1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
7 * Paul Mackerras <paulus@au1.ibm.com>
8 * Alexander Graf <agraf@suse.de>
9 * Kevin Wolf <mail@kevin-wolf.de>
11 * Description: KVM functions specific to running on Book 3S
12 * processors in hypervisor mode (specifically POWER7 and later).
14 * This file is derived from arch/powerpc/kvm/book3s.c,
15 * by Alexander Graf <agraf@suse.de>.
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
46 #include <asm/ftrace.h>
48 #include <asm/ppc-opcode.h>
49 #include <asm/asm-prototypes.h>
50 #include <asm/archrandom.h>
51 #include <asm/debug.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <linux/uaccess.h>
56 #include <asm/interrupt.h>
58 #include <asm/kvm_ppc.h>
59 #include <asm/kvm_book3s.h>
60 #include <asm/mmu_context.h>
61 #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>
76 #include <asm/hw_breakpoint.h>
77 #include <asm/kvm_book3s_uvmem.h>
78 #include <asm/ultravisor.h>
80 #include <asm/plpar_wrappers.h>
83 #include "book3s_hv.h"
85 #define CREATE_TRACE_POINTS
88 /* #define EXIT_DEBUG */
89 /* #define EXIT_DEBUG_SIMPLE */
90 /* #define EXIT_DEBUG_INT */
92 /* Used to indicate that a guest page fault needs to be handled */
93 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
94 /* Used to indicate that a guest passthrough interrupt needs to be handled */
95 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
97 /* Used as a "null" value for timebase values */
98 #define TB_NIL (~(u64)0)
100 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
102 static int dynamic_mt_modes = 6;
103 module_param(dynamic_mt_modes, int, 0644);
104 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
105 static int target_smt_mode;
106 module_param(target_smt_mode, int, 0644);
107 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
109 static bool one_vm_per_core;
110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
113 #ifdef CONFIG_KVM_XICS
114 static const struct kernel_param_ops module_param_ops = {
115 .set = param_set_int,
116 .get = param_get_int,
119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
126 /* If set, guests are allowed to create and control nested guests */
127 static bool nested = true;
128 module_param(nested, bool, S_IRUGO | S_IWUSR);
129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
131 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
134 * RWMR values for POWER8. These control the rate at which PURR
135 * and SPURR count and should be set according to the number of
136 * online threads in the vcore being run.
138 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
139 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
140 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
141 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
142 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
143 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
144 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
145 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
147 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
159 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
163 struct kvm_vcpu *vcpu;
165 while (++i < MAX_SMT_THREADS) {
166 vcpu = READ_ONCE(vc->runnable_threads[i]);
175 /* Used to traverse the list of runnable threads for a given vcore */
176 #define for_each_runnable_thread(i, vcpu, vc) \
177 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
179 static bool kvmppc_ipi_thread(int cpu)
181 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
183 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
184 if (kvmhv_on_pseries())
187 /* On POWER9 we can use msgsnd to IPI any cpu */
188 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
189 msg |= get_hard_smp_processor_id(cpu);
191 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
195 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
196 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
198 if (cpu_first_thread_sibling(cpu) ==
199 cpu_first_thread_sibling(smp_processor_id())) {
200 msg |= cpu_thread_in_core(cpu);
202 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
209 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
210 if (cpu >= 0 && cpu < nr_cpu_ids) {
211 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
215 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
223 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
226 struct rcuwait *waitp;
228 waitp = kvm_arch_vcpu_get_wait(vcpu);
229 if (rcuwait_wake_up(waitp))
230 ++vcpu->stat.generic.halt_wakeup;
232 cpu = READ_ONCE(vcpu->arch.thread_cpu);
233 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
236 /* CPU points to the first thread of the core */
238 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
239 smp_send_reschedule(cpu);
243 * We use the vcpu_load/put functions to measure stolen time.
244 * Stolen time is counted as time when either the vcpu is able to
245 * run as part of a virtual core, but the task running the vcore
246 * is preempted or sleeping, or when the vcpu needs something done
247 * in the kernel by the task running the vcpu, but that task is
248 * preempted or sleeping. Those two things have to be counted
249 * separately, since one of the vcpu tasks will take on the job
250 * of running the core, and the other vcpu tasks in the vcore will
251 * sleep waiting for it to do that, but that sleep shouldn't count
254 * Hence we accumulate stolen time when the vcpu can run as part of
255 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
256 * needs its task to do other things in the kernel (for example,
257 * service a page fault) in busy_stolen. We don't accumulate
258 * stolen time for a vcore when it is inactive, or for a vcpu
259 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
260 * a misnomer; it means that the vcpu task is not executing in
261 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
262 * the kernel. We don't have any way of dividing up that time
263 * between time that the vcpu is genuinely stopped, time that
264 * the task is actively working on behalf of the vcpu, and time
265 * that the task is preempted, so we don't count any of it as
268 * Updates to busy_stolen are protected by arch.tbacct_lock;
269 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
270 * lock. The stolen times are measured in units of timebase ticks.
271 * (Note that the != TB_NIL checks below are purely defensive;
272 * they should never fail.)
275 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb)
279 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
281 spin_lock_irqsave(&vc->stoltb_lock, flags);
283 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
286 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb)
290 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
292 spin_lock_irqsave(&vc->stoltb_lock, flags);
293 if (vc->preempt_tb != TB_NIL) {
294 vc->stolen_tb += tb - vc->preempt_tb;
295 vc->preempt_tb = TB_NIL;
297 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
300 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
302 struct kvmppc_vcore *vc = vcpu->arch.vcore;
306 if (cpu_has_feature(CPU_FTR_ARCH_300))
312 * We can test vc->runner without taking the vcore lock,
313 * because only this task ever sets vc->runner to this
314 * vcpu, and once it is set to this vcpu, only this task
315 * ever sets it to NULL.
