HARD)
896 case H_PUT_TCE_INDIRECT:
897 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
898 kvmppc_get_gpr(vcpu, 5),
899 kvmppc_get_gpr(vcpu, 6),
900 kvmppc_get_gpr(vcpu, 7));
901 if (ret == H_TOO_HARD)
905 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
906 kvmppc_get_gpr(vcpu, 5),
907 kvmppc_get_gpr(vcpu, 6),
908 kvmppc_get_gpr(vcpu, 7));
909 if (ret == H_TOO_HARD)
915 kvmppc_set_gpr(vcpu, 3, ret);
916 vcpu->arch.hcall_needed = 0;
920 static int kvmppc_hcall_impl_hv(unsigned long cmd)
928 case H_LOGICAL_CI_LOAD:
929 case H_LOGICAL_CI_STORE:
930 #ifdef CONFIG_KVM_XICS
941 /* See if it's in the real-mode table */
942 return kvmppc_hcall_impl_hv_realmode(cmd);
945 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
946 struct kvm_vcpu *vcpu)
950 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
953 * Fetch failed, so return to guest and
954 * try executing it again.
959 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
960 run->exit_reason = KVM_EXIT_DEBUG;
961 run->debug.arch.address = kvmppc_get_pc(vcpu);
964 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
969 static void do_nothing(void *x)
973 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
975 int thr, cpu, pcpu, nthreads;
979 nthreads = vcpu->kvm->arch.emul_smt_mode;
981 cpu = vcpu->vcpu_id & ~(nthreads - 1);
982 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
983 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
987 * If the vcpu is currently running on a physical cpu thread,
988 * interrupt it in order to pull it out of the guest briefly,
989 * which will update its vcore->dpdes value.
991 pcpu = READ_ONCE(v->cpu);
993 smp_call_function_single(pcpu, do_nothing, NULL, 1);
994 if (kvmppc_doorbell_pending(v))
1001 * On POWER9, emulate doorbell-related instructions in order to
1002 * give the guest the illusion of running on a multi-threaded core.
1003 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1006 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1010 struct kvm *kvm = vcpu->kvm;
1011 struct kvm_vcpu *tvcpu;
1013 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1014 return RESUME_GUEST;
1015 if (get_op(inst) != 31)
1016 return EMULATE_FAIL;
1018 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1019 switch (get_xop(inst)) {
1020 case OP_31_XOP_MSGSNDP:
1021 arg = kvmppc_get_gpr(vcpu, rb);
1022 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1025 if (arg >= kvm->arch.emul_smt_mode)
1027 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1030 if (!tvcpu->arch.doorbell_request) {
1031 tvcpu->arch.doorbell_request = 1;
1032 kvmppc_fast_vcpu_kick_hv(tvcpu);
1035 case OP_31_XOP_MSGCLRP:
1036 arg = kvmppc_get_gpr(vcpu, rb);
1037 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1039 vcpu->arch.vcore->dpdes = 0;
1040 vcpu->arch.doorbell_request = 0;
1042 case OP_31_XOP_MFSPR:
1043 switch (get_sprn(inst)) {
1048 arg = kvmppc_read_dpdes(vcpu);
1051 return EMULATE_FAIL;
1053 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1056 return EMULATE_FAIL;
1058 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1059 return RESUME_GUEST;
1062 /* Called with vcpu->arch.vcore->lock held */
1063 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1064 struct task_struct *tsk)
1066 int r = RESUME_HOST;
1068 vcpu->stat.sum_exits++;
1071 * This can happen if an interrupt occurs in the last stages
1072 * of guest entry or the first stages of guest exit (i.e. after
1073 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1074 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1075 * That can happen due to a bug, or due to a machine check
1076 * occurring at just the wrong time.
1078 if (vcpu->arch.shregs.msr & MSR_HV) {
1079 printk(KERN_EMERG "KVM trap in HV mode!\n");
1080 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1081 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1082 vcpu->arch.shregs.msr);
1083 kvmppc_dump_regs(vcpu);
1084 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1085 run->hw.hardware_exit_reason = vcpu->arch.trap;
1088 run->exit_reason = KVM_EXIT_UNKNOWN;
1089 run->ready_for_interrupt_injection = 1;
1090 switch (vcpu->arch.trap) {
1091 /* We're good on these - the host merely wanted to get our attention */
1092 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1093 vcpu->stat.dec_exits++;
1096 case BOOK3S_INTERRUPT_EXTERNAL:
1097 case BOOK3S_INTERRUPT_H_DOORBELL:
1098 case BOOK3S_INTERRUPT_H_VIRT:
1099 vcpu->stat.ext_intr_exits++;
1102 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1103 case BOOK3S_INTERRUPT_HMI:
1104 case BOOK3S_INTERRUPT_PERFMON:
1105 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1108 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1109 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1110 run->exit_reason = KVM_EXIT_NMI;
1111 run->hw.hardware_exit_reason = vcpu->arch.trap;
1112 /* Clear out the old NMI status from run->flags */
1113 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1114 /* Now set the NMI status */
1115 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1116 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1118 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1121 /* Print the MCE event to host console. */
1122 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1124 case BOOK3S_INTERRUPT_PROGRAM:
1128 * Normally program interrupts are delivered directly
1129 * to the guest by the hardware, but we can get here
1130 * as a result of a hypervisor emulation interrupt
1131 * (e40) getting turned into a 700 by BML RTAS.
1133 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1134 kvmppc_core_queue_program(vcpu, flags);
1138 case BOOK3S_INTERRUPT_SYSCALL:
1140 /* hcall - punt to userspace */
1143 /* hypercall with MSR_PR has already been handled in rmode,
1144 * and never reaches here.
1147 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1148 for (i = 0; i < 9; ++i)
1149 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1150 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1151 vcpu->arch.hcall_needed = 1;
1156 * We get these next two if the guest accesses a page which it thinks
1157 * it has mapped but which is not actually present, either because
1158 * it is for an emulated I/O device or because the corresonding
1159 * host page has been paged out. Any other HDSI/HISI interrupts
1160 * have been handled already.
1162 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1163 r = RESUME_PAGE_FAULT;
1165 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1166 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1167 vcpu->arch.fault_dsisr = 0;
1168 r = RESUME_PAGE_FAULT;
1171 * This occurs if the guest executes an illegal instruction.
1172 * If the guest debug is disabled, generate a program interrupt
1173 * to the guest. If guest debug is enabled, we need to check
1174 * whether the instruction is a software breakpoint instruction.
1175 * Accordingly return to Guest or Host.
1177 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1178 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1179 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1180 swab32(vcpu->arch.emul_inst) :
1181 vcpu->arch.emul_inst;
1182 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1183 /* Need vcore unlocked to call kvmppc_get_last_inst */
1184 spin_unlock(&vcpu->arch.vcore->lock);
1185 r = kvmppc_emulate_debug_inst(run, vcpu);
1186 spin_lock(&vcpu->arch.vcore->lock);
1188 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1193 * This occurs if the guest (kernel or userspace), does something that
1194 * is prohibited by HFSCR.
1195 * On POWER9, this could be a doorbell instruction that we need
1197 * Otherwise, we just generate a program interrupt to the guest.
1199 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1201 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1202 cpu_has_feature(CPU_FTR_ARCH_300)) {
1203 /* Need vcore unlocked to call kvmppc_get_last_inst */
1204 spin_unlock(&vcpu->arch.vcore->lock);
1205 r = kvmppc_emulate_doorbell_instr(vcpu);
1206 spin_lock(&vcpu->arch.vcore->lock);
1208 if (r == EMULATE_FAIL) {
1209 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1214 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1215 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1217 * This occurs for various TM-related instructions that
1218 * we need to emulate on POWER9 DD2.2. We have already
1219 * handled the cases where the guest was in real-suspend
1220 * mode and was transitioning to transactional state.
1222 r = kvmhv_p9_tm_emulation(vcpu);
1226 case BOOK3S_INTERRUPT_HV_RM_HARD:
1227 r = RESUME_PASSTHROUGH;
1230 kvmppc_dump_regs(vcpu);
1231 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1232 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1233 vcpu->arch.shregs.msr);
1234 run->hw.hardware_exit_reason = vcpu->arch.trap;
1242 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1243 struct kvm_sregs *sregs)
1247 memset(sregs, 0, sizeof(struct kvm_sregs));
1248 sregs->pvr = vcpu->arch.pvr;
1249 for (i = 0; i < vcpu->arch.slb_max; i++) {
1250 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1251 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1257 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1258 struct kvm_sregs *sregs)
1262 /* Only accept the same PVR as the host's, since we can't spoof it */
1263 if (sregs->pvr != vcpu->arch.pvr)
1267 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1268 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1269 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1270 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1274 vcpu->arch.slb_max = j;
1279 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1280 bool preserve_top32)
1282 struct kvm *kvm = vcpu->kvm;
1283 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1286 mutex_lock(&kvm->lock);
1287 spin_lock(&vc->lock);
1289 * If ILE (interrupt little-endian) has changed, update the
1290 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1292 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1293 struct kvm_vcpu *vcpu;
1296 kvm_for_each_vcpu(i, vcpu, kvm) {
1297 if (vcpu->arch.vcore != vc)
1299 if (new_lpcr & LPCR_ILE)
1300 vcpu->arch.intr_msr |= MSR_LE;
1302 vcpu->arch.intr_msr &= ~MSR_LE;
1307 * Userspace can only modify DPFD (default prefetch depth),
1308 * ILE (interrupt little-endian) and TC (translation control).
1309 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1311 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1312 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1315 * On POWER9, allow userspace to enable large decrementer for the
1316 * guest, whether or not the host has it enabled.
