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>
49 #include <asm/ftrace.h>
51 #include <asm/ppc-opcode.h>
52 #include <asm/asm-prototypes.h>
53 #include <asm/debug.h>
54 #include <asm/disassemble.h>
55 #include <asm/cputable.h>
56 #include <asm/cacheflush.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);
127 * RWMR values for POWER8. These control the rate at which PURR
128 * and SPURR count and should be set according to the number of
129 * online threads in the vcore being run.
131 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
132 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
133 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
134 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
135 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
136 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
137 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
138 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
140 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
152 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
156 struct kvm_vcpu *vcpu;
158 while (++i < MAX_SMT_THREADS) {
159 vcpu = READ_ONCE(vc->runnable_threads[i]);
168 /* Used to traverse the list of runnable threads for a given vcore */
169 #define for_each_runnable_thread(i, vcpu, vc) \
170 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
172 static bool kvmppc_ipi_thread(int cpu)
174 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
176 /* On POWER9 we can use msgsnd to IPI any cpu */
177 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
178 msg |= get_hard_smp_processor_id(cpu);
180 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
184 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
185 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
187 if (cpu_first_thread_sibling(cpu) ==
188 cpu_first_thread_sibling(smp_processor_id())) {
189 msg |= cpu_thread_in_core(cpu);
191 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
198 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
199 if (cpu >= 0 && cpu < nr_cpu_ids) {
200 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
204 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
212 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
215 struct swait_queue_head *wqp;
217 wqp = kvm_arch_vcpu_wq(vcpu);
218 if (swq_has_sleeper(wqp)) {
220 ++vcpu->stat.halt_wakeup;
223 cpu = READ_ONCE(vcpu->arch.thread_cpu);
224 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
227 /* CPU points to the first thread of the core */
229 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
230 smp_send_reschedule(cpu);
234 * We use the vcpu_load/put functions to measure stolen time.
235 * Stolen time is counted as time when either the vcpu is able to
236 * run as part of a virtual core, but the task running the vcore
237 * is preempted or sleeping, or when the vcpu needs something done
238 * in the kernel by the task running the vcpu, but that task is
239 * preempted or sleeping. Those two things have to be counted
240 * separately, since one of the vcpu tasks will take on the job
241 * of running the core, and the other vcpu tasks in the vcore will
242 * sleep waiting for it to do that, but that sleep shouldn't count
245 * Hence we accumulate stolen time when the vcpu can run as part of
246 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
247 * needs its task to do other things in the kernel (for example,
248 * service a page fault) in busy_stolen. We don't accumulate
249 * stolen time for a vcore when it is inactive, or for a vcpu
250 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
251 * a misnomer; it means that the vcpu task is not executing in
252 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
253 * the kernel. We don't have any way of dividing up that time
254 * between time that the vcpu is genuinely stopped, time that
255 * the task is actively working on behalf of the vcpu, and time
256 * that the task is preempted, so we don't count any of it as
259 * Updates to busy_stolen are protected by arch.tbacct_lock;
260 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
261 * lock. The stolen times are measured in units of timebase ticks.
262 * (Note that the != TB_NIL checks below are purely defensive;
263 * they should never fail.)
266 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
270 spin_lock_irqsave(&vc->stoltb_lock, flags);
271 vc->preempt_tb = mftb();
272 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
275 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
279 spin_lock_irqsave(&vc->stoltb_lock, flags);
280 if (vc->preempt_tb != TB_NIL) {
281 vc->stolen_tb += mftb() - vc->preempt_tb;
282 vc->preempt_tb = TB_NIL;
284 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
287 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
289 struct kvmppc_vcore *vc = vcpu->arch.vcore;
293 * We can test vc->runner without taking the vcore lock,
294 * because only this task ever sets vc->runner to this
295 * vcpu, and once it is set to this vcpu, only this task
296 * ever sets it to NULL.
298 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
299 kvmppc_core_end_stolen(vc);
301 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
302 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
303 vcpu->arch.busy_preempt != TB_NIL) {
304 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
305 vcpu->arch.busy_preempt = TB_NIL;
307 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
310 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
312 struct kvmppc_vcore *vc = vcpu->arch.vcore;
315 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
316 kvmppc_core_start_stolen(vc);
318 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
319 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
320 vcpu->arch.busy_preempt = mftb();
321 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
324 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
327 * Check for illegal transactional state bit combination
328 * and if we find it, force the TS field to a safe state.
330 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
332 vcpu->arch.shregs.msr = msr;
333 kvmppc_end_cede(vcpu);
336 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
338 vcpu->arch.pvr = pvr;
341 /* Dummy value used in computing PCR value below */
342 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
344 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
346 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
347 struct kvmppc_vcore *vc = vcpu->arch.vcore;
349 /* We can (emulate) our own architecture version and anything older */
350 if (cpu_has_feature(CPU_FTR_ARCH_300))
351 host_pcr_bit = PCR_ARCH_300;
352 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
353 host_pcr_bit = PCR_ARCH_207;
354 else if (cpu_has_feature(CPU_FTR_ARCH_206))
355 host_pcr_bit = PCR_ARCH_206;
357 host_pcr_bit = PCR_ARCH_205;
359 /* Determine lowest PCR bit needed to run guest in given PVR level */
360 guest_pcr_bit = host_pcr_bit;
362 switch (arch_compat) {
364 guest_pcr_bit = PCR_ARCH_205;
368 guest_pcr_bit = PCR_ARCH_206;
371 guest_pcr_bit = PCR_ARCH_207;
374 guest_pcr_bit = PCR_ARCH_300;
381 /* Check requested PCR bits don't exceed our capabilities */
382 if (guest_pcr_bit > host_pcr_bit)
385 spin_lock(&vc->lock);
386 vc->arch_compat = arch_compat;
387 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
388 vc->pcr = host_pcr_bit - guest_pcr_bit;
389 spin_unlock(&vc->lock);
394 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
398 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
399 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
400 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
401 for (r = 0; r < 16; ++r)
402 pr_err("r%2d = %.16lx r%d = %.16lx\n",
403 r, kvmppc_get_gpr(vcpu, r),
404 r+16, kvmppc_get_gpr(vcpu, r+16));
405 pr_err("ctr = %.16lx lr = %.16lx\n",
406 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
407 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
408 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
409 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
410 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
411 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
412 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
413 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
414 vcpu->arch.cr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
415 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
416 pr_err("fault dar = %.16lx dsisr = %.8x\n",
417 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
418 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
419 for (r = 0; r < vcpu->arch.slb_max; ++r)
420 pr_err(" ESID = %.16llx VSID = %.16llx\n",
421 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
422 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
423 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
424 vcpu->arch.last_inst);
427 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
429 struct kvm_vcpu *ret;
431 mutex_lock(&kvm->lock);
432 ret = kvm_get_vcpu_by_id(kvm, id);
433 mutex_unlock(&kvm->lock);
437 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
439 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
440 vpa->yield_count = cpu_to_be32(1);
443 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
444 unsigned long addr, unsigned long len)
446 /* check address is cacheline aligned */
447 if (addr & (L1_CACHE_BYTES - 1))
449 spin_lock(&vcpu->arch.vpa_update_lock);
450 if (v->next_gpa != addr || v->len != len) {
452 v->len = addr ? len : 0;
453 v->update_pending = 1;
455 spin_unlock(&vcpu->arch.vpa_update_lock);
459 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
468 static int vpa_is_registered(struct kvmppc_vpa *vpap)
470 if (vpap->update_pending)
471 return vpap->next_gpa != 0;
472 return vpap->pinned_addr != NULL;
475 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
477 unsigned long vcpuid, unsigned long vpa)
479 struct kvm *kvm = vcpu->kvm;
480 unsigned long len, nb;
482 struct kvm_vcpu *tvcpu;
485 struct kvmppc_vpa *vpap;
487 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
491 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
492 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
493 subfunc == H_VPA_REG_SLB) {
494 /* Registering new area - address must be cache-line aligned */
495 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
498 /* convert logical addr to kernel addr and read length */
499 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
502 if (subfunc == H_VPA_REG_VPA)
503 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
505 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
506 kvmppc_unpin_guest_page(kvm, va, vpa, false);
509 if (len > nb || len < sizeof(struct reg_vpa))
518 spin_lock(&tvcpu->arch.vpa_update_lock);
521 case H_VPA_REG_VPA: /* register VPA */
523 * The size of our lppaca is 1kB because of the way we align
524 * it for the guest to avoid crossing a 4kB boundary. We only
525 * use 640 bytes of the structure though, so we should accept
526 * clients that set a size of 640.
528 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
529 if (len < sizeof(struct lppaca))
531 vpap = &tvcpu->arch.vpa;
535 case H_VPA_REG_DTL: /* register DTL */
536 if (len < sizeof(struct dtl_entry))
538 len -= len % sizeof(struct dtl_entry);
540 /* Check that they have previously registered a VPA */
542 if (!vpa_is_registered(&tvcpu->arch.vpa))
545 vpap = &tvcpu->arch.dtl;
549 case H_VPA_REG_SLB: /* register SLB shadow buffer */
550 /* Check that they have previously registered a VPA */
552 if (!vpa_is_registered(&tvcpu->arch.vpa))
555 vpap = &tvcpu->arch.slb_shadow;
559 case H_VPA_DEREG_VPA: /* deregister VPA */
560 /* Check they don't still have a DTL or SLB buf registered */
562 if (vpa_is_registered(&tvcpu->arch.dtl) ||
563 vpa_is_registered(&tvcpu->arch.slb_shadow))
566 vpap = &tvcpu->arch.vpa;
570 case H_VPA_DEREG_DTL: /* deregister DTL */
571 vpap = &tvcpu->arch.dtl;
575 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
576 vpap = &tvcpu->arch.slb_shadow;
582 vpap->next_gpa = vpa;
584 vpap->update_pending = 1;
587 spin_unlock(&tvcpu->arch.vpa_update_lock);
592 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
594 struct kvm *kvm = vcpu->kvm;
600 * We need to pin the page pointed to by vpap->next_gpa,
601 * but we can't call kvmppc_pin_guest_page under the lock
602 * as it does get_user_pages() and down_read(). So we
603 * have to drop the lock, pin the page, then get the lock
604 * again and check that a new area didn't get registered
608 gpa = vpap->next_gpa;
609 spin_unlock(&vcpu->arch.vpa_update_lock);
613 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
614 spin_lock(&vcpu->arch.vpa_update_lock);
615 if (gpa == vpap->next_gpa)
617 /* sigh... unpin that one and try again */
619 kvmppc_unpin_guest_page(kvm, va, gpa, false);
622 vpap->update_pending = 0;
623 if (va && nb < vpap->len) {
625 * If it's now too short, it must be that userspace
626 * has changed the mappings underlying guest memory,
627 * so unregister the region.
629 kvmppc_unpin_guest_page(kvm, va, gpa, false);
632 if (vpap->pinned_addr)
633 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
636 vpap->pinned_addr = va;
639 vpap->pinned_end = va + vpap->len;
642 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
644 if (!(vcpu->arch.vpa.update_pending ||
645 vcpu->arch.slb_shadow.update_pending ||
646 vcpu->arch.dtl.update_pending))
649 spin_lock(&vcpu->arch.vpa_update_lock);
650 if (vcpu->arch.vpa.update_pending) {
651 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
652 if (vcpu->arch.vpa.pinned_addr)
653 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
655 if (vcpu->arch.dtl.update_pending) {
656 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
657 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
658 vcpu->arch.dtl_index = 0;
660 if (vcpu->arch.slb_shadow.update_pending)
661 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
662 spin_unlock(&vcpu->arch.vpa_update_lock);
666 * Return the accumulated stolen time for the vcore up until `now'.
667 * The caller should hold the vcore lock.
669 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
674 spin_lock_irqsave(&vc->stoltb_lock, flags);
676 if (vc->vcore_state != VCORE_INACTIVE &&
677 vc->preempt_tb != TB_NIL)
678 p += now - vc->preempt_tb;
679 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
683 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
684 struct kvmppc_vcore *vc)
686 struct dtl_entry *dt;
688 unsigned long stolen;
689 unsigned long core_stolen;
693 dt = vcpu->arch.dtl_ptr;
694 vpa = vcpu->arch.vpa.pinned_addr;
696 core_stolen = vcore_stolen_time(vc, now);
697 stolen = core_stolen - vcpu->arch.stolen_logged;
698 vcpu->arch.stolen_logged = core_stolen;
699 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
700 stolen += vcpu->arch.busy_stolen;
701 vcpu->arch.busy_stolen = 0;
702 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
705 memset(dt, 0, sizeof(struct dtl_entry));
706 dt->dispatch_reason = 7;
707 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
708 dt->timebase = cpu_to_be64(now + vc->tb_offset);
709 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
710 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
711 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
713 if (dt == vcpu->arch.dtl.pinned_end)
714 dt = vcpu->arch.dtl.pinned_addr;
715 vcpu->arch.dtl_ptr = dt;
716 /* order writing *dt vs. writing vpa->dtl_idx */
718 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
719 vcpu->arch.dtl.dirty = true;
722 /* See if there is a doorbell interrupt pending for a vcpu */
723 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
726 struct kvmppc_vcore *vc;
728 if (vcpu->arch.doorbell_request)
731 * Ensure that the read of vcore->dpdes comes after the read
732 * of vcpu->doorbell_request. This barrier matches the
733 * smb_wmb() in kvmppc_guest_entry_inject().