317 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
318 kvmppc_core_end_stolen(vc, now);
320 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
321 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
322 vcpu->arch.busy_preempt != TB_NIL) {
323 vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt;
324 vcpu->arch.busy_preempt = TB_NIL;
326 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
329 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
331 struct kvmppc_vcore *vc = vcpu->arch.vcore;
335 if (cpu_has_feature(CPU_FTR_ARCH_300))
340 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
341 kvmppc_core_start_stolen(vc, now);
343 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
344 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
345 vcpu->arch.busy_preempt = now;
346 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
349 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
351 vcpu->arch.pvr = pvr;
354 /* Dummy value used in computing PCR value below */
355 #define PCR_ARCH_31 (PCR_ARCH_300 << 1)
357 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
359 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
360 struct kvmppc_vcore *vc = vcpu->arch.vcore;
362 /* We can (emulate) our own architecture version and anything older */
363 if (cpu_has_feature(CPU_FTR_ARCH_31))
364 host_pcr_bit = PCR_ARCH_31;
365 else if (cpu_has_feature(CPU_FTR_ARCH_300))
366 host_pcr_bit = PCR_ARCH_300;
367 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
368 host_pcr_bit = PCR_ARCH_207;
369 else if (cpu_has_feature(CPU_FTR_ARCH_206))
370 host_pcr_bit = PCR_ARCH_206;
372 host_pcr_bit = PCR_ARCH_205;
374 /* Determine lowest PCR bit needed to run guest in given PVR level */
375 guest_pcr_bit = host_pcr_bit;
377 switch (arch_compat) {
379 guest_pcr_bit = PCR_ARCH_205;
383 guest_pcr_bit = PCR_ARCH_206;
386 guest_pcr_bit = PCR_ARCH_207;
389 guest_pcr_bit = PCR_ARCH_300;
392 guest_pcr_bit = PCR_ARCH_31;
399 /* Check requested PCR bits don't exceed our capabilities */
400 if (guest_pcr_bit > host_pcr_bit)
403 spin_lock(&vc->lock);
404 vc->arch_compat = arch_compat;
406 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
407 * Also set all reserved PCR bits
409 vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
410 spin_unlock(&vc->lock);
415 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
419 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
420 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
421 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
422 for (r = 0; r < 16; ++r)
423 pr_err("r%2d = %.16lx r%d = %.16lx\n",
424 r, kvmppc_get_gpr(vcpu, r),
425 r+16, kvmppc_get_gpr(vcpu, r+16));
426 pr_err("ctr = %.16lx lr = %.16lx\n",
427 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
428 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
429 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
430 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
431 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
432 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
433 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
434 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
435 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
436 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
437 pr_err("fault dar = %.16lx dsisr = %.8x\n",
438 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
439 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
440 for (r = 0; r < vcpu->arch.slb_max; ++r)
441 pr_err(" ESID = %.16llx VSID = %.16llx\n",
442 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
443 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
444 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
445 vcpu->arch.last_inst);
448 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
450 return kvm_get_vcpu_by_id(kvm, id);
453 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
455 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
456 vpa->yield_count = cpu_to_be32(1);
459 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
460 unsigned long addr, unsigned long len)
462 /* check address is cacheline aligned */
463 if (addr & (L1_CACHE_BYTES - 1))
465 spin_lock(&vcpu->arch.vpa_update_lock);
466 if (v->next_gpa != addr || v->len != len) {
468 v->len = addr ? len : 0;
469 v->update_pending = 1;
471 spin_unlock(&vcpu->arch.vpa_update_lock);
475 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
484 static int vpa_is_registered(struct kvmppc_vpa *vpap)
486 if (vpap->update_pending)
487 return vpap->next_gpa != 0;
488 return vpap->pinned_addr != NULL;
491 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
493 unsigned long vcpuid, unsigned long vpa)
495 struct kvm *kvm = vcpu->kvm;
496 unsigned long len, nb;
498 struct kvm_vcpu *tvcpu;
501 struct kvmppc_vpa *vpap;
503 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
507 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
508 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
509 subfunc == H_VPA_REG_SLB) {
510 /* Registering new area - address must be cache-line aligned */
511 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
514 /* convert logical addr to kernel addr and read length */
515 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
518 if (subfunc == H_VPA_REG_VPA)
519 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
521 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
522 kvmppc_unpin_guest_page(kvm, va, vpa, false);
525 if (len > nb || len < sizeof(struct reg_vpa))
534 spin_lock(&tvcpu->arch.vpa_update_lock);
537 case H_VPA_REG_VPA: /* register VPA */
539 * The size of our lppaca is 1kB because of the way we align
540 * it for the guest to avoid crossing a 4kB boundary. We only
541 * use 640 bytes of the structure though, so we should accept
542 * clients that set a size of 640.
544 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
545 if (len < sizeof(struct lppaca))
547 vpap = &tvcpu->arch.vpa;
551 case H_VPA_REG_DTL: /* register DTL */
552 if (len < sizeof(struct dtl_entry))
554 len -= len % sizeof(struct dtl_entry);
556 /* Check that they have previously registered a VPA */
558 if (!vpa_is_registered(&tvcpu->arch.vpa))
561 vpap = &tvcpu->arch.dtl;
565 case H_VPA_REG_SLB: /* register SLB shadow buffer */
566 /* Check that they have previously registered a VPA */
568 if (!vpa_is_registered(&tvcpu->arch.vpa))
571 vpap = &tvcpu->arch.slb_shadow;
575 case H_VPA_DEREG_VPA: /* deregister VPA */
576 /* Check they don't still have a DTL or SLB buf registered */
578 if (vpa_is_registered(&tvcpu->arch.dtl) ||
579 vpa_is_registered(&tvcpu->arch.slb_shadow))
582 vpap = &tvcpu->arch.vpa;
586 case H_VPA_DEREG_DTL: /* deregister DTL */
587 vpap = &tvcpu->arch.dtl;
591 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
592 vpap = &tvcpu->arch.slb_shadow;
598 vpap->next_gpa = vpa;
600 vpap->update_pending = 1;
603 spin_unlock(&tvcpu->arch.vpa_update_lock);
608 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
610 struct kvm *kvm = vcpu->kvm;
616 * We need to pin the page pointed to by vpap->next_gpa,
617 * but we can't call kvmppc_pin_guest_page under the lock
618 * as it does get_user_pages() and down_read(). So we
619 * have to drop the lock, pin the page, then get the lock
620 * again and check that a new area didn't get registered
624 gpa = vpap->next_gpa;
625 spin_unlock(&vcpu->arch.vpa_update_lock);
629 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
630 spin_lock(&vcpu->arch.vpa_update_lock);
631 if (gpa == vpap->next_gpa)
633 /* sigh... unpin that one and try again */
635 kvmppc_unpin_guest_page(kvm, va, gpa, false);
638 vpap->update_pending = 0;
639 if (va && nb < vpap->len) {
641 * If it's now too short, it must be that userspace
642 * has changed the mappings underlying guest memory,
643 * so unregister the region.