1318 if (cpu_has_feature(CPU_FTR_ARCH_300))
1321 /* Broken 32-bit version of LPCR must not clear top bits */
1324 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1325 spin_unlock(&vc->lock);
1326 mutex_unlock(&kvm->lock);
1329 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1330 union kvmppc_one_reg *val)
1336 case KVM_REG_PPC_DEBUG_INST:
1337 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1339 case KVM_REG_PPC_HIOR:
1340 *val = get_reg_val(id, 0);
1342 case KVM_REG_PPC_DABR:
1343 *val = get_reg_val(id, vcpu->arch.dabr);
1345 case KVM_REG_PPC_DABRX:
1346 *val = get_reg_val(id, vcpu->arch.dabrx);
1348 case KVM_REG_PPC_DSCR:
1349 *val = get_reg_val(id, vcpu->arch.dscr);
1351 case KVM_REG_PPC_PURR:
1352 *val = get_reg_val(id, vcpu->arch.purr);
1354 case KVM_REG_PPC_SPURR:
1355 *val = get_reg_val(id, vcpu->arch.spurr);
1357 case KVM_REG_PPC_AMR:
1358 *val = get_reg_val(id, vcpu->arch.amr);
1360 case KVM_REG_PPC_UAMOR:
1361 *val = get_reg_val(id, vcpu->arch.uamor);
1363 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1364 i = id - KVM_REG_PPC_MMCR0;
1365 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1367 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1368 i = id - KVM_REG_PPC_PMC1;
1369 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1371 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1372 i = id - KVM_REG_PPC_SPMC1;
1373 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1375 case KVM_REG_PPC_SIAR:
1376 *val = get_reg_val(id, vcpu->arch.siar);
1378 case KVM_REG_PPC_SDAR:
1379 *val = get_reg_val(id, vcpu->arch.sdar);
1381 case KVM_REG_PPC_SIER:
1382 *val = get_reg_val(id, vcpu->arch.sier);
1384 case KVM_REG_PPC_IAMR:
1385 *val = get_reg_val(id, vcpu->arch.iamr);
1387 case KVM_REG_PPC_PSPB:
1388 *val = get_reg_val(id, vcpu->arch.pspb);
1390 case KVM_REG_PPC_DPDES:
1391 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1393 case KVM_REG_PPC_VTB:
1394 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1396 case KVM_REG_PPC_DAWR:
1397 *val = get_reg_val(id, vcpu->arch.dawr);
1399 case KVM_REG_PPC_DAWRX:
1400 *val = get_reg_val(id, vcpu->arch.dawrx);
1402 case KVM_REG_PPC_CIABR:
1403 *val = get_reg_val(id, vcpu->arch.ciabr);
1405 case KVM_REG_PPC_CSIGR:
1406 *val = get_reg_val(id, vcpu->arch.csigr);
1408 case KVM_REG_PPC_TACR:
1409 *val = get_reg_val(id, vcpu->arch.tacr);
1411 case KVM_REG_PPC_TCSCR:
1412 *val = get_reg_val(id, vcpu->arch.tcscr);
1414 case KVM_REG_PPC_PID:
1415 *val = get_reg_val(id, vcpu->arch.pid);
1417 case KVM_REG_PPC_ACOP:
1418 *val = get_reg_val(id, vcpu->arch.acop);
1420 case KVM_REG_PPC_WORT:
1421 *val = get_reg_val(id, vcpu->arch.wort);
1423 case KVM_REG_PPC_TIDR:
1424 *val = get_reg_val(id, vcpu->arch.tid);
1426 case KVM_REG_PPC_PSSCR:
1427 *val = get_reg_val(id, vcpu->arch.psscr);
1429 case KVM_REG_PPC_VPA_ADDR:
1430 spin_lock(&vcpu->arch.vpa_update_lock);
1431 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1432 spin_unlock(&vcpu->arch.vpa_update_lock);
1434 case KVM_REG_PPC_VPA_SLB:
1435 spin_lock(&vcpu->arch.vpa_update_lock);
1436 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1437 val->vpaval.length = vcpu->arch.slb_shadow.len;
1438 spin_unlock(&vcpu->arch.vpa_update_lock);
1440 case KVM_REG_PPC_VPA_DTL:
1441 spin_lock(&vcpu->arch.vpa_update_lock);
1442 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1443 val->vpaval.length = vcpu->arch.dtl.len;
1444 spin_unlock(&vcpu->arch.vpa_update_lock);
1446 case KVM_REG_PPC_TB_OFFSET:
1447 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1449 case KVM_REG_PPC_LPCR:
1450 case KVM_REG_PPC_LPCR_64:
1451 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1453 case KVM_REG_PPC_PPR:
1454 *val = get_reg_val(id, vcpu->arch.ppr);
1456 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1457 case KVM_REG_PPC_TFHAR:
1458 *val = get_reg_val(id, vcpu->arch.tfhar);
1460 case KVM_REG_PPC_TFIAR:
1461 *val = get_reg_val(id, vcpu->arch.tfiar);
1463 case KVM_REG_PPC_TEXASR:
1464 *val = get_reg_val(id, vcpu->arch.texasr);
1466 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1467 i = id - KVM_REG_PPC_TM_GPR0;
1468 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1470 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1473 i = id - KVM_REG_PPC_TM_VSR0;
1475 for (j = 0; j < TS_FPRWIDTH; j++)
1476 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1478 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1479 val->vval = vcpu->arch.vr_tm.vr[i-32];
1485 case KVM_REG_PPC_TM_CR:
1486 *val = get_reg_val(id, vcpu->arch.cr_tm);
1488 case KVM_REG_PPC_TM_XER:
1489 *val = get_reg_val(id, vcpu->arch.xer_tm);
1491 case KVM_REG_PPC_TM_LR:
1492 *val = get_reg_val(id, vcpu->arch.lr_tm);
1494 case KVM_REG_PPC_TM_CTR:
1495 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1497 case KVM_REG_PPC_TM_FPSCR:
1498 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1500 case KVM_REG_PPC_TM_AMR:
1501 *val = get_reg_val(id, vcpu->arch.amr_tm);
1503 case KVM_REG_PPC_TM_PPR:
1504 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1506 case KVM_REG_PPC_TM_VRSAVE:
1507 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1509 case KVM_REG_PPC_TM_VSCR:
1510 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1511 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1515 case KVM_REG_PPC_TM_DSCR:
1516 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1518 case KVM_REG_PPC_TM_TAR:
1519 *val = get_reg_val(id, vcpu->arch.tar_tm);
1522 case KVM_REG_PPC_ARCH_COMPAT:
1523 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1525 case KVM_REG_PPC_DEC_EXPIRY:
1526 *val = get_reg_val(id, vcpu->arch.dec_expires +
1527 vcpu->arch.vcore->tb_offset);
1537 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1538 union kvmppc_one_reg *val)
1542 unsigned long addr, len;
1545 case KVM_REG_PPC_HIOR:
1546 /* Only allow this to be set to zero */
1547 if (set_reg_val(id, *val))
1550 case KVM_REG_PPC_DABR:
1551 vcpu->arch.dabr = set_reg_val(id, *val);
1553 case KVM_REG_PPC_DABRX:
1554 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1556 case KVM_REG_PPC_DSCR:
1557 vcpu->arch.dscr = set_reg_val(id, *val);
1559 case KVM_REG_PPC_PURR:
1560 vcpu->arch.purr = set_reg_val(id, *val);
1562 case KVM_REG_PPC_SPURR:
1563 vcpu->arch.spurr = set_reg_val(id, *val);
1565 case KVM_REG_PPC_AMR:
1566 vcpu->arch.amr = set_reg_val(id, *val);
1568 case KVM_REG_PPC_UAMOR:
1569 vcpu->arch.uamor = set_reg_val(id, *val);
1571 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1572 i = id - KVM_REG_PPC_MMCR0;
1573 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1575 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1576 i = id - KVM_REG_PPC_PMC1;
1577 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1579 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1580 i = id - KVM_REG_PPC_SPMC1;
1581 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1583 case KVM_REG_PPC_SIAR:
1584 vcpu->arch.siar = set_reg_val(id, *val);
1586 case KVM_REG_PPC_SDAR:
1587 vcpu->arch.sdar = set_reg_val(id, *val);
1589 case KVM_REG_PPC_SIER:
1590 vcpu->arch.sier = set_reg_val(id, *val);
1592 case KVM_REG_PPC_IAMR:
1593 vcpu->arch.iamr = set_reg_val(id, *val);
1595 case KVM_REG_PPC_PSPB:
1596 vcpu->arch.pspb = set_reg_val(id, *val);
1598 case KVM_REG_PPC_DPDES:
1599 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1601 case KVM_REG_PPC_VTB:
1602 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1604 case KVM_REG_PPC_DAWR:
1605 vcpu->arch.dawr = set_reg_val(id, *val);
1607 case KVM_REG_PPC_DAWRX:
1608 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1610 case KVM_REG_PPC_CIABR:
1611 vcpu->arch.ciabr = set_reg_val(id, *val);
1612 /* Don't allow setting breakpoints in hypervisor code */
1613 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1614 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1616 case KVM_REG_PPC_CSIGR:
1617 vcpu->arch.csigr = set_reg_val(id, *val);
1619 case KVM_REG_PPC_TACR:
1620 vcpu->arch.tacr = set_reg_val(id, *val);
1622 case KVM_REG_PPC_TCSCR:
1623 vcpu->arch.tcscr = set_reg_val(id, *val);
1625 case KVM_REG_PPC_PID:
1626 vcpu->arch.pid = set_reg_val(id, *val);
1628 case KVM_REG_PPC_ACOP:
1629 vcpu->arch.acop = set_reg_val(id, *val);
1631 case KVM_REG_PPC_WORT:
1632 vcpu->arch.wort = set_reg_val(id, *val);
1634 case KVM_REG_PPC_TIDR:
1635 vcpu->arch.tid = set_reg_val(id, *val);
1637 case KVM_REG_PPC_PSSCR:
1638 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1640 case KVM_REG_PPC_VPA_ADDR:
1641 addr = set_reg_val(id, *val);
1643 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1644 vcpu->arch.dtl.next_gpa))
1646 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1648 case KVM_REG_PPC_VPA_SLB:
1649 addr = val->vpaval.addr;
1650 len = val->vpaval.length;
1652 if (addr && !vcpu->arch.vpa.next_gpa)
1654 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1656 case KVM_REG_PPC_VPA_DTL:
1657 addr = val->vpaval.addr;
1658 len = val->vpaval.length;
1660 if (addr && (len < sizeof(struct dtl_entry) ||
1661 !vcpu->arch.vpa.next_gpa))
1663 len -= len % sizeof(struct dtl_entry);
1664 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1666 case KVM_REG_PPC_TB_OFFSET:
1668 * POWER9 DD1 has an erratum where writing TBU40 causes
1669 * the timebase to lose ticks. So we don't let the
1670 * timebase offset be changed on P9 DD1. (It is
1671 * initialized to zero.)
1673 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1675 /* round up to multiple of 2^24 */
1676 vcpu->arch.vcore->tb_offset =
1677 ALIGN(set_reg_val(id, *val), 1UL << 24);
1679 case KVM_REG_PPC_LPCR:
1680 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1682 case KVM_REG_PPC_LPCR_64:
1683 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1685 case KVM_REG_PPC_PPR:
1686 vcpu->arch.ppr = set_reg_val(id, *val);
1688 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1689 case KVM_REG_PPC_TFHAR:
1690 vcpu->arch.tfhar = set_reg_val(id, *val);
1692 case KVM_REG_PPC_TFIAR:
1693 vcpu->arch.tfiar = set_reg_val(id, *val);
1695 case KVM_REG_PPC_TEXASR:
1696 vcpu->arch.texasr = set_reg_val(id, *val);
1698 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1699 i = id - KVM_REG_PPC_TM_GPR0;
1700 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1702 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1705 i = id - KVM_REG_PPC_TM_VSR0;
1707 for (j = 0; j < TS_FPRWIDTH; j++)
1708 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1710 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1711 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1716 case KVM_REG_PPC_TM_CR:
1717 vcpu->arch.cr_tm = set_reg_val(id, *val);
1719 case KVM_REG_PPC_TM_XER:
1720 vcpu->arch.xer_tm = set_reg_val(id, *val);
1722 case KVM_REG_PPC_TM_LR:
1723 vcpu->arch.lr_tm = set_reg_val(id, *val);
1725 case KVM_REG_PPC_TM_CTR:
1726 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1728 case KVM_REG_PPC_TM_FPSCR:
1729 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1731 case KVM_REG_PPC_TM_AMR:
1732 vcpu->arch.amr_tm = set_reg_val(id, *val);
1734 case KVM_REG_PPC_TM_PPR:
1735 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1737 case KVM_REG_PPC_TM_VRSAVE:
1738 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1740 case KVM_REG_PPC_TM_VSCR:
1741 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1742 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1746 case KVM_REG_PPC_TM_DSCR:
1747 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1749 case KVM_REG_PPC_TM_TAR:
1750 vcpu->arch.tar_tm = set_reg_val(id, *val);
1753 case KVM_REG_PPC_ARCH_COMPAT:
1754 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1756 case KVM_REG_PPC_DEC_EXPIRY:
1757 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1758 vcpu->arch.vcore->tb_offset;
1769 * On POWER9, threads are independent and can be in different partitions.