736 vc = vcpu->arch.vcore;
737 thr = vcpu->vcpu_id - vc->first_vcpuid;
738 return !!(vc->dpdes & (1 << thr));
741 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
743 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
745 if ((!vcpu->arch.vcore->arch_compat) &&
746 cpu_has_feature(CPU_FTR_ARCH_207S))
751 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
752 unsigned long resource, unsigned long value1,
753 unsigned long value2)
756 case H_SET_MODE_RESOURCE_SET_CIABR:
757 if (!kvmppc_power8_compatible(vcpu))
762 return H_UNSUPPORTED_FLAG_START;
763 /* Guests can't breakpoint the hypervisor */
764 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
766 vcpu->arch.ciabr = value1;
768 case H_SET_MODE_RESOURCE_SET_DAWR:
769 if (!kvmppc_power8_compatible(vcpu))
771 if (!ppc_breakpoint_available())
774 return H_UNSUPPORTED_FLAG_START;
775 if (value2 & DABRX_HYP)
777 vcpu->arch.dawr = value1;
778 vcpu->arch.dawrx = value2;
785 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
787 struct kvmppc_vcore *vcore = target->arch.vcore;
790 * We expect to have been called by the real mode handler
791 * (kvmppc_rm_h_confer()) which would have directly returned
792 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
793 * have useful work to do and should not confer) so we don't
797 spin_lock(&vcore->lock);
798 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
799 vcore->vcore_state != VCORE_INACTIVE &&
801 target = vcore->runner;
802 spin_unlock(&vcore->lock);
804 return kvm_vcpu_yield_to(target);
807 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
810 struct lppaca *lppaca;
812 spin_lock(&vcpu->arch.vpa_update_lock);
813 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
815 yield_count = be32_to_cpu(lppaca->yield_count);
816 spin_unlock(&vcpu->arch.vpa_update_lock);
820 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
822 unsigned long req = kvmppc_get_gpr(vcpu, 3);
823 unsigned long target, ret = H_SUCCESS;
825 struct kvm_vcpu *tvcpu;
828 if (req <= MAX_HCALL_OPCODE &&
829 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
836 target = kvmppc_get_gpr(vcpu, 4);
837 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
842 tvcpu->arch.prodded = 1;
844 if (tvcpu->arch.ceded)
845 kvmppc_fast_vcpu_kick_hv(tvcpu);
848 target = kvmppc_get_gpr(vcpu, 4);
851 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
856 yield_count = kvmppc_get_gpr(vcpu, 5);
857 if (kvmppc_get_yield_count(tvcpu) != yield_count)
859 kvm_arch_vcpu_yield_to(tvcpu);
862 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
863 kvmppc_get_gpr(vcpu, 5),
864 kvmppc_get_gpr(vcpu, 6));
867 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
870 idx = srcu_read_lock(&vcpu->kvm->srcu);
871 rc = kvmppc_rtas_hcall(vcpu);
872 srcu_read_unlock(&vcpu->kvm->srcu, idx);
879 /* Send the error out to userspace via KVM_RUN */
881 case H_LOGICAL_CI_LOAD:
882 ret = kvmppc_h_logical_ci_load(vcpu);
883 if (ret == H_TOO_HARD)
886 case H_LOGICAL_CI_STORE:
887 ret = kvmppc_h_logical_ci_store(vcpu);
888 if (ret == H_TOO_HARD)
892 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
893 kvmppc_get_gpr(vcpu, 5),
894 kvmppc_get_gpr(vcpu, 6),
895 kvmppc_get_gpr(vcpu, 7));
896 if (ret == H_TOO_HARD)
905 if (kvmppc_xics_enabled(vcpu)) {
906 if (xive_enabled()) {
907 ret = H_NOT_AVAILABLE;
910 ret = kvmppc_xics_hcall(vcpu, req);
915 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
916 kvmppc_get_gpr(vcpu, 5),
917 kvmppc_get_gpr(vcpu, 6));
918 if (ret == H_TOO_HARD)
921 case H_PUT_TCE_INDIRECT:
922 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
923 kvmppc_get_gpr(vcpu, 5),
924 kvmppc_get_gpr(vcpu, 6),
925 kvmppc_get_gpr(vcpu, 7));
926 if (ret == H_TOO_HARD)
930 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
931 kvmppc_get_gpr(vcpu, 5),
932 kvmppc_get_gpr(vcpu, 6),
933 kvmppc_get_gpr(vcpu, 7));
934 if (ret == H_TOO_HARD)
940 kvmppc_set_gpr(vcpu, 3, ret);
941 vcpu->arch.hcall_needed = 0;
945 static int kvmppc_hcall_impl_hv(unsigned long cmd)
953 case H_LOGICAL_CI_LOAD:
954 case H_LOGICAL_CI_STORE:
955 #ifdef CONFIG_KVM_XICS
966 /* See if it's in the real-mode table */
967 return kvmppc_hcall_impl_hv_realmode(cmd);
970 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
971 struct kvm_vcpu *vcpu)
975 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
978 * Fetch failed, so return to guest and
979 * try executing it again.
984 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
985 run->exit_reason = KVM_EXIT_DEBUG;
986 run->debug.arch.address = kvmppc_get_pc(vcpu);
989 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
994 static void do_nothing(void *x)
998 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1000 int thr, cpu, pcpu, nthreads;
1002 unsigned long dpdes;
1004 nthreads = vcpu->kvm->arch.emul_smt_mode;
1006 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1007 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1008 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1012 * If the vcpu is currently running on a physical cpu thread,
1013 * interrupt it in order to pull it out of the guest briefly,
1014 * which will update its vcore->dpdes value.
1016 pcpu = READ_ONCE(v->cpu);
1018 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1019 if (kvmppc_doorbell_pending(v))
1026 * On POWER9, emulate doorbell-related instructions in order to
1027 * give the guest the illusion of running on a multi-threaded core.
1028 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1031 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1035 struct kvm *kvm = vcpu->kvm;
1036 struct kvm_vcpu *tvcpu;
1038 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1039 return RESUME_GUEST;
1040 if (get_op(inst) != 31)
1041 return EMULATE_FAIL;
1043 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1044 switch (get_xop(inst)) {
1045 case OP_31_XOP_MSGSNDP:
1046 arg = kvmppc_get_gpr(vcpu, rb);
1047 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1050 if (arg >= kvm->arch.emul_smt_mode)
1052 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1055 if (!tvcpu->arch.doorbell_request) {
1056 tvcpu->arch.doorbell_request = 1;
1057 kvmppc_fast_vcpu_kick_hv(tvcpu);
1060 case OP_31_XOP_MSGCLRP:
1061 arg = kvmppc_get_gpr(vcpu, rb);
1062 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1064 vcpu->arch.vcore->dpdes = 0;
1065 vcpu->arch.doorbell_request = 0;
1067 case OP_31_XOP_MFSPR:
1068 switch (get_sprn(inst)) {
1073 arg = kvmppc_read_dpdes(vcpu);
1076 return EMULATE_FAIL;
1078 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1081 return EMULATE_FAIL;
1083 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1084 return RESUME_GUEST;
1087 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1088 struct task_struct *tsk)
1090 int r = RESUME_HOST;
1092 vcpu->stat.sum_exits++;
1095 * This can happen if an interrupt occurs in the last stages
1096 * of guest entry or the first stages of guest exit (i.e. after
1097 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1098 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1099 * That can happen due to a bug, or due to a machine check
1100 * occurring at just the wrong time.
1102 if (vcpu->arch.shregs.msr & MSR_HV) {
1103 printk(KERN_EMERG "KVM trap in HV mode!\n");
1104 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1105 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1106 vcpu->arch.shregs.msr);
1107 kvmppc_dump_regs(vcpu);
1108 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1109 run->hw.hardware_exit_reason = vcpu->arch.trap;
1112 run->exit_reason = KVM_EXIT_UNKNOWN;
1113 run->ready_for_interrupt_injection = 1;
1114 switch (vcpu->arch.trap) {
1115 /* We're good on these - the host merely wanted to get our attention */
1116 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1117 vcpu->stat.dec_exits++;
1120 case BOOK3S_INTERRUPT_EXTERNAL:
1121 case BOOK3S_INTERRUPT_H_DOORBELL:
1122 case BOOK3S_INTERRUPT_H_VIRT:
1123 vcpu->stat.ext_intr_exits++;
1126 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1127 case BOOK3S_INTERRUPT_HMI:
1128 case BOOK3S_INTERRUPT_PERFMON:
1129 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1132 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1133 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1134 run->exit_reason = KVM_EXIT_NMI;
1135 run->hw.hardware_exit_reason = vcpu->arch.trap;
1136 /* Clear out the old NMI status from run->flags */
1137 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1138 /* Now set the NMI status */
1139 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1140 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1142 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1145 /* Print the MCE event to host console. */
1146 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1148 case BOOK3S_INTERRUPT_PROGRAM:
1152 * Normally program interrupts are delivered directly
1153 * to the guest by the hardware, but we can get here
1154 * as a result of a hypervisor emulation interrupt
1155 * (e40) getting turned into a 700 by BML RTAS.
1157 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1158 kvmppc_core_queue_program(vcpu, flags);
1162 case BOOK3S_INTERRUPT_SYSCALL:
1164 /* hcall - punt to userspace */
1167 /* hypercall with MSR_PR has already been handled in rmode,
1168 * and never reaches here.
1171 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1172 for (i = 0; i < 9; ++i)
1173 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1174 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1175 vcpu->arch.hcall_needed = 1;
1180 * We get these next two if the guest accesses a page which it thinks
1181 * it has mapped but which is not actually present, either because
1182 * it is for an emulated I/O device or because the corresonding
1183 * host page has been paged out. Any other HDSI/HISI interrupts
1184 * have been handled already.
1186 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1187 r = RESUME_PAGE_FAULT;
1189 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1190 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1191 vcpu->arch.fault_dsisr = 0;
1192 r = RESUME_PAGE_FAULT;
1195 * This occurs if the guest executes an illegal instruction.
1196 * If the guest debug is disabled, generate a program interrupt
1197 * to the guest. If guest debug is enabled, we need to check
1198 * whether the instruction is a software breakpoint instruction.
1199 * Accordingly return to Guest or Host.
1201 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1202 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1203 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1204 swab32(vcpu->arch.emul_inst) :
1205 vcpu->arch.emul_inst;
1206 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1207 r = kvmppc_emulate_debug_inst(run, vcpu);
1209 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1214 * This occurs if the guest (kernel or userspace), does something that
1215 * is prohibited by HFSCR.
1216 * On POWER9, this could be a doorbell instruction that we need
1218 * Otherwise, we just generate a program interrupt to the guest.
1220 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1222 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1223 cpu_has_feature(CPU_FTR_ARCH_300))
1224 r = kvmppc_emulate_doorbell_instr(vcpu);
1225 if (r == EMULATE_FAIL) {
1226 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1231 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1232 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1234 * This occurs for various TM-related instructions that
1235 * we need to emulate on POWER9 DD2.2. We have already
1236 * handled the cases where the guest was in real-suspend
1237 * mode and was transitioning to transactional state.
1239 r = kvmhv_p9_tm_emulation(vcpu);
1243 case BOOK3S_INTERRUPT_HV_RM_HARD:
1244 r = RESUME_PASSTHROUGH;
1247 kvmppc_dump_regs(vcpu);
1248 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1249 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1250 vcpu->arch.shregs.msr);
1251 run->hw.hardware_exit_reason = vcpu->arch.trap;
1259 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1260 struct kvm_sregs *sregs)
1264 memset(sregs, 0, sizeof(struct kvm_sregs));
1265 sregs->pvr = vcpu->arch.pvr;
1266 for (i = 0; i < vcpu->arch.slb_max; i++) {
1267 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1268 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1274 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1275 struct kvm_sregs *sregs)
1279 /* Only accept the same PVR as the host's, since we can't spoof it */
1280 if (sregs->pvr != vcpu->arch.pvr)
1284 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1285 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1286 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1287 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1291 vcpu->arch.slb_max = j;
1296 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1297 bool preserve_top32)
1299 struct kvm *kvm = vcpu->kvm;
1300 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1303 mutex_lock(&kvm->lock);
1304 spin_lock(&vc->lock);
1306 * If ILE (interrupt little-endian) has changed, update the
1307 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1309 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1310 struct kvm_vcpu *vcpu;
1313 kvm_for_each_vcpu(i, vcpu, kvm) {
1314 if (vcpu->arch.vcore != vc)
1316 if (new_lpcr & LPCR_ILE)
1317 vcpu->arch.intr_msr |= MSR_LE;
1319 vcpu->arch.intr_msr &= ~MSR_LE;
1324 * Userspace can only modify DPFD (default prefetch depth),
1325 * ILE (interrupt little-endian) and TC (translation control).
1326 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1328 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1329 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1332 * On POWER9, allow userspace to enable large decrementer for the
1333 * guest, whether or not the host has it enabled.