645 kvmppc_unpin_guest_page(kvm, va, gpa, false);
648 if (vpap->pinned_addr)
649 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
652 vpap->pinned_addr = va;
655 vpap->pinned_end = va + vpap->len;
658 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
660 if (!(vcpu->arch.vpa.update_pending ||
661 vcpu->arch.slb_shadow.update_pending ||
662 vcpu->arch.dtl.update_pending))
665 spin_lock(&vcpu->arch.vpa_update_lock);
666 if (vcpu->arch.vpa.update_pending) {
667 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
668 if (vcpu->arch.vpa.pinned_addr)
669 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
671 if (vcpu->arch.dtl.update_pending) {
672 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
673 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
674 vcpu->arch.dtl_index = 0;
676 if (vcpu->arch.slb_shadow.update_pending)
677 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
678 spin_unlock(&vcpu->arch.vpa_update_lock);
682 * Return the accumulated stolen time for the vcore up until `now'.
683 * The caller should hold the vcore lock.
685 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
690 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
692 spin_lock_irqsave(&vc->stoltb_lock, flags);
694 if (vc->vcore_state != VCORE_INACTIVE &&
695 vc->preempt_tb != TB_NIL)
696 p += now - vc->preempt_tb;
697 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
701 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
702 unsigned int pcpu, u64 now,
703 unsigned long stolen)
705 struct dtl_entry *dt;
708 dt = vcpu->arch.dtl_ptr;
709 vpa = vcpu->arch.vpa.pinned_addr;
714 dt->dispatch_reason = 7;
715 dt->preempt_reason = 0;
716 dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
717 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
718 dt->ready_to_enqueue_time = 0;
719 dt->waiting_to_ready_time = 0;
720 dt->timebase = cpu_to_be64(now);
722 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
723 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
726 if (dt == vcpu->arch.dtl.pinned_end)
727 dt = vcpu->arch.dtl.pinned_addr;
728 vcpu->arch.dtl_ptr = dt;
729 /* order writing *dt vs. writing vpa->dtl_idx */
731 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
732 vcpu->arch.dtl.dirty = true;
735 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
736 struct kvmppc_vcore *vc)
738 unsigned long stolen;
739 unsigned long core_stolen;
745 core_stolen = vcore_stolen_time(vc, now);
746 stolen = core_stolen - vcpu->arch.stolen_logged;
747 vcpu->arch.stolen_logged = core_stolen;
748 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
749 stolen += vcpu->arch.busy_stolen;
750 vcpu->arch.busy_stolen = 0;
751 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
753 __kvmppc_create_dtl_entry(vcpu, vc->pcpu, now + vc->tb_offset, stolen);
756 /* See if there is a doorbell interrupt pending for a vcpu */
757 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
760 struct kvmppc_vcore *vc;
762 if (vcpu->arch.doorbell_request)
764 if (cpu_has_feature(CPU_FTR_ARCH_300))
767 * Ensure that the read of vcore->dpdes comes after the read
768 * of vcpu->doorbell_request. This barrier matches the
769 * smp_wmb() in kvmppc_guest_entry_inject().
772 vc = vcpu->arch.vcore;
773 thr = vcpu->vcpu_id - vc->first_vcpuid;
774 return !!(vc->dpdes & (1 << thr));
777 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
779 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
781 if ((!vcpu->arch.vcore->arch_compat) &&
782 cpu_has_feature(CPU_FTR_ARCH_207S))
787 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
788 unsigned long resource, unsigned long value1,
789 unsigned long value2)
792 case H_SET_MODE_RESOURCE_SET_CIABR:
793 if (!kvmppc_power8_compatible(vcpu))
798 return H_UNSUPPORTED_FLAG_START;
799 /* Guests can't breakpoint the hypervisor */
800 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
802 vcpu->arch.ciabr = value1;
804 case H_SET_MODE_RESOURCE_SET_DAWR0:
805 if (!kvmppc_power8_compatible(vcpu))
807 if (!ppc_breakpoint_available())
810 return H_UNSUPPORTED_FLAG_START;
811 if (value2 & DABRX_HYP)
813 vcpu->arch.dawr0 = value1;
814 vcpu->arch.dawrx0 = value2;
816 case H_SET_MODE_RESOURCE_SET_DAWR1:
817 if (!kvmppc_power8_compatible(vcpu))
819 if (!ppc_breakpoint_available())
821 if (!cpu_has_feature(CPU_FTR_DAWR1))
823 if (!vcpu->kvm->arch.dawr1_enabled)
826 return H_UNSUPPORTED_FLAG_START;
827 if (value2 & DABRX_HYP)
829 vcpu->arch.dawr1 = value1;
830 vcpu->arch.dawrx1 = value2;
832 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
834 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
835 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
837 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
838 kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
839 return H_UNSUPPORTED_FLAG_START;
846 /* Copy guest memory in place - must reside within a single memslot */
847 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
850 struct kvm_memory_slot *to_memslot = NULL;
851 struct kvm_memory_slot *from_memslot = NULL;
852 unsigned long to_addr, from_addr;
855 /* Get HPA for from address */
856 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
859 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
862 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
863 if (kvm_is_error_hva(from_addr))
865 from_addr |= (from & (PAGE_SIZE - 1));
867 /* Get HPA for to address */
868 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
871 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
874 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
875 if (kvm_is_error_hva(to_addr))
877 to_addr |= (to & (PAGE_SIZE - 1));
880 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
884 mark_page_dirty(kvm, to >> PAGE_SHIFT);
888 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
889 unsigned long dest, unsigned long src)
891 u64 pg_sz = SZ_4K; /* 4K page size */
892 u64 pg_mask = SZ_4K - 1;
895 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
896 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
897 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
900 /* dest (and src if copy_page flag set) must be page aligned */
901 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
904 /* zero and/or copy the page as determined by the flags */
905 if (flags & H_COPY_PAGE) {
906 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
909 } else if (flags & H_ZERO_PAGE) {
910 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
915 /* We can ignore the remaining flags */
920 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
922 struct kvmppc_vcore *vcore = target->arch.vcore;
925 * We expect to have been called by the real mode handler
926 * (kvmppc_rm_h_confer()) which would have directly returned
927 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
928 * have useful work to do and should not confer) so we don't
931 * In the case of the P9 single vcpu per vcore case, the real
932 * mode handler is not called but no other threads are in the
935 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
936 spin_lock(&vcore->lock);
937 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
938 vcore->vcore_state != VCORE_INACTIVE &&
940 target = vcore->runner;
941 spin_unlock(&vcore->lock);
944 return kvm_vcpu_yield_to(target);
947 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
950 struct lppaca *lppaca;
952 spin_lock(&vcpu->arch.vpa_update_lock);
953 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
955 yield_count = be32_to_cpu(lppaca->yield_count);
956 spin_unlock(&vcpu->arch.vpa_update_lock);
961 * H_RPT_INVALIDATE hcall handler for nested guests.