1770 * Therefore we consider each thread to be a subcore.
1771 * There is a restriction that all threads have to be in the same
1772 * MMU mode (radix or HPT), unfortunately, but since we only support
1773 * HPT guests on a HPT host so far, that isn't an impediment yet.
1775 static int threads_per_vcore(struct kvm *kvm)
1777 if (kvm->arch.threads_indep)
1779 return threads_per_subcore;
1782 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1784 struct kvmppc_vcore *vcore;
1786 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1791 spin_lock_init(&vcore->lock);
1792 spin_lock_init(&vcore->stoltb_lock);
1793 init_swait_queue_head(&vcore->wq);
1794 vcore->preempt_tb = TB_NIL;
1795 vcore->lpcr = kvm->arch.lpcr;
1796 vcore->first_vcpuid = core * kvm->arch.smt_mode;
1798 INIT_LIST_HEAD(&vcore->preempt_list);
1803 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1804 static struct debugfs_timings_element {
1808 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1809 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1810 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1811 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1812 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1815 #define N_TIMINGS (ARRAY_SIZE(timings))
1817 struct debugfs_timings_state {
1818 struct kvm_vcpu *vcpu;
1819 unsigned int buflen;
1820 char buf[N_TIMINGS * 100];
1823 static int debugfs_timings_open(struct inode *inode, struct file *file)
1825 struct kvm_vcpu *vcpu = inode->i_private;
1826 struct debugfs_timings_state *p;
1828 p = kzalloc(sizeof(*p), GFP_KERNEL);
1832 kvm_get_kvm(vcpu->kvm);
1834 file->private_data = p;
1836 return nonseekable_open(inode, file);
1839 static int debugfs_timings_release(struct inode *inode, struct file *file)
1841 struct debugfs_timings_state *p = file->private_data;
1843 kvm_put_kvm(p->vcpu->kvm);
1848 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1849 size_t len, loff_t *ppos)
1851 struct debugfs_timings_state *p = file->private_data;
1852 struct kvm_vcpu *vcpu = p->vcpu;
1854 struct kvmhv_tb_accumulator tb;
1863 buf_end = s + sizeof(p->buf);
1864 for (i = 0; i < N_TIMINGS; ++i) {
1865 struct kvmhv_tb_accumulator *acc;
1867 acc = (struct kvmhv_tb_accumulator *)
1868 ((unsigned long)vcpu + timings[i].offset);
1870 for (loops = 0; loops < 1000; ++loops) {
1871 count = acc->seqcount;
1876 if (count == acc->seqcount) {
1884 snprintf(s, buf_end - s, "%s: stuck\n",
1887 snprintf(s, buf_end - s,
1888 "%s: %llu %llu %llu %llu\n",
1889 timings[i].name, count / 2,
1890 tb_to_ns(tb.tb_total),
1891 tb_to_ns(tb.tb_min),
1892 tb_to_ns(tb.tb_max));
1895 p->buflen = s - p->buf;
1899 if (pos >= p->buflen)
1901 if (len > p->buflen - pos)
1902 len = p->buflen - pos;
1903 n = copy_to_user(buf, p->buf + pos, len);
1913 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1914 size_t len, loff_t *ppos)
1919 static const struct file_operations debugfs_timings_ops = {
1920 .owner = THIS_MODULE,
1921 .open = debugfs_timings_open,
1922 .release = debugfs_timings_release,
1923 .read = debugfs_timings_read,
1924 .write = debugfs_timings_write,
1925 .llseek = generic_file_llseek,
1928 /* Create a debugfs directory for the vcpu */
1929 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1932 struct kvm *kvm = vcpu->kvm;
1934 snprintf(buf, sizeof(buf), "vcpu%u", id);
1935 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1937 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1938 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1940 vcpu->arch.debugfs_timings =
1941 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1942 vcpu, &debugfs_timings_ops);
1945 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1946 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1949 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1951 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1954 struct kvm_vcpu *vcpu;
1957 struct kvmppc_vcore *vcore;
1960 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1964 err = kvm_vcpu_init(vcpu, kvm, id);
1968 vcpu->arch.shared = &vcpu->arch.shregs;
1969 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1971 * The shared struct is never shared on HV,
1972 * so we can always use host endianness
1974 #ifdef __BIG_ENDIAN__
1975 vcpu->arch.shared_big_endian = true;
1977 vcpu->arch.shared_big_endian = false;
1980 vcpu->arch.mmcr[0] = MMCR0_FC;
1981 vcpu->arch.ctrl = CTRL_RUNLATCH;
1982 /* default to host PVR, since we can't spoof it */
1983 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1984 spin_lock_init(&vcpu->arch.vpa_update_lock);
1985 spin_lock_init(&vcpu->arch.tbacct_lock);
1986 vcpu->arch.busy_preempt = TB_NIL;
1987 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1990 * Set the default HFSCR for the guest from the host value.
1991 * This value is only used on POWER9.
1992 * On POWER9 DD1, TM doesn't work, so we make sure to
1993 * prevent the guest from using it.
1994 * On POWER9, we want to virtualize the doorbell facility, so we
1995 * turn off the HFSCR bit, which causes those instructions to trap.
1997 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1998 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
1999 vcpu->arch.hfscr |= HFSCR_TM;
2000 else if (!cpu_has_feature(CPU_FTR_TM_COMP))
2001 vcpu->arch.hfscr &= ~HFSCR_TM;
2002 if (cpu_has_feature(CPU_FTR_ARCH_300))
2003 vcpu->arch.hfscr &= ~HFSCR_MSGP;
2005 kvmppc_mmu_book3s_hv_init(vcpu);
2007 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2009 init_waitqueue_head(&vcpu->arch.cpu_run);
2011 mutex_lock(&kvm->lock);
2014 core = id / kvm->arch.smt_mode;
2015 if (core < KVM_MAX_VCORES) {
2016 vcore = kvm->arch.vcores[core];
2019 vcore = kvmppc_vcore_create(kvm, core);
2020 kvm->arch.vcores[core] = vcore;
2021 kvm->arch.online_vcores++;
2024 mutex_unlock(&kvm->lock);
2029 spin_lock(&vcore->lock);
2030 ++vcore->num_threads;
2031 spin_unlock(&vcore->lock);
2032 vcpu->arch.vcore = vcore;
2033 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2034 vcpu->arch.thread_cpu = -1;
2035 vcpu->arch.prev_cpu = -1;
2037 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2038 kvmppc_sanity_check(vcpu);
2040 debugfs_vcpu_init(vcpu, id);
2045 kmem_cache_free(kvm_vcpu_cache, vcpu);
2047 return ERR_PTR(err);
2050 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2051 unsigned long flags)
2058 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2060 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2062 * On POWER8 (or POWER7), the threading mode is "strict",
2063 * so we pack smt_mode vcpus per vcore.
2065 if (smt_mode > threads_per_subcore)
2069 * On POWER9, the threading mode is "loose",
2070 * so each vcpu gets its own vcore.
2075 mutex_lock(&kvm->lock);
2077 if (!kvm->arch.online_vcores) {
2078 kvm->arch.smt_mode = smt_mode;
2079 kvm->arch.emul_smt_mode = esmt;
2082 mutex_unlock(&kvm->lock);
2087 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2089 if (vpa->pinned_addr)
2090 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2094 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2096 spin_lock(&vcpu->arch.vpa_update_lock);
2097 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2098 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2099 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2100 spin_unlock(&vcpu->arch.vpa_update_lock);
2101 kvm_vcpu_uninit(vcpu);
2102 kmem_cache_free(kvm_vcpu_cache, vcpu);
2105 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2107 /* Indicate we want to get back into the guest */
2111 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2113 unsigned long dec_nsec, now;
2116 if (now > vcpu->arch.dec_expires) {
2117 /* decrementer has already gone negative */
2118 kvmppc_core_queue_dec(vcpu);
2119 kvmppc_core_prepare_to_enter(vcpu);
2122 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2124 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2125 vcpu->arch.timer_running = 1;
2128 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2130 vcpu->arch.ceded = 0;
2131 if (vcpu->arch.timer_running) {
2132 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2133 vcpu->arch.timer_running = 0;
2137 extern int __kvmppc_vcore_entry(void);
2139 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2140 struct kvm_vcpu *vcpu)
2144 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2146 spin_lock_irq(&vcpu->arch.tbacct_lock);
2148 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2149 vcpu->arch.stolen_logged;
2150 vcpu->arch.busy_preempt = now;
2151 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2152 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2154 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2157 static int kvmppc_grab_hwthread(int cpu)
2159 struct paca_struct *tpaca;
2160 long timeout = 10000;
2162 tpaca = paca_ptrs[cpu];
2164 /* Ensure the thread won't go into the kernel if it wakes */
2165 tpaca->kvm_hstate.kvm_vcpu = NULL;
2166 tpaca->kvm_hstate.kvm_vcore = NULL;
2167 tpaca->kvm_hstate.napping = 0;
2169 tpaca->kvm_hstate.hwthread_req = 1;
2172 * If the thread is already executing in the kernel (e.g. handling
2173 * a stray interrupt), wait for it to get back to nap mode.
2174 * The smp_mb() is to ensure that our setting of hwthread_req
2175 * is visible before we look at hwthread_state, so if this
2176 * races with the code at system_reset_pSeries and the thread
2177 * misses our setting of hwthread_req, we are sure to see its
2178 * setting of hwthread_state, and vice versa.
2181 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2182 if (--timeout <= 0) {
2183 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2191 static void kvmppc_release_hwthread(int cpu)
2193 struct paca_struct *tpaca;
2195 tpaca = paca_ptrs[cpu];
2196 tpaca->kvm_hstate.hwthread_req = 0;
2197 tpaca->kvm_hstate.kvm_vcpu = NULL;
2198 tpaca->kvm_hstate.kvm_vcore = NULL;
2199 tpaca->kvm_hstate.kvm_split_mode = NULL;
2202 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2206 cpu = cpu_first_thread_sibling(cpu);
2207 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2209 * Make sure setting of bit in need_tlb_flush precedes
2210 * testing of cpu_in_guest bits. The matching barrier on
2211 * the other side is the first smp_mb() in kvmppc_run_core().