1335 if (cpu_has_feature(CPU_FTR_ARCH_300))
1338 /* Broken 32-bit version of LPCR must not clear top bits */
1341 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1342 spin_unlock(&vc->lock);
1343 mutex_unlock(&kvm->lock);
1346 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1347 union kvmppc_one_reg *val)
1353 case KVM_REG_PPC_DEBUG_INST:
1354 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1356 case KVM_REG_PPC_HIOR:
1357 *val = get_reg_val(id, 0);
1359 case KVM_REG_PPC_DABR:
1360 *val = get_reg_val(id, vcpu->arch.dabr);
1362 case KVM_REG_PPC_DABRX:
1363 *val = get_reg_val(id, vcpu->arch.dabrx);
1365 case KVM_REG_PPC_DSCR:
1366 *val = get_reg_val(id, vcpu->arch.dscr);
1368 case KVM_REG_PPC_PURR:
1369 *val = get_reg_val(id, vcpu->arch.purr);
1371 case KVM_REG_PPC_SPURR:
1372 *val = get_reg_val(id, vcpu->arch.spurr);
1374 case KVM_REG_PPC_AMR:
1375 *val = get_reg_val(id, vcpu->arch.amr);
1377 case KVM_REG_PPC_UAMOR:
1378 *val = get_reg_val(id, vcpu->arch.uamor);
1380 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1381 i = id - KVM_REG_PPC_MMCR0;
1382 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1384 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1385 i = id - KVM_REG_PPC_PMC1;
1386 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1388 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1389 i = id - KVM_REG_PPC_SPMC1;
1390 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1392 case KVM_REG_PPC_SIAR:
1393 *val = get_reg_val(id, vcpu->arch.siar);
1395 case KVM_REG_PPC_SDAR:
1396 *val = get_reg_val(id, vcpu->arch.sdar);
1398 case KVM_REG_PPC_SIER:
1399 *val = get_reg_val(id, vcpu->arch.sier);
1401 case KVM_REG_PPC_IAMR:
1402 *val = get_reg_val(id, vcpu->arch.iamr);
1404 case KVM_REG_PPC_PSPB:
1405 *val = get_reg_val(id, vcpu->arch.pspb);
1407 case KVM_REG_PPC_DPDES:
1408 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1410 case KVM_REG_PPC_VTB:
1411 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1413 case KVM_REG_PPC_DAWR:
1414 *val = get_reg_val(id, vcpu->arch.dawr);
1416 case KVM_REG_PPC_DAWRX:
1417 *val = get_reg_val(id, vcpu->arch.dawrx);
1419 case KVM_REG_PPC_CIABR:
1420 *val = get_reg_val(id, vcpu->arch.ciabr);
1422 case KVM_REG_PPC_CSIGR:
1423 *val = get_reg_val(id, vcpu->arch.csigr);
1425 case KVM_REG_PPC_TACR:
1426 *val = get_reg_val(id, vcpu->arch.tacr);
1428 case KVM_REG_PPC_TCSCR:
1429 *val = get_reg_val(id, vcpu->arch.tcscr);
1431 case KVM_REG_PPC_PID:
1432 *val = get_reg_val(id, vcpu->arch.pid);
1434 case KVM_REG_PPC_ACOP:
1435 *val = get_reg_val(id, vcpu->arch.acop);
1437 case KVM_REG_PPC_WORT:
1438 *val = get_reg_val(id, vcpu->arch.wort);
1440 case KVM_REG_PPC_TIDR:
1441 *val = get_reg_val(id, vcpu->arch.tid);
1443 case KVM_REG_PPC_PSSCR:
1444 *val = get_reg_val(id, vcpu->arch.psscr);
1446 case KVM_REG_PPC_VPA_ADDR:
1447 spin_lock(&vcpu->arch.vpa_update_lock);
1448 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1449 spin_unlock(&vcpu->arch.vpa_update_lock);
1451 case KVM_REG_PPC_VPA_SLB:
1452 spin_lock(&vcpu->arch.vpa_update_lock);
1453 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1454 val->vpaval.length = vcpu->arch.slb_shadow.len;
1455 spin_unlock(&vcpu->arch.vpa_update_lock);
1457 case KVM_REG_PPC_VPA_DTL:
1458 spin_lock(&vcpu->arch.vpa_update_lock);
1459 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1460 val->vpaval.length = vcpu->arch.dtl.len;
1461 spin_unlock(&vcpu->arch.vpa_update_lock);
1463 case KVM_REG_PPC_TB_OFFSET:
1464 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1466 case KVM_REG_PPC_LPCR:
1467 case KVM_REG_PPC_LPCR_64:
1468 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1470 case KVM_REG_PPC_PPR:
1471 *val = get_reg_val(id, vcpu->arch.ppr);
1473 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1474 case KVM_REG_PPC_TFHAR:
1475 *val = get_reg_val(id, vcpu->arch.tfhar);
1477 case KVM_REG_PPC_TFIAR:
1478 *val = get_reg_val(id, vcpu->arch.tfiar);
1480 case KVM_REG_PPC_TEXASR:
1481 *val = get_reg_val(id, vcpu->arch.texasr);
1483 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1484 i = id - KVM_REG_PPC_TM_GPR0;
1485 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1487 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1490 i = id - KVM_REG_PPC_TM_VSR0;
1492 for (j = 0; j < TS_FPRWIDTH; j++)
1493 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1495 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1496 val->vval = vcpu->arch.vr_tm.vr[i-32];
1502 case KVM_REG_PPC_TM_CR:
1503 *val = get_reg_val(id, vcpu->arch.cr_tm);
1505 case KVM_REG_PPC_TM_XER:
1506 *val = get_reg_val(id, vcpu->arch.xer_tm);
1508 case KVM_REG_PPC_TM_LR:
1509 *val = get_reg_val(id, vcpu->arch.lr_tm);
1511 case KVM_REG_PPC_TM_CTR:
1512 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1514 case KVM_REG_PPC_TM_FPSCR:
1515 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1517 case KVM_REG_PPC_TM_AMR:
1518 *val = get_reg_val(id, vcpu->arch.amr_tm);
1520 case KVM_REG_PPC_TM_PPR:
1521 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1523 case KVM_REG_PPC_TM_VRSAVE:
1524 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1526 case KVM_REG_PPC_TM_VSCR:
1527 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1528 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1532 case KVM_REG_PPC_TM_DSCR:
1533 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1535 case KVM_REG_PPC_TM_TAR:
1536 *val = get_reg_val(id, vcpu->arch.tar_tm);
1539 case KVM_REG_PPC_ARCH_COMPAT:
1540 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1542 case KVM_REG_PPC_DEC_EXPIRY:
1543 *val = get_reg_val(id, vcpu->arch.dec_expires +
1544 vcpu->arch.vcore->tb_offset);
1546 case KVM_REG_PPC_ONLINE:
1547 *val = get_reg_val(id, vcpu->arch.online);
1557 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1558 union kvmppc_one_reg *val)
1562 unsigned long addr, len;
1565 case KVM_REG_PPC_HIOR:
1566 /* Only allow this to be set to zero */
1567 if (set_reg_val(id, *val))
1570 case KVM_REG_PPC_DABR:
1571 vcpu->arch.dabr = set_reg_val(id, *val);
1573 case KVM_REG_PPC_DABRX:
1574 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1576 case KVM_REG_PPC_DSCR:
1577 vcpu->arch.dscr = set_reg_val(id, *val);
1579 case KVM_REG_PPC_PURR:
1580 vcpu->arch.purr = set_reg_val(id, *val);
1582 case KVM_REG_PPC_SPURR:
1583 vcpu->arch.spurr = set_reg_val(id, *val);
1585 case KVM_REG_PPC_AMR:
1586 vcpu->arch.amr = set_reg_val(id, *val);
1588 case KVM_REG_PPC_UAMOR:
1589 vcpu->arch.uamor = set_reg_val(id, *val);
1591 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1592 i = id - KVM_REG_PPC_MMCR0;
1593 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1595 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1596 i = id - KVM_REG_PPC_PMC1;
1597 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1599 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1600 i = id - KVM_REG_PPC_SPMC1;
1601 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1603 case KVM_REG_PPC_SIAR:
1604 vcpu->arch.siar = set_reg_val(id, *val);
1606 case KVM_REG_PPC_SDAR:
1607 vcpu->arch.sdar = set_reg_val(id, *val);
1609 case KVM_REG_PPC_SIER:
1610 vcpu->arch.sier = set_reg_val(id, *val);
1612 case KVM_REG_PPC_IAMR:
1613 vcpu->arch.iamr = set_reg_val(id, *val);
1615 case KVM_REG_PPC_PSPB:
1616 vcpu->arch.pspb = set_reg_val(id, *val);
1618 case KVM_REG_PPC_DPDES:
1619 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1621 case KVM_REG_PPC_VTB:
1622 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1624 case KVM_REG_PPC_DAWR:
1625 vcpu->arch.dawr = set_reg_val(id, *val);
1627 case KVM_REG_PPC_DAWRX:
1628 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1630 case KVM_REG_PPC_CIABR:
1631 vcpu->arch.ciabr = set_reg_val(id, *val);
1632 /* Don't allow setting breakpoints in hypervisor code */
1633 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1634 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1636 case KVM_REG_PPC_CSIGR:
1637 vcpu->arch.csigr = set_reg_val(id, *val);
1639 case KVM_REG_PPC_TACR:
1640 vcpu->arch.tacr = set_reg_val(id, *val);
1642 case KVM_REG_PPC_TCSCR:
1643 vcpu->arch.tcscr = set_reg_val(id, *val);
1645 case KVM_REG_PPC_PID:
1646 vcpu->arch.pid = set_reg_val(id, *val);
1648 case KVM_REG_PPC_ACOP:
1649 vcpu->arch.acop = set_reg_val(id, *val);
1651 case KVM_REG_PPC_WORT:
1652 vcpu->arch.wort = set_reg_val(id, *val);
1654 case KVM_REG_PPC_TIDR:
1655 vcpu->arch.tid = set_reg_val(id, *val);
1657 case KVM_REG_PPC_PSSCR:
1658 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1660 case KVM_REG_PPC_VPA_ADDR:
1661 addr = set_reg_val(id, *val);
1663 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1664 vcpu->arch.dtl.next_gpa))
1666 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1668 case KVM_REG_PPC_VPA_SLB:
1669 addr = val->vpaval.addr;
1670 len = val->vpaval.length;
1672 if (addr && !vcpu->arch.vpa.next_gpa)
1674 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1676 case KVM_REG_PPC_VPA_DTL:
1677 addr = val->vpaval.addr;
1678 len = val->vpaval.length;
1680 if (addr && (len < sizeof(struct dtl_entry) ||
1681 !vcpu->arch.vpa.next_gpa))
1683 len -= len % sizeof(struct dtl_entry);
1684 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1686 case KVM_REG_PPC_TB_OFFSET:
1687 /* round up to multiple of 2^24 */
1688 vcpu->arch.vcore->tb_offset =
1689 ALIGN(set_reg_val(id, *val), 1UL << 24);
1691 case KVM_REG_PPC_LPCR:
1692 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1694 case KVM_REG_PPC_LPCR_64:
1695 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1697 case KVM_REG_PPC_PPR:
1698 vcpu->arch.ppr = set_reg_val(id, *val);
1700 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1701 case KVM_REG_PPC_TFHAR:
1702 vcpu->arch.tfhar = set_reg_val(id, *val);
1704 case KVM_REG_PPC_TFIAR:
1705 vcpu->arch.tfiar = set_reg_val(id, *val);
1707 case KVM_REG_PPC_TEXASR:
1708 vcpu->arch.texasr = set_reg_val(id, *val);
1710 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1711 i = id - KVM_REG_PPC_TM_GPR0;
1712 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1714 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1717 i = id - KVM_REG_PPC_TM_VSR0;
1719 for (j = 0; j < TS_FPRWIDTH; j++)
1720 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1722 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1723 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1728 case KVM_REG_PPC_TM_CR:
1729 vcpu->arch.cr_tm = set_reg_val(id, *val);
1731 case KVM_REG_PPC_TM_XER:
1732 vcpu->arch.xer_tm = set_reg_val(id, *val);
1734 case KVM_REG_PPC_TM_LR:
1735 vcpu->arch.lr_tm = set_reg_val(id, *val);
1737 case KVM_REG_PPC_TM_CTR:
1738 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1740 case KVM_REG_PPC_TM_FPSCR:
1741 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1743 case KVM_REG_PPC_TM_AMR:
1744 vcpu->arch.amr_tm = set_reg_val(id, *val);
1746 case KVM_REG_PPC_TM_PPR:
1747 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1749 case KVM_REG_PPC_TM_VRSAVE:
1750 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1752 case KVM_REG_PPC_TM_VSCR:
1753 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1754 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1758 case KVM_REG_PPC_TM_DSCR:
1759 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1761 case KVM_REG_PPC_TM_TAR:
1762 vcpu->arch.tar_tm = set_reg_val(id, *val);
1765 case KVM_REG_PPC_ARCH_COMPAT:
1766 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1768 case KVM_REG_PPC_DEC_EXPIRY:
1769 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1770 vcpu->arch.vcore->tb_offset;
1772 case KVM_REG_PPC_ONLINE:
1773 i = set_reg_val(id, *val);
1774 if (i && !vcpu->arch.online)
1775 atomic_inc(&vcpu->arch.vcore->online_count);
1776 else if (!i && vcpu->arch.online)
1777 atomic_dec(&vcpu->arch.vcore->online_count);
1778 vcpu->arch.online = i;
1789 * On POWER9, threads are independent and can be in different partitions.
1790 * Therefore we consider each thread to be a subcore.
1791 * There is a restriction that all threads have to be in the same
1792 * MMU mode (radix or HPT), unfortunately, but since we only support
1793 * HPT guests on a HPT host so far, that isn't an impediment yet.
1795 static int threads_per_vcore(struct kvm *kvm)
1797 if (kvm->arch.threads_indep)
1799 return threads_per_subcore;
1802 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
1804 struct kvmppc_vcore *vcore;
1806 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1811 spin_lock_init(&vcore->lock);
1812 spin_lock_init(&vcore->stoltb_lock);
1813 init_swait_queue_head(&vcore->wq);
1814 vcore->preempt_tb = TB_NIL;
1815 vcore->lpcr = kvm->arch.lpcr;
1816 vcore->first_vcpuid = id;
1818 INIT_LIST_HEAD(&vcore->preempt_list);
1823 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1824 static struct debugfs_timings_element {
1828 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1829 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1830 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1831 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1832 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1835 #define N_TIMINGS (ARRAY_SIZE(timings))
1837 struct debugfs_timings_state {
1838 struct kvm_vcpu *vcpu;
1839 unsigned int buflen;
1840 char buf[N_TIMINGS * 100];
1843 static int debugfs_timings_open(struct inode *inode, struct file *file)
1845 struct kvm_vcpu *vcpu = inode->i_private;
1846 struct debugfs_timings_state *p;
1848 p = kzalloc(sizeof(*p), GFP_KERNEL);
1852 kvm_get_kvm(vcpu->kvm);
1854 file->private_data = p;
1856 return nonseekable_open(inode, file);
1859 static int debugfs_timings_release(struct inode *inode, struct file *file)
1861 struct debugfs_timings_state *p = file->private_data;
1863 kvm_put_kvm(p->vcpu->kvm);
1868 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1869 size_t len, loff_t *ppos)
1871 struct debugfs_timings_state *p = file->private_data;
1872 struct kvm_vcpu *vcpu = p->vcpu;
1874 struct kvmhv_tb_accumulator tb;
1883 buf_end = s + sizeof(p->buf);
1884 for (i = 0; i < N_TIMINGS; ++i) {
1885 struct kvmhv_tb_accumulator *acc;
1887 acc = (struct kvmhv_tb_accumulator *)
1888 ((unsigned long)vcpu + timings[i].offset);
1890 for (loops = 0; loops < 1000; ++loops) {
1891 count = acc->seqcount;
1896 if (count == acc->seqcount) {
1904 snprintf(s, buf_end - s, "%s: stuck\n",
1907 snprintf(s, buf_end - s,
1908 "%s: %llu %llu %llu %llu\n",
1909 timings[i].name, count / 2,
1910 tb_to_ns(tb.tb_total),
1911 tb_to_ns(tb.tb_min),
1912 tb_to_ns(tb.tb_max));
1915 p->buflen = s - p->buf;
1919 if (pos >= p->buflen)
1921 if (len > p->buflen - pos)
1922 len = p->buflen - pos;
1923 n = copy_to_user(buf, p->buf + pos, len);
1933 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1934 size_t len, loff_t *ppos)
1939 static const struct file_operations debugfs_timings_ops = {
1940 .owner = THIS_MODULE,
1941 .open = debugfs_timings_open,
1942 .release = debugfs_timings_release,
1943 .read = debugfs_timings_read,
1944 .write = debugfs_timings_write,
1945 .llseek = generic_file_llseek,
1948 /* Create a debugfs directory for the vcpu */
1949 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1952 struct kvm *kvm = vcpu->kvm;
1954 snprintf(buf, sizeof(buf), "vcpu%u", id);
1955 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1957 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1958 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1960 vcpu->arch.debugfs_timings =
1961 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1962 vcpu, &debugfs_timings_ops);
1965 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1966 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1969 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1971 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1974 struct kvm_vcpu *vcpu;
1977 struct kvmppc_vcore *vcore;
1980 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1984 err = kvm_vcpu_init(vcpu, kvm, id);
1988 vcpu->arch.shared = &vcpu->arch.shregs;
1989 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1991 * The shared struct is never shared on HV,
1992 * so we can always use host endianness
1994 #ifdef __BIG_ENDIAN__
1995 vcpu->arch.shared_big_endian = true;
1997 vcpu->arch.shared_big_endian = false;
2000 vcpu->arch.mmcr[0] = MMCR0_FC;
2001 vcpu->arch.ctrl = CTRL_RUNLATCH;
2002 /* default to host PVR, since we can't spoof it */
2003 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2004 spin_lock_init(&vcpu->arch.vpa_update_lock);
2005 spin_lock_init(&vcpu->arch.tbacct_lock);
2006 vcpu->arch.busy_preempt = TB_NIL;
2007 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2010 * Set the default HFSCR for the guest from the host value.
2011 * This value is only used on POWER9.
2012 * On POWER9, we want to virtualize the doorbell facility, so we
2013 * turn off the HFSCR bit, which causes those instructions to trap.