963 * Handles only nested process-scoped invalidation requests in L0.
965 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
967 unsigned long type = kvmppc_get_gpr(vcpu, 6);
968 unsigned long pid, pg_sizes, start, end;
971 * The partition-scoped invalidations aren't handled here in L0.
973 if (type & H_RPTI_TYPE_NESTED)
976 pid = kvmppc_get_gpr(vcpu, 4);
977 pg_sizes = kvmppc_get_gpr(vcpu, 7);
978 start = kvmppc_get_gpr(vcpu, 8);
979 end = kvmppc_get_gpr(vcpu, 9);
981 do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
982 type, pg_sizes, start, end);
984 kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
988 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
989 unsigned long id, unsigned long target,
990 unsigned long type, unsigned long pg_sizes,
991 unsigned long start, unsigned long end)
993 if (!kvm_is_radix(vcpu->kvm))
994 return H_UNSUPPORTED;
1000 * Partition-scoped invalidation for nested guests.
1002 if (type & H_RPTI_TYPE_NESTED) {
1003 if (!nesting_enabled(vcpu->kvm))
1006 /* Support only cores as target */
1007 if (target != H_RPTI_TARGET_CMMU)
1010 return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1015 * Process-scoped invalidation for L1 guests.
1017 do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1018 type, pg_sizes, start, end);
1022 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1024 struct kvm *kvm = vcpu->kvm;
1025 unsigned long req = kvmppc_get_gpr(vcpu, 3);
1026 unsigned long target, ret = H_SUCCESS;
1028 struct kvm_vcpu *tvcpu;
1031 if (req <= MAX_HCALL_OPCODE &&
1032 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1037 ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1038 kvmppc_get_gpr(vcpu, 5),
1039 kvmppc_get_gpr(vcpu, 6));
1040 if (ret == H_TOO_HARD)
1044 ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1045 kvmppc_get_gpr(vcpu, 5),
1046 kvmppc_get_gpr(vcpu, 6),
1047 kvmppc_get_gpr(vcpu, 7));
1048 if (ret == H_TOO_HARD)
1052 ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1053 kvmppc_get_gpr(vcpu, 5));
1054 if (ret == H_TOO_HARD)
1058 ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1059 kvmppc_get_gpr(vcpu, 5));
1060 if (ret == H_TOO_HARD)
1064 ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1065 kvmppc_get_gpr(vcpu, 5));
1066 if (ret == H_TOO_HARD)
1070 ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1071 kvmppc_get_gpr(vcpu, 5),
1072 kvmppc_get_gpr(vcpu, 6));
1073 if (ret == H_TOO_HARD)
1077 ret = kvmppc_h_bulk_remove(vcpu);
1078 if (ret == H_TOO_HARD)
1085 target = kvmppc_get_gpr(vcpu, 4);
1086 tvcpu = kvmppc_find_vcpu(kvm, target);
1091 tvcpu->arch.prodded = 1;
1093 if (tvcpu->arch.ceded)
1094 kvmppc_fast_vcpu_kick_hv(tvcpu);
1097 target = kvmppc_get_gpr(vcpu, 4);
1100 tvcpu = kvmppc_find_vcpu(kvm, target);
1105 yield_count = kvmppc_get_gpr(vcpu, 5);
1106 if (kvmppc_get_yield_count(tvcpu) != yield_count)
1108 kvm_arch_vcpu_yield_to(tvcpu);
1110 case H_REGISTER_VPA:
1111 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1112 kvmppc_get_gpr(vcpu, 5),
1113 kvmppc_get_gpr(vcpu, 6));
1116 if (list_empty(&kvm->arch.rtas_tokens))
1119 idx = srcu_read_lock(&kvm->srcu);
1120 rc = kvmppc_rtas_hcall(vcpu);
1121 srcu_read_unlock(&kvm->srcu, idx);
1128 /* Send the error out to userspace via KVM_RUN */
1130 case H_LOGICAL_CI_LOAD:
1131 ret = kvmppc_h_logical_ci_load(vcpu);
1132 if (ret == H_TOO_HARD)
1135 case H_LOGICAL_CI_STORE:
1136 ret = kvmppc_h_logical_ci_store(vcpu);
1137 if (ret == H_TOO_HARD)
1141 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1142 kvmppc_get_gpr(vcpu, 5),
1143 kvmppc_get_gpr(vcpu, 6),
1144 kvmppc_get_gpr(vcpu, 7));
1145 if (ret == H_TOO_HARD)
1154 if (kvmppc_xics_enabled(vcpu)) {
1155 if (xics_on_xive()) {
1156 ret = H_NOT_AVAILABLE;
1157 return RESUME_GUEST;
1159 ret = kvmppc_xics_hcall(vcpu, req);
1164 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1167 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1168 kvmppc_get_gpr(vcpu, 5));
1170 #ifdef CONFIG_SPAPR_TCE_IOMMU
1172 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1173 kvmppc_get_gpr(vcpu, 5));
1174 if (ret == H_TOO_HARD)
1178 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1179 kvmppc_get_gpr(vcpu, 5),
1180 kvmppc_get_gpr(vcpu, 6));
1181 if (ret == H_TOO_HARD)
1184 case H_PUT_TCE_INDIRECT:
1185 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1186 kvmppc_get_gpr(vcpu, 5),
1187 kvmppc_get_gpr(vcpu, 6),
1188 kvmppc_get_gpr(vcpu, 7));
1189 if (ret == H_TOO_HARD)
1193 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1194 kvmppc_get_gpr(vcpu, 5),
1195 kvmppc_get_gpr(vcpu, 6),
1196 kvmppc_get_gpr(vcpu, 7));
1197 if (ret == H_TOO_HARD)
1202 if (!arch_get_random_seed_long(&vcpu->arch.regs.gpr[4]))
1205 case H_RPT_INVALIDATE:
1206 ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1207 kvmppc_get_gpr(vcpu, 5),
1208 kvmppc_get_gpr(vcpu, 6),
1209 kvmppc_get_gpr(vcpu, 7),
1210 kvmppc_get_gpr(vcpu, 8),
1211 kvmppc_get_gpr(vcpu, 9));
1214 case H_SET_PARTITION_TABLE:
1216 if (nesting_enabled(kvm))
1217 ret = kvmhv_set_partition_table(vcpu);
1219 case H_ENTER_NESTED:
1221 if (!nesting_enabled(kvm))
1223 ret = kvmhv_enter_nested_guest(vcpu);
1224 if (ret == H_INTERRUPT) {
1225 kvmppc_set_gpr(vcpu, 3, 0);