2214 for (i = 0; i < threads_per_core; ++i)
2215 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2216 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2219 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2221 struct kvm *kvm = vcpu->kvm;
2224 * With radix, the guest can do TLB invalidations itself,
2225 * and it could choose to use the local form (tlbiel) if
2226 * it is invalidating a translation that has only ever been
2227 * used on one vcpu. However, that doesn't mean it has
2228 * only ever been used on one physical cpu, since vcpus
2229 * can move around between pcpus. To cope with this, when
2230 * a vcpu moves from one pcpu to another, we need to tell
2231 * any vcpus running on the same core as this vcpu previously
2232 * ran to flush the TLB. The TLB is shared between threads,
2233 * so we use a single bit in .need_tlb_flush for all 4 threads.
2235 if (vcpu->arch.prev_cpu != pcpu) {
2236 if (vcpu->arch.prev_cpu >= 0 &&
2237 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2238 cpu_first_thread_sibling(pcpu))
2239 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2240 vcpu->arch.prev_cpu = pcpu;
2244 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2247 struct paca_struct *tpaca;
2248 struct kvm *kvm = vc->kvm;
2252 if (vcpu->arch.timer_running) {
2253 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2254 vcpu->arch.timer_running = 0;
2256 cpu += vcpu->arch.ptid;
2257 vcpu->cpu = vc->pcpu;
2258 vcpu->arch.thread_cpu = cpu;
2259 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2261 tpaca = paca_ptrs[cpu];
2262 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2263 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2264 tpaca->kvm_hstate.fake_suspend = 0;
2265 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2267 tpaca->kvm_hstate.kvm_vcore = vc;
2268 if (cpu != smp_processor_id())
2269 kvmppc_ipi_thread(cpu);
2272 static void kvmppc_wait_for_nap(int n_threads)
2274 int cpu = smp_processor_id();
2279 for (loops = 0; loops < 1000000; ++loops) {
2281 * Check if all threads are finished.
2282 * We set the vcore pointer when starting a thread
2283 * and the thread clears it when finished, so we look
2284 * for any threads that still have a non-NULL vcore ptr.
2286 for (i = 1; i < n_threads; ++i)
2287 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2289 if (i == n_threads) {
2296 for (i = 1; i < n_threads; ++i)
2297 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2298 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2302 * Check that we are on thread 0 and that any other threads in
2303 * this core are off-line. Then grab the threads so they can't
2306 static int on_primary_thread(void)
2308 int cpu = smp_processor_id();
2311 /* Are we on a primary subcore? */
2312 if (cpu_thread_in_subcore(cpu))
2316 while (++thr < threads_per_subcore)
2317 if (cpu_online(cpu + thr))
2320 /* Grab all hw threads so they can't go into the kernel */
2321 for (thr = 1; thr < threads_per_subcore; ++thr) {
2322 if (kvmppc_grab_hwthread(cpu + thr)) {
2323 /* Couldn't grab one; let the others go */
2325 kvmppc_release_hwthread(cpu + thr);
2326 } while (--thr > 0);
2334 * A list of virtual cores for each physical CPU.
2335 * These are vcores that could run but their runner VCPU tasks are
2336 * (or may be) preempted.
2338 struct preempted_vcore_list {
2339 struct list_head list;
2343 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2345 static void init_vcore_lists(void)
2349 for_each_possible_cpu(cpu) {
2350 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2351 spin_lock_init(&lp->lock);
2352 INIT_LIST_HEAD(&lp->list);
2356 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2358 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2360 vc->vcore_state = VCORE_PREEMPT;
2361 vc->pcpu = smp_processor_id();
2362 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2363 spin_lock(&lp->lock);
2364 list_add_tail(&vc->preempt_list, &lp->list);
2365 spin_unlock(&lp->lock);
2368 /* Start accumulating stolen time */
2369 kvmppc_core_start_stolen(vc);
2372 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2374 struct preempted_vcore_list *lp;
2376 kvmppc_core_end_stolen(vc);
2377 if (!list_empty(&vc->preempt_list)) {
2378 lp = &per_cpu(preempted_vcores, vc->pcpu);
2379 spin_lock(&lp->lock);
2380 list_del_init(&vc->preempt_list);
2381 spin_unlock(&lp->lock);
2383 vc->vcore_state = VCORE_INACTIVE;
2387 * This stores information about the virtual cores currently
2388 * assigned to a physical core.
2392 int max_subcore_threads;
2394 int subcore_threads[MAX_SUBCORES];
2395 struct kvmppc_vcore *vc[MAX_SUBCORES];
2399 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2400 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2402 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2404 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2406 memset(cip, 0, sizeof(*cip));
2407 cip->n_subcores = 1;
2408 cip->max_subcore_threads = vc->num_threads;
2409 cip->total_threads = vc->num_threads;
2410 cip->subcore_threads[0] = vc->num_threads;
2414 static bool subcore_config_ok(int n_subcores, int n_threads)
2417 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2418 * split-core mode, with one thread per subcore.
2420 if (cpu_has_feature(CPU_FTR_ARCH_300))
2421 return n_subcores <= 4 && n_threads == 1;
2423 /* On POWER8, can only dynamically split if unsplit to begin with */
2424 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2426 if (n_subcores > MAX_SUBCORES)
2428 if (n_subcores > 1) {
2429 if (!(dynamic_mt_modes & 2))
2431 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2435 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2438 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2440 vc->entry_exit_map = 0;
2442 vc->napping_threads = 0;
2443 vc->conferring_threads = 0;
2444 vc->tb_offset_applied = 0;
2447 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2449 int n_threads = vc->num_threads;
2452 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2455 /* Some POWER9 chips require all threads to be in the same MMU mode */
2456 if (no_mixing_hpt_and_radix &&
2457 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2460 if (n_threads < cip->max_subcore_threads)
2461 n_threads = cip->max_subcore_threads;
2462 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2464 cip->max_subcore_threads = n_threads;
2466 sub = cip->n_subcores;
2468 cip->total_threads += vc->num_threads;
2469 cip->subcore_threads[sub] = vc->num_threads;
2471 init_vcore_to_run(vc);
2472 list_del_init(&vc->preempt_list);
2478 * Work out whether it is possible to piggyback the execution of
2479 * vcore *pvc onto the execution of the other vcores described in *cip.
2481 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2484 if (cip->total_threads + pvc->num_threads > target_threads)
2487 return can_dynamic_split(pvc, cip);
2490 static void prepare_threads(struct kvmppc_vcore *vc)
2493 struct kvm_vcpu *vcpu;
2495 for_each_runnable_thread(i, vcpu, vc) {
2496 if (signal_pending(vcpu->arch.run_task))
2497 vcpu->arch.ret = -EINTR;
2498 else if (vcpu->arch.vpa.update_pending ||
2499 vcpu->arch.slb_shadow.update_pending ||
2500 vcpu->arch.dtl.update_pending)
2501 vcpu->arch.ret = RESUME_GUEST;
2504 kvmppc_remove_runnable(vc, vcpu);
2505 wake_up(&vcpu->arch.cpu_run);
2509 static void collect_piggybacks(struct core_info *cip, int target_threads)
2511 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2512 struct kvmppc_vcore *pvc, *vcnext;
2514 spin_lock(&lp->lock);
2515 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2516 if (!spin_trylock(&pvc->lock))
2518 prepare_threads(pvc);
2519 if (!pvc->n_runnable) {
2520 list_del_init(&pvc->preempt_list);
2521 if (pvc->runner == NULL) {
2522 pvc->vcore_state = VCORE_INACTIVE;
2523 kvmppc_core_end_stolen(pvc);
2525 spin_unlock(&pvc->lock);
2528 if (!can_piggyback(pvc, cip, target_threads)) {
2529 spin_unlock(&pvc->lock);
2532 kvmppc_core_end_stolen(pvc);
2533 pvc->vcore_state = VCORE_PIGGYBACK;
2534 if (cip->total_threads >= target_threads)
2537 spin_unlock(&lp->lock);
2540 static bool recheck_signals(struct core_info *cip)
2543 struct kvm_vcpu *vcpu;
2545 for (sub = 0; sub < cip->n_subcores; ++sub)
2546 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2547 if (signal_pending(vcpu->arch.run_task))
2552 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2554 int still_running = 0, i;
2557 struct kvm_vcpu *vcpu;
2559 spin_lock(&vc->lock);
2561 for_each_runnable_thread(i, vcpu, vc) {
2562 /* cancel pending dec exception if dec is positive */
2563 if (now < vcpu->arch.dec_expires &&
2564 kvmppc_core_pending_dec(vcpu))
2565 kvmppc_core_dequeue_dec(vcpu);
2567 trace_kvm_guest_exit(vcpu);
2570 if (vcpu->arch.trap)
2571 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2572 vcpu->arch.run_task);
2574 vcpu->arch.ret = ret;
2575 vcpu->arch.trap = 0;
2577 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2578 if (vcpu->arch.pending_exceptions)
2579 kvmppc_core_prepare_to_enter(vcpu);
2580 if (vcpu->arch.ceded)
2581 kvmppc_set_timer(vcpu);
2585 kvmppc_remove_runnable(vc, vcpu);
2586 wake_up(&vcpu->arch.cpu_run);
2590 if (still_running > 0) {
2591 kvmppc_vcore_preempt(vc);
2592 } else if (vc->runner) {
2593 vc->vcore_state = VCORE_PREEMPT;
2594 kvmppc_core_start_stolen(vc);
2596 vc->vcore_state = VCORE_INACTIVE;
2598 if (vc->n_runnable > 0 && vc->runner == NULL) {
2599 /* make sure there's a candidate runner awake */
2601 vcpu = next_runnable_thread(vc, &i);
2602 wake_up(&vcpu->arch.cpu_run);
2605 spin_unlock(&vc->lock);
2609 * Clear core from the list of active host cores as we are about to
2610 * enter the guest. Only do this if it is the primary thread of the
2611 * core (not if a subcore) that is entering the guest.
2613 static inline int kvmppc_clear_host_core(unsigned int cpu)
2617 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2620 * Memory barrier can be omitted here as we will do a smp_wmb()
2621 * later in kvmppc_start_thread and we need ensure that state is
2622 * visible to other CPUs only after we enter guest.
2624 core = cpu >> threads_shift;
2625 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2630 * Advertise this core as an active host core since we exited the guest
2631 * Only need to do this if it is the primary thread of the core that is
2634 static inline int kvmppc_set_host_core(unsigned int cpu)
2638 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2642 * Memory barrier can be omitted here because we do a spin_unlock
2643 * immediately after this which provides the memory barrier.
2645 core = cpu >> threads_shift;
2646 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2650 static void set_irq_happened(int trap)
2653 case BOOK3S_INTERRUPT_EXTERNAL:
2654 local_paca->irq_happened |= PACA_IRQ_EE;
2656 case BOOK3S_INTERRUPT_H_DOORBELL:
2657 local_paca->irq_happened |= PACA_IRQ_DBELL;
2659 case BOOK3S_INTERRUPT_HMI:
2660 local_paca->irq_happened |= PACA_IRQ_HMI;
2662 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2663 replay_system_reset();
2669 * Run a set of guest threads on a physical core.