2015 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
2016 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2017 vcpu->arch.hfscr |= HFSCR_TM;
2018 else if (!cpu_has_feature(CPU_FTR_TM_COMP))
2019 vcpu->arch.hfscr &= ~HFSCR_TM;
2020 if (cpu_has_feature(CPU_FTR_ARCH_300))
2021 vcpu->arch.hfscr &= ~HFSCR_MSGP;
2023 kvmppc_mmu_book3s_hv_init(vcpu);
2025 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2027 init_waitqueue_head(&vcpu->arch.cpu_run);
2029 mutex_lock(&kvm->lock);
2032 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2033 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2034 pr_devel("KVM: VCPU ID too high\n");
2035 core = KVM_MAX_VCORES;
2037 BUG_ON(kvm->arch.smt_mode != 1);
2038 core = kvmppc_pack_vcpu_id(kvm, id);
2041 core = id / kvm->arch.smt_mode;
2043 if (core < KVM_MAX_VCORES) {
2044 vcore = kvm->arch.vcores[core];
2045 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2046 pr_devel("KVM: collision on id %u", id);
2048 } else if (!vcore) {
2050 vcore = kvmppc_vcore_create(kvm,
2051 id & ~(kvm->arch.smt_mode - 1));
2052 kvm->arch.vcores[core] = vcore;
2053 kvm->arch.online_vcores++;
2056 mutex_unlock(&kvm->lock);
2061 spin_lock(&vcore->lock);
2062 ++vcore->num_threads;
2063 spin_unlock(&vcore->lock);
2064 vcpu->arch.vcore = vcore;
2065 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2066 vcpu->arch.thread_cpu = -1;
2067 vcpu->arch.prev_cpu = -1;
2069 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2070 kvmppc_sanity_check(vcpu);
2072 debugfs_vcpu_init(vcpu, id);
2077 kmem_cache_free(kvm_vcpu_cache, vcpu);
2079 return ERR_PTR(err);
2082 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2083 unsigned long flags)
2090 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2092 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2094 * On POWER8 (or POWER7), the threading mode is "strict",
2095 * so we pack smt_mode vcpus per vcore.
2097 if (smt_mode > threads_per_subcore)
2101 * On POWER9, the threading mode is "loose",
2102 * so each vcpu gets its own vcore.
2107 mutex_lock(&kvm->lock);
2109 if (!kvm->arch.online_vcores) {
2110 kvm->arch.smt_mode = smt_mode;
2111 kvm->arch.emul_smt_mode = esmt;
2114 mutex_unlock(&kvm->lock);
2119 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2121 if (vpa->pinned_addr)
2122 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2126 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2128 spin_lock(&vcpu->arch.vpa_update_lock);
2129 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2130 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2131 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2132 spin_unlock(&vcpu->arch.vpa_update_lock);
2133 kvm_vcpu_uninit(vcpu);
2134 kmem_cache_free(kvm_vcpu_cache, vcpu);
2137 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2139 /* Indicate we want to get back into the guest */
2143 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2145 unsigned long dec_nsec, now;
2148 if (now > vcpu->arch.dec_expires) {
2149 /* decrementer has already gone negative */
2150 kvmppc_core_queue_dec(vcpu);
2151 kvmppc_core_prepare_to_enter(vcpu);
2154 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2156 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2157 vcpu->arch.timer_running = 1;
2160 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2162 vcpu->arch.ceded = 0;
2163 if (vcpu->arch.timer_running) {
2164 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2165 vcpu->arch.timer_running = 0;
2169 extern int __kvmppc_vcore_entry(void);
2171 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2172 struct kvm_vcpu *vcpu)
2176 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2178 spin_lock_irq(&vcpu->arch.tbacct_lock);
2180 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2181 vcpu->arch.stolen_logged;
2182 vcpu->arch.busy_preempt = now;
2183 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2184 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2186 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2189 static int kvmppc_grab_hwthread(int cpu)
2191 struct paca_struct *tpaca;
2192 long timeout = 10000;
2194 tpaca = paca_ptrs[cpu];
2196 /* Ensure the thread won't go into the kernel if it wakes */
2197 tpaca->kvm_hstate.kvm_vcpu = NULL;
2198 tpaca->kvm_hstate.kvm_vcore = NULL;
2199 tpaca->kvm_hstate.napping = 0;
2201 tpaca->kvm_hstate.hwthread_req = 1;
2204 * If the thread is already executing in the kernel (e.g. handling
2205 * a stray interrupt), wait for it to get back to nap mode.
2206 * The smp_mb() is to ensure that our setting of hwthread_req
2207 * is visible before we look at hwthread_state, so if this
2208 * races with the code at system_reset_pSeries and the thread
2209 * misses our setting of hwthread_req, we are sure to see its
2210 * setting of hwthread_state, and vice versa.
2213 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2214 if (--timeout <= 0) {
2215 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2223 static void kvmppc_release_hwthread(int cpu)
2225 struct paca_struct *tpaca;
2227 tpaca = paca_ptrs[cpu];
2228 tpaca->kvm_hstate.hwthread_req = 0;
2229 tpaca->kvm_hstate.kvm_vcpu = NULL;
2230 tpaca->kvm_hstate.kvm_vcore = NULL;
2231 tpaca->kvm_hstate.kvm_split_mode = NULL;
2234 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2238 cpu = cpu_first_thread_sibling(cpu);
2239 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2241 * Make sure setting of bit in need_tlb_flush precedes
2242 * testing of cpu_in_guest bits. The matching barrier on
2243 * the other side is the first smp_mb() in kvmppc_run_core().
2246 for (i = 0; i < threads_per_core; ++i)
2247 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2248 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2251 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2253 struct kvm *kvm = vcpu->kvm;
2256 * With radix, the guest can do TLB invalidations itself,
2257 * and it could choose to use the local form (tlbiel) if
2258 * it is invalidating a translation that has only ever been
2259 * used on one vcpu. However, that doesn't mean it has
2260 * only ever been used on one physical cpu, since vcpus
2261 * can move around between pcpus. To cope with this, when
2262 * a vcpu moves from one pcpu to another, we need to tell
2263 * any vcpus running on the same core as this vcpu previously
2264 * ran to flush the TLB. The TLB is shared between threads,
2265 * so we use a single bit in .need_tlb_flush for all 4 threads.
2267 if (vcpu->arch.prev_cpu != pcpu) {
2268 if (vcpu->arch.prev_cpu >= 0 &&
2269 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2270 cpu_first_thread_sibling(pcpu))
2271 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2272 vcpu->arch.prev_cpu = pcpu;
2276 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2279 struct paca_struct *tpaca;
2280 struct kvm *kvm = vc->kvm;
2284 if (vcpu->arch.timer_running) {
2285 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2286 vcpu->arch.timer_running = 0;
2288 cpu += vcpu->arch.ptid;
2289 vcpu->cpu = vc->pcpu;
2290 vcpu->arch.thread_cpu = cpu;
2291 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2293 tpaca = paca_ptrs[cpu];
2294 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2295 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2296 tpaca->kvm_hstate.fake_suspend = 0;
2297 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2299 tpaca->kvm_hstate.kvm_vcore = vc;
2300 if (cpu != smp_processor_id())
2301 kvmppc_ipi_thread(cpu);
2304 static void kvmppc_wait_for_nap(int n_threads)
2306 int cpu = smp_processor_id();
2311 for (loops = 0; loops < 1000000; ++loops) {
2313 * Check if all threads are finished.
2314 * We set the vcore pointer when starting a thread
2315 * and the thread clears it when finished, so we look
2316 * for any threads that still have a non-NULL vcore ptr.
2318 for (i = 1; i < n_threads; ++i)
2319 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2321 if (i == n_threads) {
2328 for (i = 1; i < n_threads; ++i)
2329 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2330 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2334 * Check that we are on thread 0 and that any other threads in
2335 * this core are off-line. Then grab the threads so they can't
2338 static int on_primary_thread(void)
2340 int cpu = smp_processor_id();
2343 /* Are we on a primary subcore? */
2344 if (cpu_thread_in_subcore(cpu))
2348 while (++thr < threads_per_subcore)
2349 if (cpu_online(cpu + thr))
2352 /* Grab all hw threads so they can't go into the kernel */
2353 for (thr = 1; thr < threads_per_subcore; ++thr) {
2354 if (kvmppc_grab_hwthread(cpu + thr)) {
2355 /* Couldn't grab one; let the others go */
2357 kvmppc_release_hwthread(cpu + thr);
2358 } while (--thr > 0);
2366 * A list of virtual cores for each physical CPU.
2367 * These are vcores that could run but their runner VCPU tasks are
2368 * (or may be) preempted.
2370 struct preempted_vcore_list {
2371 struct list_head list;
2375 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2377 static void init_vcore_lists(void)
2381 for_each_possible_cpu(cpu) {
2382 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2383 spin_lock_init(&lp->lock);
2384 INIT_LIST_HEAD(&lp->list);
2388 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2390 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2392 vc->vcore_state = VCORE_PREEMPT;
2393 vc->pcpu = smp_processor_id();
2394 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2395 spin_lock(&lp->lock);
2396 list_add_tail(&vc->preempt_list, &lp->list);
2397 spin_unlock(&lp->lock);
2400 /* Start accumulating stolen time */
2401 kvmppc_core_start_stolen(vc);
2404 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2406 struct preempted_vcore_list *lp;
2408 kvmppc_core_end_stolen(vc);
2409 if (!list_empty(&vc->preempt_list)) {
2410 lp = &per_cpu(preempted_vcores, vc->pcpu);
2411 spin_lock(&lp->lock);
2412 list_del_init(&vc->preempt_list);
2413 spin_unlock(&lp->lock);
2415 vc->vcore_state = VCORE_INACTIVE;
2419 * This stores information about the virtual cores currently
2420 * assigned to a physical core.
2424 int max_subcore_threads;
2426 int subcore_threads[MAX_SUBCORES];
2427 struct kvmppc_vcore *vc[MAX_SUBCORES];
2431 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2432 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2434 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2436 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2438 memset(cip, 0, sizeof(*cip));
2439 cip->n_subcores = 1;
2440 cip->max_subcore_threads = vc->num_threads;
2441 cip->total_threads = vc->num_threads;
2442 cip->subcore_threads[0] = vc->num_threads;
2446 static bool subcore_config_ok(int n_subcores, int n_threads)
2449 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2450 * split-core mode, with one thread per subcore.
2452 if (cpu_has_feature(CPU_FTR_ARCH_300))
2453 return n_subcores <= 4 && n_threads == 1;
2455 /* On POWER8, can only dynamically split if unsplit to begin with */
2456 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2458 if (n_subcores > MAX_SUBCORES)
2460 if (n_subcores > 1) {
2461 if (!(dynamic_mt_modes & 2))
2463 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2467 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2470 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2472 vc->entry_exit_map = 0;
2474 vc->napping_threads = 0;
2475 vc->conferring_threads = 0;
2476 vc->tb_offset_applied = 0;
2479 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2481 int n_threads = vc->num_threads;
2484 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2487 /* Some POWER9 chips require all threads to be in the same MMU mode */
2488 if (no_mixing_hpt_and_radix &&
2489 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2492 if (n_threads < cip->max_subcore_threads)
2493 n_threads = cip->max_subcore_threads;
2494 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2496 cip->max_subcore_threads = n_threads;
2498 sub = cip->n_subcores;
2500 cip->total_threads += vc->num_threads;
2501 cip->subcore_threads[sub] = vc->num_threads;
2503 init_vcore_to_run(vc);
2504 list_del_init(&vc->preempt_list);
2510 * Work out whether it is possible to piggyback the execution of
2511 * vcore *pvc onto the execution of the other vcores described in *cip.
2513 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2516 if (cip->total_threads + pvc->num_threads > target_threads)
2519 return can_dynamic_split(pvc, cip);
2522 static void prepare_threads(struct kvmppc_vcore *vc)
2525 struct kvm_vcpu *vcpu;
2527 for_each_runnable_thread(i, vcpu, vc) {
2528 if (signal_pending(vcpu->arch.run_task))
2529 vcpu->arch.ret = -EINTR;
2530 else if (vcpu->arch.vpa.update_pending ||
2531 vcpu->arch.slb_shadow.update_pending ||
2532 vcpu->arch.dtl.update_pending)
2533 vcpu->arch.ret = RESUME_GUEST;
2536 kvmppc_remove_runnable(vc, vcpu);
2537 wake_up(&vcpu->arch.cpu_run);
2541 static void collect_piggybacks(struct core_info *cip, int target_threads)
2543 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2544 struct kvmppc_vcore *pvc, *vcnext;
2546 spin_lock(&lp->lock);
2547 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2548 if (!spin_trylock(&pvc->lock))
2550 prepare_threads(pvc);
2551 if (!pvc->n_runnable) {
2552 list_del_init(&pvc->preempt_list);
2553 if (pvc->runner == NULL) {
2554 pvc->vcore_state = VCORE_INACTIVE;
2555 kvmppc_core_end_stolen(pvc);
2557 spin_unlock(&pvc->lock);
2560 if (!can_piggyback(pvc, cip, target_threads)) {
2561 spin_unlock(&pvc->lock);
2564 kvmppc_core_end_stolen(pvc);
2565 pvc->vcore_state = VCORE_PIGGYBACK;
2566 if (cip->total_threads >= target_threads)
2569 spin_unlock(&lp->lock);
2572 static bool recheck_signals(struct core_info *cip)
2575 struct kvm_vcpu *vcpu;
2577 for (sub = 0; sub < cip->n_subcores; ++sub)
2578 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2579 if (signal_pending(vcpu->arch.run_task))
2584 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2586 int still_running = 0, i;
2589 struct kvm_vcpu *vcpu;
2591 spin_lock(&vc->lock);
2593 for_each_runnable_thread(i, vcpu, vc) {
2595 * It's safe to unlock the vcore in the loop here, because
2596 * for_each_runnable_thread() is safe against removal of
2597 * the vcpu, and the vcore state is VCORE_EXITING here,
2598 * so any vcpus becoming runnable will have their arch.trap
2599 * set to zero and can't actually run in the guest.
2601 spin_unlock(&vc->lock);
2602 /* cancel pending dec exception if dec is positive */
2603 if (now < vcpu->arch.dec_expires &&
2604 kvmppc_core_pending_dec(vcpu))
2605 kvmppc_core_dequeue_dec(vcpu);
2607 trace_kvm_guest_exit(vcpu);
2610 if (vcpu->arch.trap)
2611 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2612 vcpu->arch.run_task);
2614 vcpu->arch.ret = ret;
2615 vcpu->arch.trap = 0;
2617 spin_lock(&vc->lock);
2618 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2619 if (vcpu->arch.pending_exceptions)
2620 kvmppc_core_prepare_to_enter(vcpu);
2621 if (vcpu->arch.ceded)
2622 kvmppc_set_timer(vcpu);
2626 kvmppc_remove_runnable(vc, vcpu);
2627 wake_up(&vcpu->arch.cpu_run);
2631 if (still_running > 0) {
2632 kvmppc_vcore_preempt(vc);
2633 } else if (vc->runner) {
2634 vc->vcore_state = VCORE_PREEMPT;
2635 kvmppc_core_start_stolen(vc);
2637 vc->vcore_state = VCORE_INACTIVE;
2639 if (vc->n_runnable > 0 && vc->runner == NULL) {
2640 /* make sure there's a candidate runner awake */
2642 vcpu = next_runnable_thread(vc, &i);
2643 wake_up(&vcpu->arch.cpu_run);
2646 spin_unlock(&vc->lock);
2650 * Clear core from the list of active host cores as we are about to
2651 * enter the guest. Only do this if it is the primary thread of the
2652 * core (not if a subcore) that is entering the guest.