1226 vcpu->arch.hcall_needed = 0;
1228 } else if (ret == H_TOO_HARD) {
1229 kvmppc_set_gpr(vcpu, 3, 0);
1230 vcpu->arch.hcall_needed = 0;
1234 case H_TLB_INVALIDATE:
1236 if (nesting_enabled(kvm))
1237 ret = kvmhv_do_nested_tlbie(vcpu);
1239 case H_COPY_TOFROM_GUEST:
1241 if (nesting_enabled(kvm))
1242 ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1245 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1246 kvmppc_get_gpr(vcpu, 5),
1247 kvmppc_get_gpr(vcpu, 6));
1250 ret = H_UNSUPPORTED;
1251 if (kvmppc_get_srr1(vcpu) & MSR_S)
1252 ret = kvmppc_h_svm_page_in(kvm,
1253 kvmppc_get_gpr(vcpu, 4),
1254 kvmppc_get_gpr(vcpu, 5),
1255 kvmppc_get_gpr(vcpu, 6));
1257 case H_SVM_PAGE_OUT:
1258 ret = H_UNSUPPORTED;
1259 if (kvmppc_get_srr1(vcpu) & MSR_S)
1260 ret = kvmppc_h_svm_page_out(kvm,
1261 kvmppc_get_gpr(vcpu, 4),
1262 kvmppc_get_gpr(vcpu, 5),
1263 kvmppc_get_gpr(vcpu, 6));
1265 case H_SVM_INIT_START:
1266 ret = H_UNSUPPORTED;
1267 if (kvmppc_get_srr1(vcpu) & MSR_S)
1268 ret = kvmppc_h_svm_init_start(kvm);
1270 case H_SVM_INIT_DONE:
1271 ret = H_UNSUPPORTED;
1272 if (kvmppc_get_srr1(vcpu) & MSR_S)
1273 ret = kvmppc_h_svm_init_done(kvm);
1275 case H_SVM_INIT_ABORT:
1277 * Even if that call is made by the Ultravisor, the SSR1 value
1278 * is the guest context one, with the secure bit clear as it has
1279 * not yet been secured. So we can't check it here.
1280 * Instead the kvm->arch.secure_guest flag is checked inside
1281 * kvmppc_h_svm_init_abort().
1283 ret = kvmppc_h_svm_init_abort(kvm);
1289 WARN_ON_ONCE(ret == H_TOO_HARD);
1290 kvmppc_set_gpr(vcpu, 3, ret);
1291 vcpu->arch.hcall_needed = 0;
1292 return RESUME_GUEST;
1296 * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1297 * handlers in book3s_hv_rmhandlers.S.
1299 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1300 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1302 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1304 vcpu->arch.shregs.msr |= MSR_EE;
1305 vcpu->arch.ceded = 1;
1307 if (vcpu->arch.prodded) {
1308 vcpu->arch.prodded = 0;
1310 vcpu->arch.ceded = 0;
1314 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1320 case H_REGISTER_VPA:
1322 case H_LOGICAL_CI_LOAD:
1323 case H_LOGICAL_CI_STORE:
1324 #ifdef CONFIG_KVM_XICS
1333 case H_RPT_INVALIDATE:
1337 /* See if it's in the real-mode table */
1338 return kvmppc_hcall_impl_hv_realmode(cmd);
1341 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1345 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1348 * Fetch failed, so return to guest and
1349 * try executing it again.
1351 return RESUME_GUEST;
1354 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1355 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1356 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1359 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1360 return RESUME_GUEST;
1364 static void do_nothing(void *x)
1368 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1370 int thr, cpu, pcpu, nthreads;
1372 unsigned long dpdes;
1374 nthreads = vcpu->kvm->arch.emul_smt_mode;
1376 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1377 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1378 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1382 * If the vcpu is currently running on a physical cpu thread,
1383 * interrupt it in order to pull it out of the guest briefly,
1384 * which will update its vcore->dpdes value.
1386 pcpu = READ_ONCE(v->cpu);
1388 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1389 if (kvmppc_doorbell_pending(v))
1396 * On POWER9, emulate doorbell-related instructions in order to
1397 * give the guest the illusion of running on a multi-threaded core.
1398 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1401 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1405 struct kvm *kvm = vcpu->kvm;
1406 struct kvm_vcpu *tvcpu;
1408 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1409 return RESUME_GUEST;
1410 if (get_op(inst) != 31)
1411 return EMULATE_FAIL;
1413 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1414 switch (get_xop(inst)) {
1415 case OP_31_XOP_MSGSNDP:
1416 arg = kvmppc_get_gpr(vcpu, rb);
1417 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1420 if (arg >= kvm->arch.emul_smt_mode)
1422 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1425 if (!tvcpu->arch.doorbell_request) {
1426 tvcpu->arch.doorbell_request = 1;
1427 kvmppc_fast_vcpu_kick_hv(tvcpu);
1430 case OP_31_XOP_MSGCLRP:
1431 arg = kvmppc_get_gpr(vcpu, rb);
1432 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1434 vcpu->arch.vcore->dpdes = 0;
1435 vcpu->arch.doorbell_request = 0;
1437 case OP_31_XOP_MFSPR:
1438 switch (get_sprn(inst)) {
1443 arg = kvmppc_read_dpdes(vcpu);
1446 return EMULATE_FAIL;
1448 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1451 return EMULATE_FAIL;
1453 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1454 return RESUME_GUEST;
1458 * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1459 * HFSCR_PM is cleared for next entry. If the guest then tries to access
1460 * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1461 * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1462 * allow the guest access to continue.