2670 * Called with vc->lock held.
2672 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2674 struct kvm_vcpu *vcpu;
2677 struct core_info core_info;
2678 struct kvmppc_vcore *pvc;
2679 struct kvm_split_mode split_info, *sip;
2680 int split, subcore_size, active;
2683 unsigned long cmd_bit, stat_bit;
2686 int controlled_threads;
2692 * Remove from the list any threads that have a signal pending
2693 * or need a VPA update done
2695 prepare_threads(vc);
2697 /* if the runner is no longer runnable, let the caller pick a new one */
2698 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2704 init_vcore_to_run(vc);
2705 vc->preempt_tb = TB_NIL;
2708 * Number of threads that we will be controlling: the same as
2709 * the number of threads per subcore, except on POWER9,
2710 * where it's 1 because the threads are (mostly) independent.
2712 controlled_threads = threads_per_vcore(vc->kvm);
2715 * Make sure we are running on primary threads, and that secondary
2716 * threads are offline. Also check if the number of threads in this
2717 * guest are greater than the current system threads per guest.
2718 * On POWER9, we need to be not in independent-threads mode if
2719 * this is a HPT guest on a radix host machine where the
2720 * CPU threads may not be in different MMU modes.
2722 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
2723 !kvm_is_radix(vc->kvm);
2724 if (((controlled_threads > 1) &&
2725 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2726 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2727 for_each_runnable_thread(i, vcpu, vc) {
2728 vcpu->arch.ret = -EBUSY;
2729 kvmppc_remove_runnable(vc, vcpu);
2730 wake_up(&vcpu->arch.cpu_run);
2736 * See if we could run any other vcores on the physical core
2737 * along with this one.
2739 init_core_info(&core_info, vc);
2740 pcpu = smp_processor_id();
2741 target_threads = controlled_threads;
2742 if (target_smt_mode && target_smt_mode < target_threads)
2743 target_threads = target_smt_mode;
2744 if (vc->num_threads < target_threads)
2745 collect_piggybacks(&core_info, target_threads);
2748 * On radix, arrange for TLB flushing if necessary.
2749 * This has to be done before disabling interrupts since
2750 * it uses smp_call_function().
2752 pcpu = smp_processor_id();
2753 if (kvm_is_radix(vc->kvm)) {
2754 for (sub = 0; sub < core_info.n_subcores; ++sub)
2755 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2756 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2760 * Hard-disable interrupts, and check resched flag and signals.
2761 * If we need to reschedule or deliver a signal, clean up
2762 * and return without going into the guest(s).
2763 * If the mmu_ready flag has been cleared, don't go into the
2764 * guest because that means a HPT resize operation is in progress.
2766 local_irq_disable();
2768 if (lazy_irq_pending() || need_resched() ||
2769 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
2771 vc->vcore_state = VCORE_INACTIVE;
2772 /* Unlock all except the primary vcore */
2773 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2774 pvc = core_info.vc[sub];
2775 /* Put back on to the preempted vcores list */
2776 kvmppc_vcore_preempt(pvc);
2777 spin_unlock(&pvc->lock);
2779 for (i = 0; i < controlled_threads; ++i)
2780 kvmppc_release_hwthread(pcpu + i);
2784 kvmppc_clear_host_core(pcpu);
2786 /* Decide on micro-threading (split-core) mode */
2787 subcore_size = threads_per_subcore;
2788 cmd_bit = stat_bit = 0;
2789 split = core_info.n_subcores;
2791 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2792 && !cpu_has_feature(CPU_FTR_ARCH_300);
2794 if (split > 1 || hpt_on_radix) {
2796 memset(&split_info, 0, sizeof(split_info));
2797 for (sub = 0; sub < core_info.n_subcores; ++sub)
2798 split_info.vc[sub] = core_info.vc[sub];
2801 if (split == 2 && (dynamic_mt_modes & 2)) {
2802 cmd_bit = HID0_POWER8_1TO2LPAR;
2803 stat_bit = HID0_POWER8_2LPARMODE;
2806 cmd_bit = HID0_POWER8_1TO4LPAR;
2807 stat_bit = HID0_POWER8_4LPARMODE;
2809 subcore_size = MAX_SMT_THREADS / split;
2810 split_info.rpr = mfspr(SPRN_RPR);
2811 split_info.pmmar = mfspr(SPRN_PMMAR);
2812 split_info.ldbar = mfspr(SPRN_LDBAR);
2813 split_info.subcore_size = subcore_size;
2815 split_info.subcore_size = 1;
2817 /* Use the split_info for LPCR/LPIDR changes */
2818 split_info.lpcr_req = vc->lpcr;
2819 split_info.lpidr_req = vc->kvm->arch.lpid;
2820 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2821 split_info.do_set = 1;
2825 /* order writes to split_info before kvm_split_mode pointer */
2829 for (thr = 0; thr < controlled_threads; ++thr) {
2830 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2832 paca->kvm_hstate.tid = thr;
2833 paca->kvm_hstate.napping = 0;
2834 paca->kvm_hstate.kvm_split_mode = sip;
2837 /* Initiate micro-threading (split-core) on POWER8 if required */
2839 unsigned long hid0 = mfspr(SPRN_HID0);
2841 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2843 mtspr(SPRN_HID0, hid0);
2846 hid0 = mfspr(SPRN_HID0);
2847 if (hid0 & stat_bit)
2853 /* Start all the threads */
2855 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2856 thr = is_power8 ? subcore_thread_map[sub] : sub;
2859 pvc = core_info.vc[sub];
2860 pvc->pcpu = pcpu + thr;
2861 for_each_runnable_thread(i, vcpu, pvc) {
2862 kvmppc_start_thread(vcpu, pvc);
2863 kvmppc_create_dtl_entry(vcpu, pvc);
2864 trace_kvm_guest_enter(vcpu);
2865 if (!vcpu->arch.ptid)
2867 active |= 1 << (thr + vcpu->arch.ptid);
2870 * We need to start the first thread of each subcore
2871 * even if it doesn't have a vcpu.
2874 kvmppc_start_thread(NULL, pvc);
2878 * Ensure that split_info.do_nap is set after setting
2879 * the vcore pointer in the PACA of the secondaries.
2884 * When doing micro-threading, poke the inactive threads as well.
2885 * This gets them to the nap instruction after kvm_do_nap,
2886 * which reduces the time taken to unsplit later.
2887 * For POWER9 HPT guest on radix host, we need all the secondary
2888 * threads woken up so they can do the LPCR/LPIDR change.
2890 if (cmd_bit || hpt_on_radix) {
2891 split_info.do_nap = 1; /* ask secondaries to nap when done */
2892 for (thr = 1; thr < threads_per_subcore; ++thr)
2893 if (!(active & (1 << thr)))
2894 kvmppc_ipi_thread(pcpu + thr);
2897 vc->vcore_state = VCORE_RUNNING;
2900 trace_kvmppc_run_core(vc, 0);
2902 for (sub = 0; sub < core_info.n_subcores; ++sub)
2903 spin_unlock(&core_info.vc[sub]->lock);
2906 * Interrupts will be enabled once we get into the guest,
2907 * so tell lockdep that we're about to enable interrupts.
2909 trace_hardirqs_on();
2911 guest_enter_irqoff();
2913 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2915 this_cpu_disable_ftrace();
2917 trap = __kvmppc_vcore_entry();
2919 this_cpu_enable_ftrace();
2921 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2923 trace_hardirqs_off();
2924 set_irq_happened(trap);
2926 spin_lock(&vc->lock);
2927 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2928 vc->vcore_state = VCORE_EXITING;
2930 /* wait for secondary threads to finish writing their state to memory */
2931 kvmppc_wait_for_nap(controlled_threads);
2933 /* Return to whole-core mode if we split the core earlier */
2935 unsigned long hid0 = mfspr(SPRN_HID0);
2936 unsigned long loops = 0;
2938 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2939 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2941 mtspr(SPRN_HID0, hid0);
2944 hid0 = mfspr(SPRN_HID0);
2945 if (!(hid0 & stat_bit))
2950 } else if (hpt_on_radix) {
2951 /* Wait for all threads to have seen final sync */
2952 for (thr = 1; thr < controlled_threads; ++thr) {
2953 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2955 while (paca->kvm_hstate.kvm_split_mode) {
2962 split_info.do_nap = 0;
2964 kvmppc_set_host_core(pcpu);
2969 /* Let secondaries go back to the offline loop */
2970 for (i = 0; i < controlled_threads; ++i) {
2971 kvmppc_release_hwthread(pcpu + i);
2972 if (sip && sip->napped[i])
2973 kvmppc_ipi_thread(pcpu + i);
2974 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2977 spin_unlock(&vc->lock);
2979 /* make sure updates to secondary vcpu structs are visible now */
2984 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2985 pvc = core_info.vc[sub];
2986 post_guest_process(pvc, pvc == vc);
2989 spin_lock(&vc->lock);
2992 vc->vcore_state = VCORE_INACTIVE;
2993 trace_kvmppc_run_core(vc, 1);
2997 * Wait for some other vcpu thread to execute us, and
2998 * wake us up when we need to handle something in the host.
3000 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3001 struct kvm_vcpu *vcpu, int wait_state)
3005 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3006 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3007 spin_unlock(&vc->lock);
3009 spin_lock(&vc->lock);
3011 finish_wait(&vcpu->arch.cpu_run, &wait);
3014 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3017 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3018 vc->halt_poll_ns = 10000;
3020 vc->halt_poll_ns *= halt_poll_ns_grow;
3023 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3025 if (halt_poll_ns_shrink == 0)
3026 vc->halt_poll_ns = 0;
3028 vc->halt_poll_ns /= halt_poll_ns_shrink;
3031 #ifdef CONFIG_KVM_XICS
3032 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3034 if (!xive_enabled())
3036 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3037 vcpu->arch.xive_saved_state.cppr;
3040 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3044 #endif /* CONFIG_KVM_XICS */
3046 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3048 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3049 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3056 * Check to see if any of the runnable vcpus on the vcore have pending
3057 * exceptions or are no longer ceded
3059 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3061 struct kvm_vcpu *vcpu;
3064 for_each_runnable_thread(i, vcpu, vc) {
3065 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3073 * All the vcpus in this vcore are idle, so wait for a decrementer
3074 * or external interrupt to one of the vcpus. vc->lock is held.