2654 static inline int kvmppc_clear_host_core(unsigned int cpu)
2658 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2661 * Memory barrier can be omitted here as we will do a smp_wmb()
2662 * later in kvmppc_start_thread and we need ensure that state is
2663 * visible to other CPUs only after we enter guest.
2665 core = cpu >> threads_shift;
2666 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2671 * Advertise this core as an active host core since we exited the guest
2672 * Only need to do this if it is the primary thread of the core that is
2675 static inline int kvmppc_set_host_core(unsigned int cpu)
2679 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2683 * Memory barrier can be omitted here because we do a spin_unlock
2684 * immediately after this which provides the memory barrier.
2686 core = cpu >> threads_shift;
2687 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2691 static void set_irq_happened(int trap)
2694 case BOOK3S_INTERRUPT_EXTERNAL:
2695 local_paca->irq_happened |= PACA_IRQ_EE;
2697 case BOOK3S_INTERRUPT_H_DOORBELL:
2698 local_paca->irq_happened |= PACA_IRQ_DBELL;
2700 case BOOK3S_INTERRUPT_HMI:
2701 local_paca->irq_happened |= PACA_IRQ_HMI;
2703 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2704 replay_system_reset();
2710 * Run a set of guest threads on a physical core.
2711 * Called with vc->lock held.
2713 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2715 struct kvm_vcpu *vcpu;
2718 struct core_info core_info;
2719 struct kvmppc_vcore *pvc;
2720 struct kvm_split_mode split_info, *sip;
2721 int split, subcore_size, active;
2724 unsigned long cmd_bit, stat_bit;
2727 int controlled_threads;
2733 * Remove from the list any threads that have a signal pending
2734 * or need a VPA update done
2736 prepare_threads(vc);
2738 /* if the runner is no longer runnable, let the caller pick a new one */
2739 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2745 init_vcore_to_run(vc);
2746 vc->preempt_tb = TB_NIL;
2749 * Number of threads that we will be controlling: the same as
2750 * the number of threads per subcore, except on POWER9,
2751 * where it's 1 because the threads are (mostly) independent.
2753 controlled_threads = threads_per_vcore(vc->kvm);
2756 * Make sure we are running on primary threads, and that secondary
2757 * threads are offline. Also check if the number of threads in this
2758 * guest are greater than the current system threads per guest.
2759 * On POWER9, we need to be not in independent-threads mode if
2760 * this is a HPT guest on a radix host machine where the
2761 * CPU threads may not be in different MMU modes.
2763 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
2764 !kvm_is_radix(vc->kvm);
2765 if (((controlled_threads > 1) &&
2766 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2767 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2768 for_each_runnable_thread(i, vcpu, vc) {
2769 vcpu->arch.ret = -EBUSY;
2770 kvmppc_remove_runnable(vc, vcpu);
2771 wake_up(&vcpu->arch.cpu_run);
2777 * See if we could run any other vcores on the physical core
2778 * along with this one.
2780 init_core_info(&core_info, vc);
2781 pcpu = smp_processor_id();
2782 target_threads = controlled_threads;
2783 if (target_smt_mode && target_smt_mode < target_threads)
2784 target_threads = target_smt_mode;
2785 if (vc->num_threads < target_threads)
2786 collect_piggybacks(&core_info, target_threads);
2789 * On radix, arrange for TLB flushing if necessary.
2790 * This has to be done before disabling interrupts since
2791 * it uses smp_call_function().
2793 pcpu = smp_processor_id();
2794 if (kvm_is_radix(vc->kvm)) {
2795 for (sub = 0; sub < core_info.n_subcores; ++sub)
2796 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2797 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2801 * Hard-disable interrupts, and check resched flag and signals.
2802 * If we need to reschedule or deliver a signal, clean up
2803 * and return without going into the guest(s).
2804 * If the mmu_ready flag has been cleared, don't go into the
2805 * guest because that means a HPT resize operation is in progress.
2807 local_irq_disable();
2809 if (lazy_irq_pending() || need_resched() ||
2810 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
2812 vc->vcore_state = VCORE_INACTIVE;
2813 /* Unlock all except the primary vcore */
2814 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2815 pvc = core_info.vc[sub];
2816 /* Put back on to the preempted vcores list */
2817 kvmppc_vcore_preempt(pvc);
2818 spin_unlock(&pvc->lock);
2820 for (i = 0; i < controlled_threads; ++i)
2821 kvmppc_release_hwthread(pcpu + i);
2825 kvmppc_clear_host_core(pcpu);
2827 /* Decide on micro-threading (split-core) mode */
2828 subcore_size = threads_per_subcore;
2829 cmd_bit = stat_bit = 0;
2830 split = core_info.n_subcores;
2832 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2833 && !cpu_has_feature(CPU_FTR_ARCH_300);
2835 if (split > 1 || hpt_on_radix) {
2837 memset(&split_info, 0, sizeof(split_info));
2838 for (sub = 0; sub < core_info.n_subcores; ++sub)
2839 split_info.vc[sub] = core_info.vc[sub];
2842 if (split == 2 && (dynamic_mt_modes & 2)) {
2843 cmd_bit = HID0_POWER8_1TO2LPAR;
2844 stat_bit = HID0_POWER8_2LPARMODE;
2847 cmd_bit = HID0_POWER8_1TO4LPAR;
2848 stat_bit = HID0_POWER8_4LPARMODE;
2850 subcore_size = MAX_SMT_THREADS / split;
2851 split_info.rpr = mfspr(SPRN_RPR);
2852 split_info.pmmar = mfspr(SPRN_PMMAR);
2853 split_info.ldbar = mfspr(SPRN_LDBAR);
2854 split_info.subcore_size = subcore_size;
2856 split_info.subcore_size = 1;
2858 /* Use the split_info for LPCR/LPIDR changes */
2859 split_info.lpcr_req = vc->lpcr;
2860 split_info.lpidr_req = vc->kvm->arch.lpid;
2861 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2862 split_info.do_set = 1;
2866 /* order writes to split_info before kvm_split_mode pointer */
2870 for (thr = 0; thr < controlled_threads; ++thr) {
2871 struct paca_struct *paca = paca_ptrs[pcpu + thr];
2873 paca->kvm_hstate.tid = thr;
2874 paca->kvm_hstate.napping = 0;
2875 paca->kvm_hstate.kvm_split_mode = sip;
2878 /* Initiate micro-threading (split-core) on POWER8 if required */
2880 unsigned long hid0 = mfspr(SPRN_HID0);
2882 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2884 mtspr(SPRN_HID0, hid0);
2887 hid0 = mfspr(SPRN_HID0);
2888 if (hid0 & stat_bit)
2895 * On POWER8, set RWMR register.
2896 * Since it only affects PURR and SPURR, it doesn't affect
2897 * the host, so we don't save/restore the host value.
2900 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
2901 int n_online = atomic_read(&vc->online_count);
2904 * Use the 8-thread value if we're doing split-core
2905 * or if the vcore's online count looks bogus.
2907 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
2908 n_online >= 1 && n_online <= MAX_SMT_THREADS)
2909 rwmr_val = p8_rwmr_values[n_online];
2910 mtspr(SPRN_RWMR, rwmr_val);
2913 /* Start all the threads */
2915 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2916 thr = is_power8 ? subcore_thread_map[sub] : sub;
2919 pvc = core_info.vc[sub];
2920 pvc->pcpu = pcpu + thr;
2921 for_each_runnable_thread(i, vcpu, pvc) {
2922 kvmppc_start_thread(vcpu, pvc);
2923 kvmppc_create_dtl_entry(vcpu, pvc);
2924 trace_kvm_guest_enter(vcpu);
2925 if (!vcpu->arch.ptid)
2927 active |= 1 << (thr + vcpu->arch.ptid);
2930 * We need to start the first thread of each subcore
2931 * even if it doesn't have a vcpu.
2934 kvmppc_start_thread(NULL, pvc);
2938 * Ensure that split_info.do_nap is set after setting
2939 * the vcore pointer in the PACA of the secondaries.
2944 * When doing micro-threading, poke the inactive threads as well.
2945 * This gets them to the nap instruction after kvm_do_nap,
2946 * which reduces the time taken to unsplit later.
2947 * For POWER9 HPT guest on radix host, we need all the secondary
2948 * threads woken up so they can do the LPCR/LPIDR change.
2950 if (cmd_bit || hpt_on_radix) {
2951 split_info.do_nap = 1; /* ask secondaries to nap when done */
2952 for (thr = 1; thr < threads_per_subcore; ++thr)
2953 if (!(active & (1 << thr)))
2954 kvmppc_ipi_thread(pcpu + thr);
2957 vc->vcore_state = VCORE_RUNNING;
2960 trace_kvmppc_run_core(vc, 0);
2962 for (sub = 0; sub < core_info.n_subcores; ++sub)
2963 spin_unlock(&core_info.vc[sub]->lock);
2965 if (kvm_is_radix(vc->kvm)) {
2969 * Do we need to flush the process scoped TLB for the LPAR?
2971 * On POWER9, individual threads can come in here, but the
2972 * TLB is shared between the 4 threads in a core, hence
2973 * invalidating on one thread invalidates for all.
2974 * Thus we make all 4 threads use the same bit here.
2976 * Hash must be flushed in realmode in order to use tlbiel.
2978 mtspr(SPRN_LPID, vc->kvm->arch.lpid);
2981 if (cpu_has_feature(CPU_FTR_ARCH_300))
2984 if (cpumask_test_cpu(tmp, &vc->kvm->arch.need_tlb_flush)) {
2985 radix__local_flush_tlb_lpid_guest(vc->kvm->arch.lpid);
2986 /* Clear the bit after the TLB flush */
2987 cpumask_clear_cpu(tmp, &vc->kvm->arch.need_tlb_flush);
2992 * Interrupts will be enabled once we get into the guest,
2993 * so tell lockdep that we're about to enable interrupts.
2995 trace_hardirqs_on();
2997 guest_enter_irqoff();
2999 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3001 this_cpu_disable_ftrace();
3003 trap = __kvmppc_vcore_entry();
3005 this_cpu_enable_ftrace();
3007 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3009 trace_hardirqs_off();
3010 set_irq_happened(trap);
3012 spin_lock(&vc->lock);
3013 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3014 vc->vcore_state = VCORE_EXITING;
3016 /* wait for secondary threads to finish writing their state to memory */
3017 kvmppc_wait_for_nap(controlled_threads);
3019 /* Return to whole-core mode if we split the core earlier */
3021 unsigned long hid0 = mfspr(SPRN_HID0);
3022 unsigned long loops = 0;
3024 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3025 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3027 mtspr(SPRN_HID0, hid0);
3030 hid0 = mfspr(SPRN_HID0);
3031 if (!(hid0 & stat_bit))
3036 } else if (hpt_on_radix) {
3037 /* Wait for all threads to have seen final sync */
3038 for (thr = 1; thr < controlled_threads; ++thr) {
3039 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3041 while (paca->kvm_hstate.kvm_split_mode) {
3048 split_info.do_nap = 0;
3050 kvmppc_set_host_core(pcpu);
3055 /* Let secondaries go back to the offline loop */
3056 for (i = 0; i < controlled_threads; ++i) {
3057 kvmppc_release_hwthread(pcpu + i);
3058 if (sip && sip->napped[i])
3059 kvmppc_ipi_thread(pcpu + i);
3060 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3063 spin_unlock(&vc->lock);
3065 /* make sure updates to secondary vcpu structs are visible now */
3070 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3071 pvc = core_info.vc[sub];
3072 post_guest_process(pvc, pvc == vc);
3075 spin_lock(&vc->lock);
3078 vc->vcore_state = VCORE_INACTIVE;
3079 trace_kvmppc_run_core(vc, 1);
3083 * Load up hypervisor-mode registers on P9.
3085 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit)
3087 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3089 u64 tb, purr, spurr;
3091 unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3092 unsigned long host_ciabr = mfspr(SPRN_CIABR);
3093 unsigned long host_dawr = mfspr(SPRN_DAWR);
3094 unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3095 unsigned long host_psscr = mfspr(SPRN_PSSCR);
3096 unsigned long host_pidr = mfspr(SPRN_PID);
3098 hdec = time_limit - mftb();
3100 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3101 mtspr(SPRN_HDEC, hdec);
3103 if (vc->tb_offset) {
3104 u64 new_tb = mftb() + vc->tb_offset;
3105 mtspr(SPRN_TBU40, new_tb);
3107 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3108 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3109 vc->tb_offset_applied = vc->tb_offset;
3113 mtspr(SPRN_PCR, vc->pcr);
3114 mtspr(SPRN_DPDES, vc->dpdes);
3115 mtspr(SPRN_VTB, vc->vtb);
3117 local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3118 local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3119 mtspr(SPRN_PURR, vcpu->arch.purr);
3120 mtspr(SPRN_SPURR, vcpu->arch.spurr);
3122 if (cpu_has_feature(CPU_FTR_DAWR)) {
3123 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3124 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3126 mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3127 mtspr(SPRN_IC, vcpu->arch.ic);
3128 mtspr(SPRN_PID, vcpu->arch.pid);
3130 mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3131 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3133 mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3135 mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3136 mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3137 mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3138 mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3140 mtspr(SPRN_AMOR, ~0UL);
3142 mtspr(SPRN_LPCR, vc->lpcr);
3145 kvmppc_xive_push_vcpu(vcpu);
3147 mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3148 mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3150 trap = __kvmhv_vcpu_entry_p9(vcpu);
3152 /* Advance host PURR/SPURR by the amount used by guest */
3153 purr = mfspr(SPRN_PURR);
3154 spurr = mfspr(SPRN_SPURR);
3155 mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3156 purr - vcpu->arch.purr);
3157 mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3158 spurr - vcpu->arch.spurr);
3159 vcpu->arch.purr = purr;
3160 vcpu->arch.spurr = spurr;
3162 vcpu->arch.ic = mfspr(SPRN_IC);
3163 vcpu->arch.pid = mfspr(SPRN_PID);
3164 vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3166 vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3167 vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3168 vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3169 vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3171 mtspr(SPRN_PSSCR, host_psscr);
3172 mtspr(SPRN_HFSCR, host_hfscr);
3173 mtspr(SPRN_CIABR, host_ciabr);
3174 mtspr(SPRN_DAWR, host_dawr);
3175 mtspr(SPRN_DAWRX, host_dawrx);
3176 mtspr(SPRN_PID, host_pidr);
3179 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3180 * case we interrupted the guest between a tlbie and a ptesync.