1464 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1466 if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1467 return EMULATE_FAIL;
1469 vcpu->arch.hfscr |= HFSCR_PM;
1471 return RESUME_GUEST;
1474 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1476 if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1477 return EMULATE_FAIL;
1479 vcpu->arch.hfscr |= HFSCR_EBB;
1481 return RESUME_GUEST;
1484 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1486 if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1487 return EMULATE_FAIL;
1489 vcpu->arch.hfscr |= HFSCR_TM;
1491 return RESUME_GUEST;
1494 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1495 struct task_struct *tsk)
1497 struct kvm_run *run = vcpu->run;
1498 int r = RESUME_HOST;
1500 vcpu->stat.sum_exits++;
1503 * This can happen if an interrupt occurs in the last stages
1504 * of guest entry or the first stages of guest exit (i.e. after
1505 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1506 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1507 * That can happen due to a bug, or due to a machine check
1508 * occurring at just the wrong time.
1510 if (vcpu->arch.shregs.msr & MSR_HV) {
1511 printk(KERN_EMERG "KVM trap in HV mode!\n");
1512 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1513 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1514 vcpu->arch.shregs.msr);
1515 kvmppc_dump_regs(vcpu);
1516 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1517 run->hw.hardware_exit_reason = vcpu->arch.trap;
1520 run->exit_reason = KVM_EXIT_UNKNOWN;
1521 run->ready_for_interrupt_injection = 1;
1522 switch (vcpu->arch.trap) {
1523 /* We're good on these - the host merely wanted to get our attention */
1524 case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1525 WARN_ON_ONCE(1); /* Should never happen */
1526 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1528 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1529 vcpu->stat.dec_exits++;
1532 case BOOK3S_INTERRUPT_EXTERNAL:
1533 case BOOK3S_INTERRUPT_H_DOORBELL:
1534 case BOOK3S_INTERRUPT_H_VIRT:
1535 vcpu->stat.ext_intr_exits++;
1538 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1539 case BOOK3S_INTERRUPT_HMI:
1540 case BOOK3S_INTERRUPT_PERFMON:
1541 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1544 case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1545 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1546 DEFAULT_RATELIMIT_BURST);
1548 * Print the MCE event to host console. Ratelimit so the guest
1549 * can't flood the host log.
1551 if (__ratelimit(&rs))
1552 machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1555 * If the guest can do FWNMI, exit to userspace so it can
1556 * deliver a FWNMI to the guest.
1557 * Otherwise we synthesize a machine check for the guest
1558 * so that it knows that the machine check occurred.
1560 if (!vcpu->kvm->arch.fwnmi_enabled) {
1561 ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1562 kvmppc_core_queue_machine_check(vcpu, flags);
1567 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1568 run->exit_reason = KVM_EXIT_NMI;
1569 run->hw.hardware_exit_reason = vcpu->arch.trap;
1570 /* Clear out the old NMI status from run->flags */
1571 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1572 /* Now set the NMI status */
1573 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1574 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1576 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1581 case BOOK3S_INTERRUPT_PROGRAM:
1585 * Normally program interrupts are delivered directly
1586 * to the guest by the hardware, but we can get here
1587 * as a result of a hypervisor emulation interrupt
1588 * (e40) getting turned into a 700 by BML RTAS.
1590 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1591 kvmppc_core_queue_program(vcpu, flags);
1595 case BOOK3S_INTERRUPT_SYSCALL:
1599 if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1601 * Guest userspace executed sc 1. This can only be
1602 * reached by the P9 path because the old path
1603 * handles this case in realmode hcall handlers.
1605 if (!kvmhv_vcpu_is_radix(vcpu)) {
1607 * A guest could be running PR KVM, so this
1608 * may be a PR KVM hcall. It must be reflected
1609 * to the guest kernel as a sc interrupt.
1611 kvmppc_core_queue_syscall(vcpu);
1614 * Radix guests can not run PR KVM or nested HV
1615 * hash guests which might run PR KVM, so this
1616 * is always a privilege fault. Send a program
1617 * check to guest kernel.
1619 kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1626 * hcall - gather args and set exit_reason. This will next be
1627 * handled by kvmppc_pseries_do_hcall which may be able to deal
1628 * with it and resume guest, or may punt to userspace.
1630 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1631 for (i = 0; i < 9; ++i)
1632 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1633 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1634 vcpu->arch.hcall_needed = 1;
1639 * We get these next two if the guest accesses a page which it thinks
1640 * it has mapped but which is not actually present, either because
1641 * it is for an emulated I/O device or because the corresonding
1642 * host page has been paged out.
1644 * Any other HDSI/HISI interrupts have been handled already for P7/8
1645 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1646 * fault handling is done here.
1648 case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1652 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1653 unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1654 r = RESUME_GUEST; /* Just retry if it's the canary */
1658 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1660 * Radix doesn't require anything, and pre-ISAv3.0 hash
1661 * already attempted to handle this in rmhandlers. The
1662 * hash fault handling below is v3 only (it uses ASDR
1665 r = RESUME_PAGE_FAULT;
1669 if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1670 kvmppc_core_queue_data_storage(vcpu,
1671 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1676 if (!(vcpu->arch.shregs.msr & MSR_DR))
1677 vsid = vcpu->kvm->arch.vrma_slb_v;
1679 vsid = vcpu->arch.fault_gpa;
1681 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1682 vsid, vcpu->arch.fault_dsisr, true);
1685 } else if (err == -1 || err == -2) {
1686 r = RESUME_PAGE_FAULT;
1688 kvmppc_core_queue_data_storage(vcpu,
1689 vcpu->arch.fault_dar, err);
1694 case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1698 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1699 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1700 DSISR_SRR1_MATCH_64S;
1701 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1703 * Radix doesn't require anything, and pre-ISAv3.0 hash
1704 * already attempted to handle this in rmhandlers. The
1705 * hash fault handling below is v3 only (it uses ASDR
1708 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1709 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1710 r = RESUME_PAGE_FAULT;
1714 if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1715 kvmppc_core_queue_inst_storage(vcpu,
1716 vcpu->arch.fault_dsisr);
1721 if (!(vcpu->arch.shregs.msr & MSR_IR))
1722 vsid = vcpu->kvm->arch.vrma_slb_v;
1724 vsid = vcpu->arch.fault_gpa;
1726 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1727 vsid, vcpu->arch.fault_dsisr, false);
1730 } else if (err == -1) {
1731 r = RESUME_PAGE_FAULT;
1733 kvmppc_core_queue_inst_storage(vcpu, err);
1740 * This occurs if the guest executes an illegal instruction.