3076 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3078 ktime_t cur, start_poll, start_wait;
3081 DECLARE_SWAITQUEUE(wait);
3083 /* Poll for pending exceptions and ceded state */
3084 cur = start_poll = ktime_get();
3085 if (vc->halt_poll_ns) {
3086 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3087 ++vc->runner->stat.halt_attempted_poll;
3089 vc->vcore_state = VCORE_POLLING;
3090 spin_unlock(&vc->lock);
3093 if (kvmppc_vcore_check_block(vc)) {
3098 } while (single_task_running() && ktime_before(cur, stop));
3100 spin_lock(&vc->lock);
3101 vc->vcore_state = VCORE_INACTIVE;
3104 ++vc->runner->stat.halt_successful_poll;
3109 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3111 if (kvmppc_vcore_check_block(vc)) {
3112 finish_swait(&vc->wq, &wait);
3114 /* If we polled, count this as a successful poll */
3115 if (vc->halt_poll_ns)
3116 ++vc->runner->stat.halt_successful_poll;
3120 start_wait = ktime_get();
3122 vc->vcore_state = VCORE_SLEEPING;
3123 trace_kvmppc_vcore_blocked(vc, 0);
3124 spin_unlock(&vc->lock);
3126 finish_swait(&vc->wq, &wait);
3127 spin_lock(&vc->lock);
3128 vc->vcore_state = VCORE_INACTIVE;
3129 trace_kvmppc_vcore_blocked(vc, 1);
3130 ++vc->runner->stat.halt_successful_wait;
3135 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3137 /* Attribute wait time */
3139 vc->runner->stat.halt_wait_ns +=
3140 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3141 /* Attribute failed poll time */
3142 if (vc->halt_poll_ns)
3143 vc->runner->stat.halt_poll_fail_ns +=
3144 ktime_to_ns(start_wait) -
3145 ktime_to_ns(start_poll);
3147 /* Attribute successful poll time */
3148 if (vc->halt_poll_ns)
3149 vc->runner->stat.halt_poll_success_ns +=
3151 ktime_to_ns(start_poll);
3154 /* Adjust poll time */
3156 if (block_ns <= vc->halt_poll_ns)
3158 /* We slept and blocked for longer than the max halt time */
3159 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3160 shrink_halt_poll_ns(vc);
3161 /* We slept and our poll time is too small */
3162 else if (vc->halt_poll_ns < halt_poll_ns &&
3163 block_ns < halt_poll_ns)
3164 grow_halt_poll_ns(vc);
3165 if (vc->halt_poll_ns > halt_poll_ns)
3166 vc->halt_poll_ns = halt_poll_ns;
3168 vc->halt_poll_ns = 0;
3170 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3173 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3176 struct kvm *kvm = vcpu->kvm;
3178 mutex_lock(&kvm->lock);
3179 if (!kvm->arch.mmu_ready) {
3180 if (!kvm_is_radix(kvm))
3181 r = kvmppc_hv_setup_htab_rma(vcpu);
3183 if (cpu_has_feature(CPU_FTR_ARCH_300))
3184 kvmppc_setup_partition_table(kvm);
3185 kvm->arch.mmu_ready = 1;
3188 mutex_unlock(&kvm->lock);
3192 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3195 struct kvmppc_vcore *vc;
3198 trace_kvmppc_run_vcpu_enter(vcpu);
3200 kvm_run->exit_reason = 0;
3201 vcpu->arch.ret = RESUME_GUEST;
3202 vcpu->arch.trap = 0;
3203 kvmppc_update_vpas(vcpu);
3206 * Synchronize with other threads in this virtual core
3208 vc = vcpu->arch.vcore;
3209 spin_lock(&vc->lock);
3210 vcpu->arch.ceded = 0;
3211 vcpu->arch.run_task = current;
3212 vcpu->arch.kvm_run = kvm_run;
3213 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3214 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3215 vcpu->arch.busy_preempt = TB_NIL;
3216 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3220 * This happens the first time this is called for a vcpu.
3221 * If the vcore is already running, we may be able to start
3222 * this thread straight away and have it join in.
3224 if (!signal_pending(current)) {
3225 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3226 vc->vcore_state == VCORE_RUNNING) &&
3227 !VCORE_IS_EXITING(vc)) {
3228 kvmppc_create_dtl_entry(vcpu, vc);
3229 kvmppc_start_thread(vcpu, vc);
3230 trace_kvm_guest_enter(vcpu);
3231 } else if (vc->vcore_state == VCORE_SLEEPING) {
3237 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3238 !signal_pending(current)) {
3239 /* See if the MMU is ready to go */
3240 if (!vcpu->kvm->arch.mmu_ready) {
3241 spin_unlock(&vc->lock);
3242 r = kvmhv_setup_mmu(vcpu);
3243 spin_lock(&vc->lock);
3245 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3246 kvm_run->fail_entry.
3247 hardware_entry_failure_reason = 0;
3253 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3254 kvmppc_vcore_end_preempt(vc);
3256 if (vc->vcore_state != VCORE_INACTIVE) {
3257 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3260 for_each_runnable_thread(i, v, vc) {
3261 kvmppc_core_prepare_to_enter(v);
3262 if (signal_pending(v->arch.run_task)) {
3263 kvmppc_remove_runnable(vc, v);
3264 v->stat.signal_exits++;
3265 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3266 v->arch.ret = -EINTR;
3267 wake_up(&v->arch.cpu_run);
3270 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3273 for_each_runnable_thread(i, v, vc) {
3274 if (!kvmppc_vcpu_woken(v))
3275 n_ceded += v->arch.ceded;
3280 if (n_ceded == vc->n_runnable) {
3281 kvmppc_vcore_blocked(vc);
3282 } else if (need_resched()) {
3283 kvmppc_vcore_preempt(vc);
3284 /* Let something else run */
3285 cond_resched_lock(&vc->lock);
3286 if (vc->vcore_state == VCORE_PREEMPT)
3287 kvmppc_vcore_end_preempt(vc);
3289 kvmppc_run_core(vc);
3294 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3295 (vc->vcore_state == VCORE_RUNNING ||
3296 vc->vcore_state == VCORE_EXITING ||
3297 vc->vcore_state == VCORE_PIGGYBACK))
3298 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3300 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3301 kvmppc_vcore_end_preempt(vc);
3303 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3304 kvmppc_remove_runnable(vc, vcpu);
3305 vcpu->stat.signal_exits++;
3306 kvm_run->exit_reason = KVM_EXIT_INTR;
3307 vcpu->arch.ret = -EINTR;
3310 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3311 /* Wake up some vcpu to run the core */
3313 v = next_runnable_thread(vc, &i);
3314 wake_up(&v->arch.cpu_run);
3317 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3318 spin_unlock(&vc->lock);
3319 return vcpu->arch.ret;
3322 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3326 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3327 unsigned long user_tar = 0;
3328 unsigned int user_vrsave;
3331 if (!vcpu->arch.sane) {
3332 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3337 * Don't allow entry with a suspended transaction, because
3338 * the guest entry/exit code will lose it.
3339 * If the guest has TM enabled, save away their TM-related SPRs
3340 * (they will get restored by the TM unavailable interrupt).
3342 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3343 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3344 (current->thread.regs->msr & MSR_TM)) {
3345 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3346 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3347 run->fail_entry.hardware_entry_failure_reason = 0;
3350 /* Enable TM so we can read the TM SPRs */
3351 mtmsr(mfmsr() | MSR_TM);
3352 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3353 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3354 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3355 current->thread.regs->msr &= ~MSR_TM;
3359 kvmppc_core_prepare_to_enter(vcpu);
3361 /* No need to go into the guest when all we'll do is come back out */
3362 if (signal_pending(current)) {
3363 run->exit_reason = KVM_EXIT_INTR;
3368 atomic_inc(&kvm->arch.vcpus_running);
3369 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3372 flush_all_to_thread(current);
3374 /* Save userspace EBB and other register values */
3375 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3376 ebb_regs[0] = mfspr(SPRN_EBBHR);
3377 ebb_regs[1] = mfspr(SPRN_EBBRR);
3378 ebb_regs[2] = mfspr(SPRN_BESCR);
3379 user_tar = mfspr(SPRN_TAR);
3381 user_vrsave = mfspr(SPRN_VRSAVE);
3383 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3384 vcpu->arch.pgdir = current->mm->pgd;
3385 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3388 r = kvmppc_run_vcpu(run, vcpu);
3390 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3391 !(vcpu->arch.shregs.msr & MSR_PR)) {
3392 trace_kvm_hcall_enter(vcpu);
3393 r = kvmppc_pseries_do_hcall(vcpu);
3394 trace_kvm_hcall_exit(vcpu, r);
3395 kvmppc_core_prepare_to_enter(vcpu);
3396 } else if (r == RESUME_PAGE_FAULT) {
3397 srcu_idx = srcu_read_lock(&kvm->srcu);
3398 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3399 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3400 srcu_read_unlock(&kvm->srcu, srcu_idx);
3401 } else if (r == RESUME_PASSTHROUGH) {
3402 if (WARN_ON(xive_enabled()))
3405 r = kvmppc_xics_rm_complete(vcpu, 0);
3407 } while (is_kvmppc_resume_guest(r));
3409 /* Restore userspace EBB and other register values */
3410 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3411 mtspr(SPRN_EBBHR, ebb_regs[0]);
3412 mtspr(SPRN_EBBRR, ebb_regs[1]);
3413 mtspr(SPRN_BESCR, ebb_regs[2]);
3414 mtspr(SPRN_TAR, user_tar);
3415 mtspr(SPRN_FSCR, current->thread.fscr);
3417 mtspr(SPRN_VRSAVE, user_vrsave);
3419 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3420 atomic_dec(&kvm->arch.vcpus_running);
3424 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3425 int shift, int sllp)
3427 (*sps)->page_shift = shift;
3428 (*sps)->slb_enc = sllp;
3429 (*sps)->enc[0].page_shift = shift;
3430 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3432 * Add 16MB MPSS support (may get filtered out by userspace)
3435 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3437 (*sps)->enc[1].page_shift = 24;
3438 (*sps)->enc[1].pte_enc = penc;
3444 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3445 struct kvm_ppc_smmu_info *info)
3447 struct kvm_ppc_one_seg_page_size *sps;
3450 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3451 * POWER7 doesn't support keys for instruction accesses,
3452 * POWER8 and POWER9 do.
3454 info->data_keys = 32;
3455 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3457 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3458 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3459 info->slb_size = 32;
3461 /* We only support these sizes for now, and no muti-size segments */
3462 sps = &info->sps[0];
3463 kvmppc_add_seg_page_size(&sps, 12, 0);
3464 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3465 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3471 * Get (and clear) the dirty memory log for a memory slot.
3473 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3474 struct kvm_dirty_log *log)
3476 struct kvm_memslots *slots;
3477 struct kvm_memory_slot *memslot;
3480 unsigned long *buf, *p;
3481 struct kvm_vcpu *vcpu;
3483 mutex_lock(&kvm->slots_lock);
3486 if (log->slot >= KVM_USER_MEM_SLOTS)
3489 slots = kvm_memslots(kvm);
3490 memslot = id_to_memslot(slots, log->slot);
3492 if (!memslot->dirty_bitmap)
3496 * Use second half of bitmap area because both HPT and radix
3497 * accumulate bits in the first half.
3499 n = kvm_dirty_bitmap_bytes(memslot);
3500 buf = memslot->dirty_bitmap + n / sizeof(long);
3503 if (kvm_is_radix(kvm))
3504 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3506 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3511 * We accumulate dirty bits in the first half of the
3512 * memslot's dirty_bitmap area, for when pages are paged
3513 * out or modified by the host directly. Pick up these
3514 * bits and add them to the map.