3182 asm volatile("eieio; tlbsync; ptesync");
3184 mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid); /* restore host LPID */
3187 vc->dpdes = mfspr(SPRN_DPDES);
3188 vc->vtb = mfspr(SPRN_VTB);
3189 mtspr(SPRN_DPDES, 0);
3193 if (vc->tb_offset_applied) {
3194 u64 new_tb = mftb() - vc->tb_offset_applied;
3195 mtspr(SPRN_TBU40, new_tb);
3197 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3198 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3199 vc->tb_offset_applied = 0;
3202 mtspr(SPRN_HDEC, 0x7fffffff);
3203 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3209 * Virtual-mode guest entry for POWER9 and later when the host and
3210 * guest are both using the radix MMU. The LPIDR has already been set.
3212 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit)
3214 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3215 unsigned long host_dscr = mfspr(SPRN_DSCR);
3216 unsigned long host_tidr = mfspr(SPRN_TIDR);
3217 unsigned long host_iamr = mfspr(SPRN_IAMR);
3222 dec = mfspr(SPRN_DEC);
3225 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3226 local_paca->kvm_hstate.dec_expires = dec + tb;
3227 if (local_paca->kvm_hstate.dec_expires < time_limit)
3228 time_limit = local_paca->kvm_hstate.dec_expires;
3230 vcpu->arch.ceded = 0;
3232 kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */
3234 kvmppc_subcore_enter_guest();
3236 vc->entry_exit_map = 1;
3239 if (vcpu->arch.vpa.pinned_addr) {
3240 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3241 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3242 lp->yield_count = cpu_to_be32(yield_count);
3243 vcpu->arch.vpa.dirty = 1;
3246 if (cpu_has_feature(CPU_FTR_TM) ||
3247 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3248 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3250 kvmhv_load_guest_pmu(vcpu);
3252 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3253 load_fp_state(&vcpu->arch.fp);
3254 #ifdef CONFIG_ALTIVEC
3255 load_vr_state(&vcpu->arch.vr);
3258 mtspr(SPRN_DSCR, vcpu->arch.dscr);
3259 mtspr(SPRN_IAMR, vcpu->arch.iamr);
3260 mtspr(SPRN_PSPB, vcpu->arch.pspb);
3261 mtspr(SPRN_FSCR, vcpu->arch.fscr);
3262 mtspr(SPRN_TAR, vcpu->arch.tar);
3263 mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3264 mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3265 mtspr(SPRN_BESCR, vcpu->arch.bescr);
3266 mtspr(SPRN_WORT, vcpu->arch.wort);
3267 mtspr(SPRN_TIDR, vcpu->arch.tid);
3268 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3269 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3270 mtspr(SPRN_AMR, vcpu->arch.amr);
3271 mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3273 if (!(vcpu->arch.ctrl & 1))
3274 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3276 mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3278 if (vcpu->arch.doorbell_request) {
3281 vcpu->arch.doorbell_request = 0;
3284 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit);
3286 vcpu->arch.slb_max = 0;
3287 dec = mfspr(SPRN_DEC);
3289 vcpu->arch.dec_expires = dec + tb;
3291 vcpu->arch.thread_cpu = -1;
3292 vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3294 vcpu->arch.iamr = mfspr(SPRN_IAMR);
3295 vcpu->arch.pspb = mfspr(SPRN_PSPB);
3296 vcpu->arch.fscr = mfspr(SPRN_FSCR);
3297 vcpu->arch.tar = mfspr(SPRN_TAR);
3298 vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3299 vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3300 vcpu->arch.bescr = mfspr(SPRN_BESCR);
3301 vcpu->arch.wort = mfspr(SPRN_WORT);
3302 vcpu->arch.tid = mfspr(SPRN_TIDR);
3303 vcpu->arch.amr = mfspr(SPRN_AMR);
3304 vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3305 vcpu->arch.dscr = mfspr(SPRN_DSCR);
3307 mtspr(SPRN_PSPB, 0);
3308 mtspr(SPRN_WORT, 0);
3310 mtspr(SPRN_UAMOR, 0);
3311 mtspr(SPRN_DSCR, host_dscr);
3312 mtspr(SPRN_TIDR, host_tidr);
3313 mtspr(SPRN_IAMR, host_iamr);
3314 mtspr(SPRN_PSPB, 0);
3316 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3317 store_fp_state(&vcpu->arch.fp);
3318 #ifdef CONFIG_ALTIVEC
3319 store_vr_state(&vcpu->arch.vr);
3322 if (cpu_has_feature(CPU_FTR_TM) ||
3323 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3324 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3327 if (vcpu->arch.vpa.pinned_addr) {
3328 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3329 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3330 lp->yield_count = cpu_to_be32(yield_count);
3331 vcpu->arch.vpa.dirty = 1;
3332 save_pmu = lp->pmcregs_in_use;
3335 kvmhv_save_guest_pmu(vcpu, save_pmu);
3337 vc->entry_exit_map = 0x101;
3340 mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3342 kvmhv_load_host_pmu();
3344 kvmppc_subcore_exit_guest();
3350 * Wait for some other vcpu thread to execute us, and
3351 * wake us up when we need to handle something in the host.
3353 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3354 struct kvm_vcpu *vcpu, int wait_state)
3358 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3359 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3360 spin_unlock(&vc->lock);
3362 spin_lock(&vc->lock);
3364 finish_wait(&vcpu->arch.cpu_run, &wait);
3367 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3370 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3371 vc->halt_poll_ns = 10000;
3373 vc->halt_poll_ns *= halt_poll_ns_grow;
3376 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3378 if (halt_poll_ns_shrink == 0)
3379 vc->halt_poll_ns = 0;
3381 vc->halt_poll_ns /= halt_poll_ns_shrink;
3384 #ifdef CONFIG_KVM_XICS
3385 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3387 if (!xive_enabled())
3389 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3390 vcpu->arch.xive_saved_state.cppr;
3393 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3397 #endif /* CONFIG_KVM_XICS */
3399 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3401 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3402 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3409 * Check to see if any of the runnable vcpus on the vcore have pending
3410 * exceptions or are no longer ceded
3412 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3414 struct kvm_vcpu *vcpu;
3417 for_each_runnable_thread(i, vcpu, vc) {
3418 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3426 * All the vcpus in this vcore are idle, so wait for a decrementer
3427 * or external interrupt to one of the vcpus. vc->lock is held.
3429 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3431 ktime_t cur, start_poll, start_wait;
3434 DECLARE_SWAITQUEUE(wait);
3436 /* Poll for pending exceptions and ceded state */
3437 cur = start_poll = ktime_get();
3438 if (vc->halt_poll_ns) {
3439 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3440 ++vc->runner->stat.halt_attempted_poll;
3442 vc->vcore_state = VCORE_POLLING;
3443 spin_unlock(&vc->lock);
3446 if (kvmppc_vcore_check_block(vc)) {
3451 } while (single_task_running() && ktime_before(cur, stop));
3453 spin_lock(&vc->lock);
3454 vc->vcore_state = VCORE_INACTIVE;
3457 ++vc->runner->stat.halt_successful_poll;
3462 prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3464 if (kvmppc_vcore_check_block(vc)) {
3465 finish_swait(&vc->wq, &wait);
3467 /* If we polled, count this as a successful poll */
3468 if (vc->halt_poll_ns)
3469 ++vc->runner->stat.halt_successful_poll;
3473 start_wait = ktime_get();
3475 vc->vcore_state = VCORE_SLEEPING;
3476 trace_kvmppc_vcore_blocked(vc, 0);
3477 spin_unlock(&vc->lock);
3479 finish_swait(&vc->wq, &wait);
3480 spin_lock(&vc->lock);
3481 vc->vcore_state = VCORE_INACTIVE;
3482 trace_kvmppc_vcore_blocked(vc, 1);
3483 ++vc->runner->stat.halt_successful_wait;
3488 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3490 /* Attribute wait time */
3492 vc->runner->stat.halt_wait_ns +=
3493 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3494 /* Attribute failed poll time */
3495 if (vc->halt_poll_ns)
3496 vc->runner->stat.halt_poll_fail_ns +=
3497 ktime_to_ns(start_wait) -
3498 ktime_to_ns(start_poll);
3500 /* Attribute successful poll time */
3501 if (vc->halt_poll_ns)
3502 vc->runner->stat.halt_poll_success_ns +=
3504 ktime_to_ns(start_poll);
3507 /* Adjust poll time */
3509 if (block_ns <= vc->halt_poll_ns)
3511 /* We slept and blocked for longer than the max halt time */
3512 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3513 shrink_halt_poll_ns(vc);
3514 /* We slept and our poll time is too small */
3515 else if (vc->halt_poll_ns < halt_poll_ns &&
3516 block_ns < halt_poll_ns)
3517 grow_halt_poll_ns(vc);
3518 if (vc->halt_poll_ns > halt_poll_ns)
3519 vc->halt_poll_ns = halt_poll_ns;
3521 vc->halt_poll_ns = 0;
3523 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3526 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3529 struct kvm *kvm = vcpu->kvm;
3531 mutex_lock(&kvm->lock);
3532 if (!kvm->arch.mmu_ready) {
3533 if (!kvm_is_radix(kvm))
3534 r = kvmppc_hv_setup_htab_rma(vcpu);
3536 if (cpu_has_feature(CPU_FTR_ARCH_300))
3537 kvmppc_setup_partition_table(kvm);
3538 kvm->arch.mmu_ready = 1;
3541 mutex_unlock(&kvm->lock);
3545 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3548 struct kvmppc_vcore *vc;
3551 trace_kvmppc_run_vcpu_enter(vcpu);
3553 kvm_run->exit_reason = 0;
3554 vcpu->arch.ret = RESUME_GUEST;
3555 vcpu->arch.trap = 0;
3556 kvmppc_update_vpas(vcpu);
3559 * Synchronize with other threads in this virtual core
3561 vc = vcpu->arch.vcore;
3562 spin_lock(&vc->lock);
3563 vcpu->arch.ceded = 0;
3564 vcpu->arch.run_task = current;
3565 vcpu->arch.kvm_run = kvm_run;
3566 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3567 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3568 vcpu->arch.busy_preempt = TB_NIL;
3569 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3573 * This happens the first time this is called for a vcpu.
3574 * If the vcore is already running, we may be able to start
3575 * this thread straight away and have it join in.
3577 if (!signal_pending(current)) {
3578 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3579 vc->vcore_state == VCORE_RUNNING) &&
3580 !VCORE_IS_EXITING(vc)) {
3581 kvmppc_create_dtl_entry(vcpu, vc);
3582 kvmppc_start_thread(vcpu, vc);
3583 trace_kvm_guest_enter(vcpu);
3584 } else if (vc->vcore_state == VCORE_SLEEPING) {
3585 swake_up_one(&vc->wq);
3590 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3591 !signal_pending(current)) {
3592 /* See if the MMU is ready to go */
3593 if (!vcpu->kvm->arch.mmu_ready) {
3594 spin_unlock(&vc->lock);
3595 r = kvmhv_setup_mmu(vcpu);
3596 spin_lock(&vc->lock);
3598 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3599 kvm_run->fail_entry.
3600 hardware_entry_failure_reason = 0;
3606 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3607 kvmppc_vcore_end_preempt(vc);
3609 if (vc->vcore_state != VCORE_INACTIVE) {
3610 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3613 for_each_runnable_thread(i, v, vc) {
3614 kvmppc_core_prepare_to_enter(v);
3615 if (signal_pending(v->arch.run_task)) {
3616 kvmppc_remove_runnable(vc, v);
3617 v->stat.signal_exits++;
3618 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3619 v->arch.ret = -EINTR;
3620 wake_up(&v->arch.cpu_run);
3623 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3626 for_each_runnable_thread(i, v, vc) {
3627 if (!kvmppc_vcpu_woken(v))
3628 n_ceded += v->arch.ceded;
3633 if (n_ceded == vc->n_runnable) {
3634 kvmppc_vcore_blocked(vc);
3635 } else if (need_resched()) {
3636 kvmppc_vcore_preempt(vc);
3637 /* Let something else run */
3638 cond_resched_lock(&vc->lock);
3639 if (vc->vcore_state == VCORE_PREEMPT)
3640 kvmppc_vcore_end_preempt(vc);
3642 kvmppc_run_core(vc);
3647 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3648 (vc->vcore_state == VCORE_RUNNING ||
3649 vc->vcore_state == VCORE_EXITING ||
3650 vc->vcore_state == VCORE_PIGGYBACK))
3651 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3653 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3654 kvmppc_vcore_end_preempt(vc);
3656 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3657 kvmppc_remove_runnable(vc, vcpu);
3658 vcpu->stat.signal_exits++;
3659 kvm_run->exit_reason = KVM_EXIT_INTR;
3660 vcpu->arch.ret = -EINTR;
3663 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3664 /* Wake up some vcpu to run the core */
3666 v = next_runnable_thread(vc, &i);
3667 wake_up(&v->arch.cpu_run);
3670 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3671 spin_unlock(&vc->lock);
3672 return vcpu->arch.ret;
3675 static int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
3676 struct kvm_vcpu *vcpu, u64 time_limit)
3678 int trap, r, pcpu, pcpu0;
3680 struct kvmppc_vcore *vc;
3681 struct kvm *kvm = vcpu->kvm;
3683 trace_kvmppc_run_vcpu_enter(vcpu);
3685 kvm_run->exit_reason = 0;
3686 vcpu->arch.ret = RESUME_GUEST;
3687 vcpu->arch.trap = 0;
3689 vc = vcpu->arch.vcore;
3690 vcpu->arch.ceded = 0;
3691 vcpu->arch.run_task = current;
3692 vcpu->arch.kvm_run = kvm_run;
3693 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3694 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3695 vcpu->arch.busy_preempt = TB_NIL;
3696 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
3697 vc->runnable_threads[0] = vcpu;
3701 /* See if the MMU is ready to go */
3702 if (!kvm->arch.mmu_ready) {
3703 r = kvmhv_setup_mmu(vcpu);
3705 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3706 kvm_run->fail_entry.