1741 * If the guest debug is disabled, generate a program interrupt
1742 * to the guest. If guest debug is enabled, we need to check
1743 * whether the instruction is a software breakpoint instruction.
1744 * Accordingly return to Guest or Host.
1746 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1747 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1748 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1749 swab32(vcpu->arch.emul_inst) :
1750 vcpu->arch.emul_inst;
1751 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1752 r = kvmppc_emulate_debug_inst(vcpu);
1754 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1759 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1760 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1762 * This occurs for various TM-related instructions that
1763 * we need to emulate on POWER9 DD2.2. We have already
1764 * handled the cases where the guest was in real-suspend
1765 * mode and was transitioning to transactional state.
1767 r = kvmhv_p9_tm_emulation(vcpu);
1770 fallthrough; /* go to facility unavailable handler */
1774 * This occurs if the guest (kernel or userspace), does something that
1775 * is prohibited by HFSCR.
1776 * On POWER9, this could be a doorbell instruction that we need
1778 * Otherwise, we just generate a program interrupt to the guest.
1780 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1781 u64 cause = vcpu->arch.hfscr >> 56;
1784 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1785 if (cause == FSCR_MSGP_LG)
1786 r = kvmppc_emulate_doorbell_instr(vcpu);
1787 if (cause == FSCR_PM_LG)
1788 r = kvmppc_pmu_unavailable(vcpu);
1789 if (cause == FSCR_EBB_LG)
1790 r = kvmppc_ebb_unavailable(vcpu);
1791 if (cause == FSCR_TM_LG)
1792 r = kvmppc_tm_unavailable(vcpu);
1794 if (r == EMULATE_FAIL) {
1795 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1801 case BOOK3S_INTERRUPT_HV_RM_HARD:
1802 r = RESUME_PASSTHROUGH;
1805 kvmppc_dump_regs(vcpu);
1806 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1807 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1808 vcpu->arch.shregs.msr);
1809 run->hw.hardware_exit_reason = vcpu->arch.trap;
1817 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1819 struct kvm_nested_guest *nested = vcpu->arch.nested;
1823 vcpu->stat.sum_exits++;
1826 * This can happen if an interrupt occurs in the last stages
1827 * of guest entry or the first stages of guest exit (i.e. after
1828 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1829 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1830 * That can happen due to a bug, or due to a machine check
1831 * occurring at just the wrong time.
1833 if (vcpu->arch.shregs.msr & MSR_HV) {
1834 pr_emerg("KVM trap in HV mode while nested!\n");
1835 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1836 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1837 vcpu->arch.shregs.msr);
1838 kvmppc_dump_regs(vcpu);
1841 switch (vcpu->arch.trap) {
1842 /* We're good on these - the host merely wanted to get our attention */
1843 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1844 vcpu->stat.dec_exits++;
1847 case BOOK3S_INTERRUPT_EXTERNAL:
1848 vcpu->stat.ext_intr_exits++;
1851 case BOOK3S_INTERRUPT_H_DOORBELL:
1852 case BOOK3S_INTERRUPT_H_VIRT:
1853 vcpu->stat.ext_intr_exits++;
1856 /* These need to go to the nested HV */
1857 case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1858 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1859 vcpu->stat.dec_exits++;
1862 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1863 case BOOK3S_INTERRUPT_HMI:
1864 case BOOK3S_INTERRUPT_PERFMON:
1865 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1868 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1870 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1871 DEFAULT_RATELIMIT_BURST);
1872 /* Pass the machine check to the L1 guest */
1874 /* Print the MCE event to host console. */
1875 if (__ratelimit(&rs))
1876 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1880 * We get these next two if the guest accesses a page which it thinks
1881 * it has mapped but which is not actually present, either because
1882 * it is for an emulated I/O device or because the corresonding
1883 * host page has been paged out.
1885 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1886 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1887 r = kvmhv_nested_page_fault(vcpu);
1888 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1890 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1891 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1892 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1893 DSISR_SRR1_MATCH_64S;
1894 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1895 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1896 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1897 r = kvmhv_nested_page_fault(vcpu);
1898 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1901 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1902 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1904 * This occurs for various TM-related instructions that
1905 * we need to emulate on POWER9 DD2.2. We have already
1906 * handled the cases where the guest was in real-suspend
1907 * mode and was transitioning to transactional state.
1909 r = kvmhv_p9_tm_emulation(vcpu);
1912 fallthrough; /* go to facility unavailable handler */
1915 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1916 u64 cause = vcpu->arch.hfscr >> 56;
1919 * Only pass HFU interrupts to the L1 if the facility is
1920 * permitted but disabled by the L1's HFSCR, otherwise
1921 * the interrupt does not make sense to the L1 so turn
1924 if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) ||
1925 (nested->hfscr & (1UL << cause))) {
1926 vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST;
1929 * If the fetch failed, return to guest and
1930 * try executing it again.
1932 r = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1933 &vcpu->arch.emul_inst);
1934 if (r != EMULATE_DONE)
1945 case BOOK3S_INTERRUPT_HV_RM_HARD:
1946 vcpu->arch.trap = 0;
1948 if (!xics_on_xive())
1949 kvmppc_xics_rm_complete(vcpu, 0);
1951 case BOOK3S_INTERRUPT_SYSCALL:
1953 unsigned long req = kvmppc_get_gpr(vcpu, 3);
1956 * The H_RPT_INVALIDATE hcalls issued by nested
1957 * guests for process-scoped invalidations when
1958 * GTSE=0, are handled here in L0.
1960 if (req == H_RPT_INVALIDATE) {
1961 r = kvmppc_nested_h_rpt_invalidate(vcpu);
1976 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1977 struct kvm_sregs *sregs)
1981 memset(sregs, 0, sizeof(struct kvm_sregs));
1982 sregs->pvr = vcpu->arch.pvr;
1983 for (i = 0; i < vcpu->arch.slb_max; i++) {
1984 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1985 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1991 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1992 struct kvm_sregs *sregs)
1996 /* Only accept the same PVR as the host's, since we can't spoof it */
1997 if (sregs->pvr != vcpu->arch.pvr)
2001 for (i = 0; i < vcpu->arch.slb_nr; i++) {
2002 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2003 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2004 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2008 vcpu->arch.slb_max = j;
2014 * Enforce limits on guest LPCR values based on hardware availability,
2015 * guest configuration, and possibly hypervisor support and security
2018 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2020 /* LPCR_TC only applies to HPT guests */
2021 if (kvm_is_radix(kvm))
2024 /* On POWER8 and above, userspace can modify AIL */
2025 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2027 if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2028 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2030 * On some POWER9s we force AIL off for radix guests to prevent
2031 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2032 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2033 * be cached, which the host TLB management does not expect.