3516 p = memslot->dirty_bitmap;
3517 for (i = 0; i < n / sizeof(long); ++i)
3518 buf[i] |= xchg(&p[i], 0);
3520 /* Harvest dirty bits from VPA and DTL updates */
3521 /* Note: we never modify the SLB shadow buffer areas */
3522 kvm_for_each_vcpu(i, vcpu, kvm) {
3523 spin_lock(&vcpu->arch.vpa_update_lock);
3524 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3525 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3526 spin_unlock(&vcpu->arch.vpa_update_lock);
3530 if (copy_to_user(log->dirty_bitmap, buf, n))
3535 mutex_unlock(&kvm->slots_lock);
3539 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3540 struct kvm_memory_slot *dont)
3542 if (!dont || free->arch.rmap != dont->arch.rmap) {
3543 vfree(free->arch.rmap);
3544 free->arch.rmap = NULL;
3548 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3549 unsigned long npages)
3551 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
3552 if (!slot->arch.rmap)
3558 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3559 struct kvm_memory_slot *memslot,
3560 const struct kvm_userspace_memory_region *mem)
3565 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3566 const struct kvm_userspace_memory_region *mem,
3567 const struct kvm_memory_slot *old,
3568 const struct kvm_memory_slot *new)
3570 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3573 * If we are making a new memslot, it might make
3574 * some address that was previously cached as emulated
3575 * MMIO be no longer emulated MMIO, so invalidate
3576 * all the caches of emulated MMIO translations.
3579 atomic64_inc(&kvm->arch.mmio_update);
3583 * Update LPCR values in kvm->arch and in vcores.
3584 * Caller must hold kvm->lock.
3586 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3591 if ((kvm->arch.lpcr & mask) == lpcr)
3594 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3596 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3597 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3600 spin_lock(&vc->lock);
3601 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3602 spin_unlock(&vc->lock);
3603 if (++cores_done >= kvm->arch.online_vcores)
3608 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3613 void kvmppc_setup_partition_table(struct kvm *kvm)
3615 unsigned long dw0, dw1;
3617 if (!kvm_is_radix(kvm)) {
3618 /* PS field - page size for VRMA */
3619 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3620 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3621 /* HTABSIZE and HTABORG fields */
3622 dw0 |= kvm->arch.sdr1;
3624 /* Second dword as set by userspace */
3625 dw1 = kvm->arch.process_table;
3627 dw0 = PATB_HR | radix__get_tree_size() |
3628 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3629 dw1 = PATB_GR | kvm->arch.process_table;
3632 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3636 * Set up HPT (hashed page table) and RMA (real-mode area).
3637 * Must be called with kvm->lock held.
3639 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3642 struct kvm *kvm = vcpu->kvm;
3644 struct kvm_memory_slot *memslot;
3645 struct vm_area_struct *vma;
3646 unsigned long lpcr = 0, senc;
3647 unsigned long psize, porder;
3650 /* Allocate hashed page table (if not done already) and reset it */
3651 if (!kvm->arch.hpt.virt) {
3652 int order = KVM_DEFAULT_HPT_ORDER;
3653 struct kvm_hpt_info info;
3655 err = kvmppc_allocate_hpt(&info, order);
3656 /* If we get here, it means userspace didn't specify a
3657 * size explicitly. So, try successively smaller
3658 * sizes if the default failed. */
3659 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3660 err = kvmppc_allocate_hpt(&info, order);
3663 pr_err("KVM: Couldn't alloc HPT\n");
3667 kvmppc_set_hpt(kvm, &info);
3670 /* Look up the memslot for guest physical address 0 */
3671 srcu_idx = srcu_read_lock(&kvm->srcu);
3672 memslot = gfn_to_memslot(kvm, 0);
3674 /* We must have some memory at 0 by now */
3676 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3679 /* Look up the VMA for the start of this memory slot */
3680 hva = memslot->userspace_addr;
3681 down_read(¤t->mm->mmap_sem);
3682 vma = find_vma(current->mm, hva);
3683 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3686 psize = vma_kernel_pagesize(vma);
3688 up_read(¤t->mm->mmap_sem);
3690 /* We can handle 4k, 64k or 16M pages in the VRMA */
3691 if (psize >= 0x1000000)
3693 else if (psize >= 0x10000)
3697 porder = __ilog2(psize);
3699 senc = slb_pgsize_encoding(psize);
3700 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3701 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3702 /* Create HPTEs in the hash page table for the VRMA */
3703 kvmppc_map_vrma(vcpu, memslot, porder);
3705 /* Update VRMASD field in the LPCR */
3706 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3707 /* the -4 is to account for senc values starting at 0x10 */
3708 lpcr = senc << (LPCR_VRMASD_SH - 4);
3709 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3712 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3716 srcu_read_unlock(&kvm->srcu, srcu_idx);
3721 up_read(¤t->mm->mmap_sem);
3725 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3726 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3728 kvmppc_free_radix(kvm);
3729 kvmppc_update_lpcr(kvm, LPCR_VPM1,
3730 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3731 kvmppc_rmap_reset(kvm);
3732 kvm->arch.radix = 0;
3733 kvm->arch.process_table = 0;
3737 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3738 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3742 err = kvmppc_init_vm_radix(kvm);
3746 kvmppc_free_hpt(&kvm->arch.hpt);
3747 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3748 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3749 kvm->arch.radix = 1;
3753 #ifdef CONFIG_KVM_XICS
3755 * Allocate a per-core structure for managing state about which cores are
3756 * running in the host versus the guest and for exchanging data between
3757 * real mode KVM and CPU running in the host.
3758 * This is only done for the first VM.
3759 * The allocated structure stays even if all VMs have stopped.
3760 * It is only freed when the kvm-hv module is unloaded.
3761 * It's OK for this routine to fail, we just don't support host
3762 * core operations like redirecting H_IPI wakeups.
3764 void kvmppc_alloc_host_rm_ops(void)
3766 struct kvmppc_host_rm_ops *ops;
3767 unsigned long l_ops;
3771 /* Not the first time here ? */
3772 if (kvmppc_host_rm_ops_hv != NULL)
3775 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3779 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3780 ops->rm_core = kzalloc(size, GFP_KERNEL);
3782 if (!ops->rm_core) {
3789 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3790 if (!cpu_online(cpu))
3793 core = cpu >> threads_shift;
3794 ops->rm_core[core].rm_state.in_host = 1;
3797 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3800 * Make the contents of the kvmppc_host_rm_ops structure visible
3801 * to other CPUs before we assign it to the global variable.
3802 * Do an atomic assignment (no locks used here), but if someone
3803 * beats us to it, just free our copy and return.
3806 l_ops = (unsigned long) ops;
3808 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3810 kfree(ops->rm_core);
3815 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3816 "ppc/kvm_book3s:prepare",
3817 kvmppc_set_host_core,
3818 kvmppc_clear_host_core);
3822 void kvmppc_free_host_rm_ops(void)
3824 if (kvmppc_host_rm_ops_hv) {
3825 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3826 kfree(kvmppc_host_rm_ops_hv->rm_core);
3827 kfree(kvmppc_host_rm_ops_hv);
3828 kvmppc_host_rm_ops_hv = NULL;
3833 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3835 unsigned long lpcr, lpid;
3839 /* Allocate the guest's logical partition ID */
3841 lpid = kvmppc_alloc_lpid();
3844 kvm->arch.lpid = lpid;
3846 kvmppc_alloc_host_rm_ops();
3849 * Since we don't flush the TLB when tearing down a VM,
3850 * and this lpid might have previously been used,
3851 * make sure we flush on each core before running the new VM.
3852 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3853 * does this flush for us.
3855 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3856 cpumask_setall(&kvm->arch.need_tlb_flush);
3858 /* Start out with the default set of hcalls enabled */
3859 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3860 sizeof(kvm->arch.enabled_hcalls));
3862 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3863 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3865 /* Init LPCR for virtual RMA mode */
3866 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3867 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3868 lpcr &= LPCR_PECE | LPCR_LPES;
3869 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3870 LPCR_VPM0 | LPCR_VPM1;
3871 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3872 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3873 /* On POWER8 turn on online bit to enable PURR/SPURR */
3874 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3877 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3878 * Set HVICE bit to enable hypervisor virtualization interrupts.
3879 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3880 * be unnecessary but better safe than sorry in case we re-enable
3881 * EE in HV mode with this LPCR still set)
3883 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3885 lpcr |= LPCR_HVICE | LPCR_HEIC;
3888 * If xive is enabled, we route 0x500 interrupts directly
3896 * If the host uses radix, the guest starts out as radix.
3898 if (radix_enabled()) {
3899 kvm->arch.radix = 1;
3900 kvm->arch.mmu_ready = 1;
3902 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3903 ret = kvmppc_init_vm_radix(kvm);
3905 kvmppc_free_lpid(kvm->arch.lpid);
3908 kvmppc_setup_partition_table(kvm);
3911 kvm->arch.lpcr = lpcr;
3913 /* Initialization for future HPT resizes */
3914 kvm->arch.resize_hpt = NULL;
3917 * Work out how many sets the TLB has, for the use of
3918 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3920 if (radix_enabled())
3921 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
3922 else if (cpu_has_feature(CPU_FTR_ARCH_300))
3923 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
3924 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3925 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
3927 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
3930 * Track that we now have a HV mode VM active. This blocks secondary
3931 * CPU threads from coming online.
3932 * On POWER9, we only need to do this if the "indep_threads_mode"
3933 * module parameter has been set to N.
3935 if (cpu_has_feature(CPU_FTR_ARCH_300))
3936 kvm->arch.threads_indep = indep_threads_mode;
3937 if (!kvm->arch.threads_indep)
3938 kvm_hv_vm_activated();
3941 * Initialize smt_mode depending on processor.