3707 hardware_entry_failure_reason = 0;
3716 kvmppc_update_vpas(vcpu);
3718 init_vcore_to_run(vc);
3719 vc->preempt_tb = TB_NIL;
3722 pcpu = smp_processor_id();
3724 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3726 local_irq_disable();
3728 if (signal_pending(current))
3730 if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
3733 kvmppc_core_prepare_to_enter(vcpu);
3735 kvmppc_clear_host_core(pcpu);
3737 local_paca->kvm_hstate.tid = 0;
3738 local_paca->kvm_hstate.napping = 0;
3739 local_paca->kvm_hstate.kvm_split_mode = NULL;
3740 kvmppc_start_thread(vcpu, vc);
3741 kvmppc_create_dtl_entry(vcpu, vc);
3742 trace_kvm_guest_enter(vcpu);
3744 vc->vcore_state = VCORE_RUNNING;
3745 trace_kvmppc_run_core(vc, 0);
3747 mtspr(SPRN_LPID, vc->kvm->arch.lpid);
3750 /* See comment above in kvmppc_run_core() about this */
3752 if (cpu_has_feature(CPU_FTR_ARCH_300))
3755 if (cpumask_test_cpu(pcpu0, &kvm->arch.need_tlb_flush)) {
3756 radix__local_flush_tlb_lpid_guest(kvm->arch.lpid);
3757 /* Clear the bit after the TLB flush */
3758 cpumask_clear_cpu(pcpu0, &kvm->arch.need_tlb_flush);
3761 trace_hardirqs_on();
3762 guest_enter_irqoff();
3764 srcu_idx = srcu_read_lock(&kvm->srcu);
3766 this_cpu_disable_ftrace();
3768 trap = kvmhv_p9_guest_entry(vcpu, time_limit);
3769 vcpu->arch.trap = trap;
3771 this_cpu_enable_ftrace();
3773 srcu_read_unlock(&kvm->srcu, srcu_idx);
3775 mtspr(SPRN_LPID, kvm->arch.host_lpid);
3778 trace_hardirqs_off();
3779 set_irq_happened(trap);
3781 kvmppc_set_host_core(pcpu);
3786 cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
3790 /* cancel pending decrementer exception if DEC is now positive */
3791 if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
3792 kvmppc_core_dequeue_dec(vcpu);
3794 trace_kvm_guest_exit(vcpu);
3797 r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
3800 if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
3801 !kvmppc_vcpu_woken(vcpu)) {
3802 kvmppc_set_timer(vcpu);
3803 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
3804 if (signal_pending(current)) {
3805 vcpu->stat.signal_exits++;
3806 kvm_run->exit_reason = KVM_EXIT_INTR;
3807 vcpu->arch.ret = -EINTR;
3810 spin_lock(&vc->lock);
3811 kvmppc_vcore_blocked(vc);
3812 spin_unlock(&vc->lock);
3815 vcpu->arch.ceded = 0;
3817 vc->vcore_state = VCORE_INACTIVE;
3818 trace_kvmppc_run_core(vc, 1);
3821 kvmppc_remove_runnable(vc, vcpu);
3822 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3824 return vcpu->arch.ret;
3827 vcpu->stat.signal_exits++;
3828 kvm_run->exit_reason = KVM_EXIT_INTR;
3829 vcpu->arch.ret = -EINTR;
3836 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3840 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3841 unsigned long user_tar = 0;
3842 unsigned int user_vrsave;
3845 if (!vcpu->arch.sane) {
3846 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3851 * Don't allow entry with a suspended transaction, because
3852 * the guest entry/exit code will lose it.
3853 * If the guest has TM enabled, save away their TM-related SPRs
3854 * (they will get restored by the TM unavailable interrupt).
3856 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3857 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3858 (current->thread.regs->msr & MSR_TM)) {
3859 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3860 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3861 run->fail_entry.hardware_entry_failure_reason = 0;
3864 /* Enable TM so we can read the TM SPRs */
3865 mtmsr(mfmsr() | MSR_TM);
3866 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3867 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3868 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3869 current->thread.regs->msr &= ~MSR_TM;
3874 * Force online to 1 for the sake of old userspace which doesn't
3877 if (!vcpu->arch.online) {
3878 atomic_inc(&vcpu->arch.vcore->online_count);
3879 vcpu->arch.online = 1;
3882 kvmppc_core_prepare_to_enter(vcpu);
3884 /* No need to go into the guest when all we'll do is come back out */
3885 if (signal_pending(current)) {
3886 run->exit_reason = KVM_EXIT_INTR;
3891 atomic_inc(&kvm->arch.vcpus_running);
3892 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3895 flush_all_to_thread(current);
3897 /* Save userspace EBB and other register values */
3898 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3899 ebb_regs[0] = mfspr(SPRN_EBBHR);
3900 ebb_regs[1] = mfspr(SPRN_EBBRR);
3901 ebb_regs[2] = mfspr(SPRN_BESCR);
3902 user_tar = mfspr(SPRN_TAR);
3904 user_vrsave = mfspr(SPRN_VRSAVE);
3906 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3907 vcpu->arch.pgdir = current->mm->pgd;
3908 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3911 if (kvm->arch.threads_indep && kvm_is_radix(kvm))
3912 r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0);
3914 r = kvmppc_run_vcpu(run, vcpu);
3916 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3917 !(vcpu->arch.shregs.msr & MSR_PR)) {
3918 trace_kvm_hcall_enter(vcpu);
3919 r = kvmppc_pseries_do_hcall(vcpu);
3920 trace_kvm_hcall_exit(vcpu, r);
3921 kvmppc_core_prepare_to_enter(vcpu);
3922 } else if (r == RESUME_PAGE_FAULT) {
3923 srcu_idx = srcu_read_lock(&kvm->srcu);
3924 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3925 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3926 srcu_read_unlock(&kvm->srcu, srcu_idx);
3927 } else if (r == RESUME_PASSTHROUGH) {
3928 if (WARN_ON(xive_enabled()))
3931 r = kvmppc_xics_rm_complete(vcpu, 0);
3933 } while (is_kvmppc_resume_guest(r));
3935 /* Restore userspace EBB and other register values */
3936 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3937 mtspr(SPRN_EBBHR, ebb_regs[0]);
3938 mtspr(SPRN_EBBRR, ebb_regs[1]);
3939 mtspr(SPRN_BESCR, ebb_regs[2]);
3940 mtspr(SPRN_TAR, user_tar);
3941 mtspr(SPRN_FSCR, current->thread.fscr);
3943 mtspr(SPRN_VRSAVE, user_vrsave);
3945 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3946 atomic_dec(&kvm->arch.vcpus_running);
3950 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3951 int shift, int sllp)
3953 (*sps)->page_shift = shift;
3954 (*sps)->slb_enc = sllp;
3955 (*sps)->enc[0].page_shift = shift;
3956 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3958 * Add 16MB MPSS support (may get filtered out by userspace)
3961 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3963 (*sps)->enc[1].page_shift = 24;
3964 (*sps)->enc[1].pte_enc = penc;
3970 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3971 struct kvm_ppc_smmu_info *info)
3973 struct kvm_ppc_one_seg_page_size *sps;
3976 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3977 * POWER7 doesn't support keys for instruction accesses,
3978 * POWER8 and POWER9 do.
3980 info->data_keys = 32;
3981 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3983 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3984 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3985 info->slb_size = 32;
3987 /* We only support these sizes for now, and no muti-size segments */
3988 sps = &info->sps[0];
3989 kvmppc_add_seg_page_size(&sps, 12, 0);
3990 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3991 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3997 * Get (and clear) the dirty memory log for a memory slot.
3999 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4000 struct kvm_dirty_log *log)
4002 struct kvm_memslots *slots;
4003 struct kvm_memory_slot *memslot;
4006 unsigned long *buf, *p;
4007 struct kvm_vcpu *vcpu;
4009 mutex_lock(&kvm->slots_lock);
4012 if (log->slot >= KVM_USER_MEM_SLOTS)
4015 slots = kvm_memslots(kvm);
4016 memslot = id_to_memslot(slots, log->slot);
4018 if (!memslot->dirty_bitmap)
4022 * Use second half of bitmap area because both HPT and radix
4023 * accumulate bits in the first half.
4025 n = kvm_dirty_bitmap_bytes(memslot);
4026 buf = memslot->dirty_bitmap + n / sizeof(long);
4029 if (kvm_is_radix(kvm))
4030 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4032 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4037 * We accumulate dirty bits in the first half of the
4038 * memslot's dirty_bitmap area, for when pages are paged
4039 * out or modified by the host directly. Pick up these
4040 * bits and add them to the map.
4042 p = memslot->dirty_bitmap;
4043 for (i = 0; i < n / sizeof(long); ++i)
4044 buf[i] |= xchg(&p[i], 0);
4046 /* Harvest dirty bits from VPA and DTL updates */
4047 /* Note: we never modify the SLB shadow buffer areas */
4048 kvm_for_each_vcpu(i, vcpu, kvm) {
4049 spin_lock(&vcpu->arch.vpa_update_lock);
4050 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4051 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4052 spin_unlock(&vcpu->arch.vpa_update_lock);
4056 if (copy_to_user(log->dirty_bitmap, buf, n))
4061 mutex_unlock(&kvm->slots_lock);
4065 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4066 struct kvm_memory_slot *dont)
4068 if (!dont || free->arch.rmap != dont->arch.rmap) {
4069 vfree(free->arch.rmap);
4070 free->arch.rmap = NULL;
4074 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4075 unsigned long npages)
4077 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4078 if (!slot->arch.rmap)
4084 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4085 struct kvm_memory_slot *memslot,
4086 const struct kvm_userspace_memory_region *mem)
4091 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4092 const struct kvm_userspace_memory_region *mem,
4093 const struct kvm_memory_slot *old,
4094 const struct kvm_memory_slot *new)
4096 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4099 * If we are making a new memslot, it might make
4100 * some address that was previously cached as emulated
4101 * MMIO be no longer emulated MMIO, so invalidate
4102 * all the caches of emulated MMIO translations.
4105 atomic64_inc(&kvm->arch.mmio_update);
4109 * Update LPCR values in kvm->arch and in vcores.
4110 * Caller must hold kvm->lock.
4112 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4117 if ((kvm->arch.lpcr & mask) == lpcr)
4120 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4122 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4123 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4126 spin_lock(&vc->lock);
4127 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4128 spin_unlock(&vc->lock);
4129 if (++cores_done >= kvm->arch.online_vcores)
4134 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4139 void kvmppc_setup_partition_table(struct kvm *kvm)
4141 unsigned long dw0, dw1;
4143 if (!kvm_is_radix(kvm)) {
4144 /* PS field - page size for VRMA */
4145 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4146 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4147 /* HTABSIZE and HTABORG fields */
4148 dw0 |= kvm->arch.sdr1;
4150 /* Second dword as set by userspace */
4151 dw1 = kvm->arch.process_table;
4153 dw0 = PATB_HR | radix__get_tree_size() |
4154 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4155 dw1 = PATB_GR | kvm->arch.process_table;
4158 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
4162 * Set up HPT (hashed page table) and RMA (real-mode area).
4163 * Must be called with kvm->lock held.
4165 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4168 struct kvm *kvm = vcpu->kvm;
4170 struct kvm_memory_slot *memslot;
4171 struct vm_area_struct *vma;
4172 unsigned long lpcr = 0, senc;
4173 unsigned long psize, porder;
4176 /* Allocate hashed page table (if not done already) and reset it */
4177 if (!kvm->arch.hpt.virt) {
4178 int order = KVM_DEFAULT_HPT_ORDER;
4179 struct kvm_hpt_info info;
4181 err = kvmppc_allocate_hpt(&info, order);
4182 /* If we get here, it means userspace didn't specify a
4183 * size explicitly. So, try successively smaller
4184 * sizes if the default failed. */
4185 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4186 err = kvmppc_allocate_hpt(&info, order);
4189 pr_err("KVM: Couldn't alloc HPT\n");
4193 kvmppc_set_hpt(kvm, &info);
4196 /* Look up the memslot for guest physical address 0 */
4197 srcu_idx = srcu_read_lock(&kvm->srcu);
4198 memslot = gfn_to_memslot(kvm, 0);
4200 /* We must have some memory at 0 by now */
4202 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4205 /* Look up the VMA for the start of this memory slot */
4206 hva = memslot->userspace_addr;
4207 down_read(¤t->mm->mmap_sem);
4208 vma = find_vma(current->mm, hva);
4209 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4212 psize = vma_kernel_pagesize(vma);
4214 up_read(¤t->mm->mmap_sem);
4216 /* We can handle 4k, 64k or 16M pages in the VRMA */
4217 if (psize >= 0x1000000)
4219 else if (psize >= 0x10000)
4223 porder = __ilog2(psize);
4225 senc = slb_pgsize_encoding(psize);
4226 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4227 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4228 /* Create HPTEs in the hash page table for the VRMA */
4229 kvmppc_map_vrma(vcpu, memslot, porder);
4231 /* Update VRMASD field in the LPCR */
4232 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4233 /* the -4 is to account for senc values starting at 0x10 */
4234 lpcr = senc << (LPCR_VRMASD_SH - 4);
4235 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4238 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4242 srcu_read_unlock(&kvm->srcu, srcu_idx);
4247 up_read(¤t->mm->mmap_sem);
4251 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4252 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4254 kvmppc_free_radix(kvm);
4255 kvmppc_update_lpcr(kvm, LPCR_VPM1,
4256 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4257 kvmppc_rmap_reset(kvm);
4258 kvm->arch.radix = 0;
4259 kvm->arch.process_table = 0;
4263 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4264 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4268 err = kvmppc_init_vm_radix(kvm);
4272 kvmppc_free_hpt(&kvm->arch.hpt);
4273 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4274 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4275 kvm->arch.radix = 1;
4279 #ifdef CONFIG_KVM_XICS
4281 * Allocate a per-core structure for managing state about which cores are
4282 * running in the host versus the guest and for exchanging data between
4283 * real mode KVM and CPU running in the host.
4284 * This is only done for the first VM.
4285 * The allocated structure stays even if all VMs have stopped.
4286 * It is only freed when the kvm-hv module is unloaded.
4287 * It's OK for this routine to fail, we just don't support host
4288 * core operations like redirecting H_IPI wakeups.
4290 void kvmppc_alloc_host_rm_ops(void)
4292 struct kvmppc_host_rm_ops *ops;
4293 unsigned long l_ops;
4297 /* Not the first time here ? */
4298 if (kvmppc_host_rm_ops_hv != NULL)
4301 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4305 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4306 ops->rm_core = kzalloc(size, GFP_KERNEL);
4308 if (!ops->rm_core) {
4315 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4316 if (!cpu_online(cpu))
4319 core = cpu >> threads_shift;
4320 ops->rm_core[core].rm_state.in_host = 1;
4323 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4326 * Make the contents of the kvmppc_host_rm_ops structure visible
4327 * to other CPUs before we assign it to the global variable.
4328 * Do an atomic assignment (no locks used here), but if someone
4329 * beats us to it, just free our copy and return.
4332 l_ops = (unsigned long) ops;
4334 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4336 kfree(ops->rm_core);
4341 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4342 "ppc/kvm_book3s:prepare",
4343 kvmppc_set_host_core,
4344 kvmppc_clear_host_core);
4348 void kvmppc_free_host_rm_ops(void)
4350 if (kvmppc_host_rm_ops_hv) {
4351 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4352 kfree(kvmppc_host_rm_ops_hv->rm_core);
4353 kfree(kvmppc_host_rm_ops_hv);
4354 kvmppc_host_rm_ops_hv = NULL;
4359 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4361 unsigned long lpcr, lpid;
4365 /* Allocate the guest's logical partition ID */
4367 lpid = kvmppc_alloc_lpid();
4370 kvm->arch.lpid = lpid;
4372 kvmppc_alloc_host_rm_ops();
4375 * Since we don't flush the TLB when tearing down a VM,
4376 * and this lpid might have previously been used,
4377 * make sure we flush on each core before running the new VM.