3942 * POWER8 and earlier have to use "strict" threading, where
3943 * all vCPUs in a vcore have to run on the same (sub)core,
3944 * whereas on POWER9 the threads can each run a different
3947 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3948 kvm->arch.smt_mode = threads_per_subcore;
3950 kvm->arch.smt_mode = 1;
3951 kvm->arch.emul_smt_mode = 1;
3954 * Create a debugfs directory for the VM
3956 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3957 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3958 kvmppc_mmu_debugfs_init(kvm);
3963 static void kvmppc_free_vcores(struct kvm *kvm)
3967 for (i = 0; i < KVM_MAX_VCORES; ++i)
3968 kfree(kvm->arch.vcores[i]);
3969 kvm->arch.online_vcores = 0;
3972 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3974 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3976 if (!kvm->arch.threads_indep)
3977 kvm_hv_vm_deactivated();
3979 kvmppc_free_vcores(kvm);
3981 kvmppc_free_lpid(kvm->arch.lpid);
3983 if (kvm_is_radix(kvm))
3984 kvmppc_free_radix(kvm);
3986 kvmppc_free_hpt(&kvm->arch.hpt);
3988 kvmppc_free_pimap(kvm);
3991 /* We don't need to emulate any privileged instructions or dcbz */
3992 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3993 unsigned int inst, int *advance)
3995 return EMULATE_FAIL;
3998 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4001 return EMULATE_FAIL;
4004 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4007 return EMULATE_FAIL;
4010 static int kvmppc_core_check_processor_compat_hv(void)
4012 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
4013 !cpu_has_feature(CPU_FTR_ARCH_206))
4019 #ifdef CONFIG_KVM_XICS
4021 void kvmppc_free_pimap(struct kvm *kvm)
4023 kfree(kvm->arch.pimap);
4026 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4028 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4031 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4033 struct irq_desc *desc;
4034 struct kvmppc_irq_map *irq_map;
4035 struct kvmppc_passthru_irqmap *pimap;
4036 struct irq_chip *chip;
4039 if (!kvm_irq_bypass)
4042 desc = irq_to_desc(host_irq);
4046 mutex_lock(&kvm->lock);
4048 pimap = kvm->arch.pimap;
4049 if (pimap == NULL) {
4050 /* First call, allocate structure to hold IRQ map */
4051 pimap = kvmppc_alloc_pimap();
4052 if (pimap == NULL) {
4053 mutex_unlock(&kvm->lock);
4056 kvm->arch.pimap = pimap;
4060 * For now, we only support interrupts for which the EOI operation
4061 * is an OPAL call followed by a write to XIRR, since that's
4062 * what our real-mode EOI code does, or a XIVE interrupt
4064 chip = irq_data_get_irq_chip(&desc->irq_data);
4065 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4066 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4067 host_irq, guest_gsi);
4068 mutex_unlock(&kvm->lock);
4073 * See if we already have an entry for this guest IRQ number.
4074 * If it's mapped to a hardware IRQ number, that's an error,
4075 * otherwise re-use this entry.
4077 for (i = 0; i < pimap->n_mapped; i++) {
4078 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4079 if (pimap->mapped[i].r_hwirq) {
4080 mutex_unlock(&kvm->lock);
4087 if (i == KVMPPC_PIRQ_MAPPED) {
4088 mutex_unlock(&kvm->lock);
4089 return -EAGAIN; /* table is full */
4092 irq_map = &pimap->mapped[i];
4094 irq_map->v_hwirq = guest_gsi;
4095 irq_map->desc = desc;
4098 * Order the above two stores before the next to serialize with
4099 * the KVM real mode handler.
4102 irq_map->r_hwirq = desc->irq_data.hwirq;
4104 if (i == pimap->n_mapped)
4108 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4110 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4112 irq_map->r_hwirq = 0;
4114 mutex_unlock(&kvm->lock);
4119 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4121 struct irq_desc *desc;
4122 struct kvmppc_passthru_irqmap *pimap;
4125 if (!kvm_irq_bypass)
4128 desc = irq_to_desc(host_irq);
4132 mutex_lock(&kvm->lock);
4133 if (!kvm->arch.pimap)
4136 pimap = kvm->arch.pimap;
4138 for (i = 0; i < pimap->n_mapped; i++) {
4139 if (guest_gsi == pimap->mapped[i].v_hwirq)
4143 if (i == pimap->n_mapped) {
4144 mutex_unlock(&kvm->lock);
4149 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4151 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4153 /* invalidate the entry (what do do on error from the above ?) */
4154 pimap->mapped[i].r_hwirq = 0;
4157 * We don't free this structure even when the count goes to
4158 * zero. The structure is freed when we destroy the VM.
4161 mutex_unlock(&kvm->lock);
4165 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4166 struct irq_bypass_producer *prod)
4169 struct kvm_kernel_irqfd *irqfd =
4170 container_of(cons, struct kvm_kernel_irqfd, consumer);
4172 irqfd->producer = prod;
4174 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4176 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4177 prod->irq, irqfd->gsi, ret);
4182 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4183 struct irq_bypass_producer *prod)
4186 struct kvm_kernel_irqfd *irqfd =
4187 container_of(cons, struct kvm_kernel_irqfd, consumer);
4189 irqfd->producer = NULL;
4192 * When producer of consumer is unregistered, we change back to
4193 * default external interrupt handling mode - KVM real mode
4194 * will switch back to host.
4196 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4198 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4199 prod->irq, irqfd->gsi, ret);
4203 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4204 unsigned int ioctl, unsigned long arg)
4206 struct kvm *kvm __maybe_unused = filp->private_data;
4207 void __user *argp = (void __user *)arg;
4212 case KVM_PPC_ALLOCATE_HTAB: {
4216 if (get_user(htab_order, (u32 __user *)argp))
4218 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4225 case KVM_PPC_GET_HTAB_FD: {
4226 struct kvm_get_htab_fd ghf;
4229 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4231 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4235 case KVM_PPC_RESIZE_HPT_PREPARE: {
4236 struct kvm_ppc_resize_hpt rhpt;
4239 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4242 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4246 case KVM_PPC_RESIZE_HPT_COMMIT: {
4247 struct kvm_ppc_resize_hpt rhpt;
4250 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4253 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4265 * List of hcall numbers to enable by default.
4266 * For compatibility with old userspace, we enable by default
4267 * all hcalls that were implemented before the hcall-enabling
4268 * facility was added. Note this list should not include H_RTAS.
4270 static unsigned int default_hcall_list[] = {
4284 #ifdef CONFIG_KVM_XICS
4295 static void init_default_hcalls(void)
4300 for (i = 0; default_hcall_list[i]; ++i) {
4301 hcall = default_hcall_list[i];
4302 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4303 __set_bit(hcall / 4, default_enabled_hcalls);
4307 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4313 /* If not on a POWER9, reject it */
4314 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4317 /* If any unknown flags set, reject it */
4318 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4321 /* GR (guest radix) bit in process_table field must match */
4322 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4323 if (!!(cfg->process_table & PATB_GR) != radix)
4326 /* Process table size field must be reasonable, i.e. <= 24 */
4327 if ((cfg->process_table & PRTS_MASK) > 24)
4330 /* We can change a guest to/from radix now, if the host is radix */
4331 if (radix && !radix_enabled())
4334 mutex_lock(&kvm->lock);
4335 if (radix != kvm_is_radix(kvm)) {
4336 if (kvm->arch.mmu_ready) {
4337 kvm->arch.mmu_ready = 0;
4338 /* order mmu_ready vs. vcpus_running */
4340 if (atomic_read(&kvm->arch.vcpus_running)) {
4341 kvm->arch.mmu_ready = 1;
4347 err = kvmppc_switch_mmu_to_radix(kvm);
4349 err = kvmppc_switch_mmu_to_hpt(kvm);
4354 kvm->arch.process_table = cfg->process_table;
4355 kvmppc_setup_partition_table(kvm);
4357 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4358 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4362 mutex_unlock(&kvm->lock);
4366 static struct kvmppc_ops kvm_ops_hv = {
4367 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4368 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4369 .get_one_reg = kvmppc_get_one_reg_hv,
4370 .set_one_reg = kvmppc_set_one_reg_hv,
4371 .vcpu_load = kvmppc_core_vcpu_load_hv,
4372 .vcpu_put = kvmppc_core_vcpu_put_hv,
4373 .set_msr = kvmppc_set_msr_hv,
4374 .vcpu_run = kvmppc_vcpu_run_hv,
4375 .vcpu_create = kvmppc_core_vcpu_create_hv,
4376 .vcpu_free = kvmppc_core_vcpu_free_hv,
4377 .check_requests = kvmppc_core_check_requests_hv,
4378 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4379 .flush_memslot = kvmppc_core_flush_memslot_hv,
4380 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4381 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4382 .unmap_hva_range = kvm_unmap_hva_range_hv,
4383 .age_hva = kvm_age_hva_hv,
4384 .test_age_hva = kvm_test_age_hva_hv,
4385 .set_spte_hva = kvm_set_spte_hva_hv,
4386 .mmu_destroy = kvmppc_mmu_destroy_hv,
4387 .free_memslot = kvmppc_core_free_memslot_hv,
4388 .create_memslot = kvmppc_core_create_memslot_hv,
4389 .init_vm = kvmppc_core_init_vm_hv,
4390 .destroy_vm = kvmppc_core_destroy_vm_hv,
4391 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4392 .emulate_op = kvmppc_core_emulate_op_hv,
4393 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4394 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4395 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4396 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4397 .hcall_implemented = kvmppc_hcall_impl_hv,
4398 #ifdef CONFIG_KVM_XICS
4399 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4400 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4402 .configure_mmu = kvmhv_configure_mmu,
4403 .get_rmmu_info = kvmhv_get_rmmu_info,
4404 .set_smt_mode = kvmhv_set_smt_mode,
4407 static int kvm_init_subcore_bitmap(void)
4410 int nr_cores = cpu_nr_cores();
4411 struct sibling_subcore_state *sibling_subcore_state;
4413 for (i = 0; i < nr_cores; i++) {
4414 int first_cpu = i * threads_per_core;
4415 int node = cpu_to_node(first_cpu);
4417 /* Ignore if it is already allocated. */
4418 if (paca_ptrs[first_cpu]->sibling_subcore_state)
4421 sibling_subcore_state =
4422 kmalloc_node(sizeof(struct sibling_subcore_state),
4424 if (!sibling_subcore_state)
4427 memset(sibling_subcore_state, 0,
4428 sizeof(struct sibling_subcore_state));
4430 for (j = 0; j < threads_per_core; j++) {
4431 int cpu = first_cpu + j;
4433 paca_ptrs[cpu]->sibling_subcore_state =
4434 sibling_subcore_state;
4440 static int kvmppc_radix_possible(void)
4442 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4445 static int kvmppc_book3s_init_hv(void)
4449 * FIXME!! Do we need to check on all cpus ?
4451 r = kvmppc_core_check_processor_compat_hv();
4455 r = kvm_init_subcore_bitmap();
4460 * We need a way of accessing the XICS interrupt controller,
4461 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
4462 * indirectly, via OPAL.
4465 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4466 struct device_node *np;
4468 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4470 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4476 kvm_ops_hv.owner = THIS_MODULE;
4477 kvmppc_hv_ops = &kvm_ops_hv;
4479 init_default_hcalls();
4483 r = kvmppc_mmu_hv_init();
4487 if (kvmppc_radix_possible())
4488 r = kvmppc_radix_init();
4491 * POWER9 chips before version 2.02 can't have some threads in
4492 * HPT mode and some in radix mode on the same core.
4494 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4495 unsigned int pvr = mfspr(SPRN_PVR);
4496 if ((pvr >> 16) == PVR_POWER9 &&
4497 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
4498 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
4499 no_mixing_hpt_and_radix = true;
4505 static void kvmppc_book3s_exit_hv(void)
4507 kvmppc_free_host_rm_ops();
4508 if (kvmppc_radix_possible())
4509 kvmppc_radix_exit();
4510 kvmppc_hv_ops = NULL;
4513 module_init(kvmppc_book3s_init_hv);
4514 module_exit(kvmppc_book3s_exit_hv);
4515 MODULE_LICENSE("GPL");
4516 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4517 MODULE_ALIAS("devname:kvm");