4378 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4379 * does this flush for us.
4381 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4382 cpumask_setall(&kvm->arch.need_tlb_flush);
4384 /* Start out with the default set of hcalls enabled */
4385 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4386 sizeof(kvm->arch.enabled_hcalls));
4388 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4389 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4391 /* Init LPCR for virtual RMA mode */
4392 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4393 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4394 lpcr &= LPCR_PECE | LPCR_LPES;
4395 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4396 LPCR_VPM0 | LPCR_VPM1;
4397 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4398 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4399 /* On POWER8 turn on online bit to enable PURR/SPURR */
4400 if (cpu_has_feature(CPU_FTR_ARCH_207S))
4403 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4404 * Set HVICE bit to enable hypervisor virtualization interrupts.
4405 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4406 * be unnecessary but better safe than sorry in case we re-enable
4407 * EE in HV mode with this LPCR still set)
4409 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4411 lpcr |= LPCR_HVICE | LPCR_HEIC;
4414 * If xive is enabled, we route 0x500 interrupts directly
4422 * If the host uses radix, the guest starts out as radix.
4424 if (radix_enabled()) {
4425 kvm->arch.radix = 1;
4426 kvm->arch.mmu_ready = 1;
4428 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4429 ret = kvmppc_init_vm_radix(kvm);
4431 kvmppc_free_lpid(kvm->arch.lpid);
4434 kvmppc_setup_partition_table(kvm);
4437 kvm->arch.lpcr = lpcr;
4439 /* Initialization for future HPT resizes */
4440 kvm->arch.resize_hpt = NULL;
4443 * Work out how many sets the TLB has, for the use of
4444 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4446 if (radix_enabled())
4447 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
4448 else if (cpu_has_feature(CPU_FTR_ARCH_300))
4449 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
4450 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4451 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
4453 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
4456 * Track that we now have a HV mode VM active. This blocks secondary
4457 * CPU threads from coming online.
4458 * On POWER9, we only need to do this if the "indep_threads_mode"
4459 * module parameter has been set to N.
4461 if (cpu_has_feature(CPU_FTR_ARCH_300))
4462 kvm->arch.threads_indep = indep_threads_mode;
4463 if (!kvm->arch.threads_indep)
4464 kvm_hv_vm_activated();
4467 * Initialize smt_mode depending on processor.
4468 * POWER8 and earlier have to use "strict" threading, where
4469 * all vCPUs in a vcore have to run on the same (sub)core,
4470 * whereas on POWER9 the threads can each run a different
4473 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4474 kvm->arch.smt_mode = threads_per_subcore;
4476 kvm->arch.smt_mode = 1;
4477 kvm->arch.emul_smt_mode = 1;
4480 * Create a debugfs directory for the VM
4482 snprintf(buf, sizeof(buf), "vm%d", current->pid);
4483 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4484 kvmppc_mmu_debugfs_init(kvm);
4489 static void kvmppc_free_vcores(struct kvm *kvm)
4493 for (i = 0; i < KVM_MAX_VCORES; ++i)
4494 kfree(kvm->arch.vcores[i]);
4495 kvm->arch.online_vcores = 0;
4498 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4500 debugfs_remove_recursive(kvm->arch.debugfs_dir);
4502 if (!kvm->arch.threads_indep)
4503 kvm_hv_vm_deactivated();
4505 kvmppc_free_vcores(kvm);
4507 kvmppc_free_lpid(kvm->arch.lpid);
4509 if (kvm_is_radix(kvm))
4510 kvmppc_free_radix(kvm);
4512 kvmppc_free_hpt(&kvm->arch.hpt);
4514 kvmppc_free_pimap(kvm);
4517 /* We don't need to emulate any privileged instructions or dcbz */
4518 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4519 unsigned int inst, int *advance)
4521 return EMULATE_FAIL;
4524 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4527 return EMULATE_FAIL;
4530 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4533 return EMULATE_FAIL;
4536 static int kvmppc_core_check_processor_compat_hv(void)
4538 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
4539 !cpu_has_feature(CPU_FTR_ARCH_206))
4545 #ifdef CONFIG_KVM_XICS
4547 void kvmppc_free_pimap(struct kvm *kvm)
4549 kfree(kvm->arch.pimap);
4552 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4554 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4557 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4559 struct irq_desc *desc;
4560 struct kvmppc_irq_map *irq_map;
4561 struct kvmppc_passthru_irqmap *pimap;
4562 struct irq_chip *chip;
4565 if (!kvm_irq_bypass)
4568 desc = irq_to_desc(host_irq);
4572 mutex_lock(&kvm->lock);
4574 pimap = kvm->arch.pimap;
4575 if (pimap == NULL) {
4576 /* First call, allocate structure to hold IRQ map */
4577 pimap = kvmppc_alloc_pimap();
4578 if (pimap == NULL) {
4579 mutex_unlock(&kvm->lock);
4582 kvm->arch.pimap = pimap;
4586 * For now, we only support interrupts for which the EOI operation
4587 * is an OPAL call followed by a write to XIRR, since that's
4588 * what our real-mode EOI code does, or a XIVE interrupt
4590 chip = irq_data_get_irq_chip(&desc->irq_data);
4591 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4592 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4593 host_irq, guest_gsi);
4594 mutex_unlock(&kvm->lock);
4599 * See if we already have an entry for this guest IRQ number.
4600 * If it's mapped to a hardware IRQ number, that's an error,
4601 * otherwise re-use this entry.
4603 for (i = 0; i < pimap->n_mapped; i++) {
4604 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4605 if (pimap->mapped[i].r_hwirq) {
4606 mutex_unlock(&kvm->lock);
4613 if (i == KVMPPC_PIRQ_MAPPED) {
4614 mutex_unlock(&kvm->lock);
4615 return -EAGAIN; /* table is full */
4618 irq_map = &pimap->mapped[i];
4620 irq_map->v_hwirq = guest_gsi;
4621 irq_map->desc = desc;
4624 * Order the above two stores before the next to serialize with
4625 * the KVM real mode handler.
4628 irq_map->r_hwirq = desc->irq_data.hwirq;
4630 if (i == pimap->n_mapped)
4634 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4636 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4638 irq_map->r_hwirq = 0;
4640 mutex_unlock(&kvm->lock);
4645 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4647 struct irq_desc *desc;
4648 struct kvmppc_passthru_irqmap *pimap;
4651 if (!kvm_irq_bypass)
4654 desc = irq_to_desc(host_irq);
4658 mutex_lock(&kvm->lock);
4659 if (!kvm->arch.pimap)
4662 pimap = kvm->arch.pimap;
4664 for (i = 0; i < pimap->n_mapped; i++) {
4665 if (guest_gsi == pimap->mapped[i].v_hwirq)
4669 if (i == pimap->n_mapped) {
4670 mutex_unlock(&kvm->lock);
4675 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4677 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4679 /* invalidate the entry (what do do on error from the above ?) */
4680 pimap->mapped[i].r_hwirq = 0;
4683 * We don't free this structure even when the count goes to
4684 * zero. The structure is freed when we destroy the VM.
4687 mutex_unlock(&kvm->lock);
4691 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4692 struct irq_bypass_producer *prod)
4695 struct kvm_kernel_irqfd *irqfd =
4696 container_of(cons, struct kvm_kernel_irqfd, consumer);
4698 irqfd->producer = prod;
4700 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4702 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4703 prod->irq, irqfd->gsi, ret);
4708 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4709 struct irq_bypass_producer *prod)
4712 struct kvm_kernel_irqfd *irqfd =
4713 container_of(cons, struct kvm_kernel_irqfd, consumer);
4715 irqfd->producer = NULL;
4718 * When producer of consumer is unregistered, we change back to
4719 * default external interrupt handling mode - KVM real mode
4720 * will switch back to host.
4722 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4724 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4725 prod->irq, irqfd->gsi, ret);
4729 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4730 unsigned int ioctl, unsigned long arg)
4732 struct kvm *kvm __maybe_unused = filp->private_data;
4733 void __user *argp = (void __user *)arg;
4738 case KVM_PPC_ALLOCATE_HTAB: {
4742 if (get_user(htab_order, (u32 __user *)argp))
4744 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4751 case KVM_PPC_GET_HTAB_FD: {
4752 struct kvm_get_htab_fd ghf;
4755 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4757 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4761 case KVM_PPC_RESIZE_HPT_PREPARE: {
4762 struct kvm_ppc_resize_hpt rhpt;
4765 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4768 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4772 case KVM_PPC_RESIZE_HPT_COMMIT: {
4773 struct kvm_ppc_resize_hpt rhpt;
4776 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4779 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4791 * List of hcall numbers to enable by default.
4792 * For compatibility with old userspace, we enable by default
4793 * all hcalls that were implemented before the hcall-enabling
4794 * facility was added. Note this list should not include H_RTAS.
4796 static unsigned int default_hcall_list[] = {
4810 #ifdef CONFIG_KVM_XICS
4821 static void init_default_hcalls(void)
4826 for (i = 0; default_hcall_list[i]; ++i) {
4827 hcall = default_hcall_list[i];
4828 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4829 __set_bit(hcall / 4, default_enabled_hcalls);
4833 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4839 /* If not on a POWER9, reject it */
4840 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4843 /* If any unknown flags set, reject it */
4844 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4847 /* GR (guest radix) bit in process_table field must match */
4848 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4849 if (!!(cfg->process_table & PATB_GR) != radix)
4852 /* Process table size field must be reasonable, i.e. <= 24 */
4853 if ((cfg->process_table & PRTS_MASK) > 24)
4856 /* We can change a guest to/from radix now, if the host is radix */
4857 if (radix && !radix_enabled())
4860 mutex_lock(&kvm->lock);
4861 if (radix != kvm_is_radix(kvm)) {
4862 if (kvm->arch.mmu_ready) {
4863 kvm->arch.mmu_ready = 0;
4864 /* order mmu_ready vs. vcpus_running */
4866 if (atomic_read(&kvm->arch.vcpus_running)) {
4867 kvm->arch.mmu_ready = 1;
4873 err = kvmppc_switch_mmu_to_radix(kvm);
4875 err = kvmppc_switch_mmu_to_hpt(kvm);
4880 kvm->arch.process_table = cfg->process_table;
4881 kvmppc_setup_partition_table(kvm);
4883 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4884 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4888 mutex_unlock(&kvm->lock);
4892 static struct kvmppc_ops kvm_ops_hv = {
4893 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4894 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4895 .get_one_reg = kvmppc_get_one_reg_hv,
4896 .set_one_reg = kvmppc_set_one_reg_hv,
4897 .vcpu_load = kvmppc_core_vcpu_load_hv,
4898 .vcpu_put = kvmppc_core_vcpu_put_hv,
4899 .set_msr = kvmppc_set_msr_hv,
4900 .vcpu_run = kvmppc_vcpu_run_hv,
4901 .vcpu_create = kvmppc_core_vcpu_create_hv,
4902 .vcpu_free = kvmppc_core_vcpu_free_hv,
4903 .check_requests = kvmppc_core_check_requests_hv,
4904 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4905 .flush_memslot = kvmppc_core_flush_memslot_hv,
4906 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4907 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4908 .unmap_hva_range = kvm_unmap_hva_range_hv,
4909 .age_hva = kvm_age_hva_hv,
4910 .test_age_hva = kvm_test_age_hva_hv,
4911 .set_spte_hva = kvm_set_spte_hva_hv,
4912 .mmu_destroy = kvmppc_mmu_destroy_hv,
4913 .free_memslot = kvmppc_core_free_memslot_hv,
4914 .create_memslot = kvmppc_core_create_memslot_hv,
4915 .init_vm = kvmppc_core_init_vm_hv,
4916 .destroy_vm = kvmppc_core_destroy_vm_hv,
4917 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4918 .emulate_op = kvmppc_core_emulate_op_hv,
4919 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4920 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4921 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4922 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4923 .hcall_implemented = kvmppc_hcall_impl_hv,
4924 #ifdef CONFIG_KVM_XICS
4925 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4926 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4928 .configure_mmu = kvmhv_configure_mmu,
4929 .get_rmmu_info = kvmhv_get_rmmu_info,
4930 .set_smt_mode = kvmhv_set_smt_mode,
4933 static int kvm_init_subcore_bitmap(void)
4936 int nr_cores = cpu_nr_cores();
4937 struct sibling_subcore_state *sibling_subcore_state;
4939 for (i = 0; i < nr_cores; i++) {
4940 int first_cpu = i * threads_per_core;
4941 int node = cpu_to_node(first_cpu);
4943 /* Ignore if it is already allocated. */
4944 if (paca_ptrs[first_cpu]->sibling_subcore_state)
4947 sibling_subcore_state =
4948 kmalloc_node(sizeof(struct sibling_subcore_state),
4950 if (!sibling_subcore_state)
4953 memset(sibling_subcore_state, 0,
4954 sizeof(struct sibling_subcore_state));
4956 for (j = 0; j < threads_per_core; j++) {
4957 int cpu = first_cpu + j;
4959 paca_ptrs[cpu]->sibling_subcore_state =
4960 sibling_subcore_state;
4966 static int kvmppc_radix_possible(void)
4968 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4971 static int kvmppc_book3s_init_hv(void)
4975 * FIXME!! Do we need to check on all cpus ?
4977 r = kvmppc_core_check_processor_compat_hv();
4981 r = kvm_init_subcore_bitmap();
4986 * We need a way of accessing the XICS interrupt controller,
4987 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
4988 * indirectly, via OPAL.
4991 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4992 struct device_node *np;
4994 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4996 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4999 /* presence of intc confirmed - node can be dropped again */
5004 kvm_ops_hv.owner = THIS_MODULE;
5005 kvmppc_hv_ops = &kvm_ops_hv;
5007 init_default_hcalls();
5011 r = kvmppc_mmu_hv_init();
5015 if (kvmppc_radix_possible())
5016 r = kvmppc_radix_init();
5019 * POWER9 chips before version 2.02 can't have some threads in
5020 * HPT mode and some in radix mode on the same core.
5022 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5023 unsigned int pvr = mfspr(SPRN_PVR);
5024 if ((pvr >> 16) == PVR_POWER9 &&
5025 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5026 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5027 no_mixing_hpt_and_radix = true;
5033 static void kvmppc_book3s_exit_hv(void)
5035 kvmppc_free_host_rm_ops();
5036 if (kvmppc_radix_possible())
5037 kvmppc_radix_exit();
5038 kvmppc_hv_ops = NULL;
5041 module_init(kvmppc_book3s_init_hv);
5042 module_exit(kvmppc_book3s_exit_hv);
5043 MODULE_LICENSE("GPL");
5044 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5045 MODULE_ALIAS("devname:kvm");