2035 if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2039 * On POWER9, allow userspace to enable large decrementer for the
2040 * guest, whether or not the host has it enabled.
2042 if (!cpu_has_feature(CPU_FTR_ARCH_300))
2048 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2050 if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2051 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2052 lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2056 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2057 bool preserve_top32)
2059 struct kvm *kvm = vcpu->kvm;
2060 struct kvmppc_vcore *vc = vcpu->arch.vcore;
2063 spin_lock(&vc->lock);
2066 * Userspace can only modify
2067 * DPFD (default prefetch depth), ILE (interrupt little-endian),
2068 * TC (translation control), AIL (alternate interrupt location),
2069 * LD (large decrementer).
2070 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2072 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2074 /* Broken 32-bit version of LPCR must not clear top bits */
2078 new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2079 (vc->lpcr & ~mask) | (new_lpcr & mask));
2082 * If ILE (interrupt little-endian) has changed, update the
2083 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2085 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2086 struct kvm_vcpu *vcpu;
2089 kvm_for_each_vcpu(i, vcpu, kvm) {
2090 if (vcpu->arch.vcore != vc)
2092 if (new_lpcr & LPCR_ILE)
2093 vcpu->arch.intr_msr |= MSR_LE;
2095 vcpu->arch.intr_msr &= ~MSR_LE;
2099 vc->lpcr = new_lpcr;
2101 spin_unlock(&vc->lock);
2104 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2105 union kvmppc_one_reg *val)
2111 case KVM_REG_PPC_DEBUG_INST:
2112 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2114 case KVM_REG_PPC_HIOR:
2115 *val = get_reg_val(id, 0);
2117 case KVM_REG_PPC_DABR:
2118 *val = get_reg_val(id, vcpu->arch.dabr);
2120 case KVM_REG_PPC_DABRX:
2121 *val = get_reg_val(id, vcpu->arch.dabrx);
2123 case KVM_REG_PPC_DSCR:
2124 *val = get_reg_val(id, vcpu->arch.dscr);
2126 case KVM_REG_PPC_PURR:
2127 *val = get_reg_val(id, vcpu->arch.purr);
2129 case KVM_REG_PPC_SPURR:
2130 *val = get_reg_val(id, vcpu->arch.spurr);
2132 case KVM_REG_PPC_AMR:
2133 *val = get_reg_val(id, vcpu->arch.amr);
2135 case KVM_REG_PPC_UAMOR:
2136 *val = get_reg_val(id, vcpu->arch.uamor);
2138 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2139 i = id - KVM_REG_PPC_MMCR0;
2140 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
2142 case KVM_REG_PPC_MMCR2:
2143 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
2145 case KVM_REG_PPC_MMCRA:
2146 *val = get_reg_val(id, vcpu->arch.mmcra);
2148 case KVM_REG_PPC_MMCRS:
2149 *val = get_reg_val(id, vcpu->arch.mmcrs);
2151 case KVM_REG_PPC_MMCR3:
2152 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2154 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2155 i = id - KVM_REG_PPC_PMC1;
2156 *val = get_reg_val(id, vcpu->arch.pmc[i]);
2158 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2159 i = id - KVM_REG_PPC_SPMC1;
2160 *val = get_reg_val(id, vcpu->arch.spmc[i]);
2162 case KVM_REG_PPC_SIAR:
2163 *val = get_reg_val(id, vcpu->arch.siar);
2165 case KVM_REG_PPC_SDAR:
2166 *val = get_reg_val(id, vcpu->arch.sdar);
2168 case KVM_REG_PPC_SIER:
2169 *val = get_reg_val(id, vcpu->arch.sier[0]);
2171 case KVM_REG_PPC_SIER2:
2172 *val = get_reg_val(id, vcpu->arch.sier[1]);
2174 case KVM_REG_PPC_SIER3:
2175 *val = get_reg_val(id, vcpu->arch.sier[2]);
2177 case KVM_REG_PPC_IAMR:
2178 *val = get_reg_val(id, vcpu->arch.iamr);
2180 case KVM_REG_PPC_PSPB:
2181 *val = get_reg_val(id, vcpu->arch.pspb);
2183 case KVM_REG_PPC_DPDES:
2185 * On POWER9, where we are emulating msgsndp etc.,
2186 * we return 1 bit for each vcpu, which can come from
2187 * either vcore->dpdes or doorbell_request.
2188 * On POWER8, doorbell_request is 0.
2190 if (cpu_has_feature(CPU_FTR_ARCH_300))
2191 *val = get_reg_val(id, vcpu->arch.doorbell_request);
2193 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2195 case KVM_REG_PPC_VTB:
2196 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
2198 case KVM_REG_PPC_DAWR:
2199 *val = get_reg_val(id, vcpu->arch.dawr0);
2201 case KVM_REG_PPC_DAWRX:
2202 *val = get_reg_val(id, vcpu->arch.dawrx0);
2204 case KVM_REG_PPC_DAWR1:
2205 *val = get_reg_val(id, vcpu->arch.dawr1);
2207 case KVM_REG_PPC_DAWRX1:
2208 *val = get_reg_val(id, vcpu->arch.dawrx1);
2210 case KVM_REG_PPC_CIABR:
2211 *val = get_reg_val(id, vcpu->arch.ciabr);
2213 case KVM_REG_PPC_CSIGR:
2214 *val = get_reg_val(id, vcpu->arch.csigr);
2216 case KVM_REG_PPC_TACR:
2217 *val = get_reg_val(id, vcpu->arch.tacr);
2219 case KVM_REG_PPC_TCSCR:
2220 *val = get_reg_val(id, vcpu->arch.tcscr);
2222 case KVM_REG_PPC_PID:
2223 *val = get_reg_val(id, vcpu->arch.pid);
2225 case KVM_REG_PPC_ACOP:
2226 *val = get_reg_val(id, vcpu->arch.acop);
2228 case KVM_REG_PPC_WORT:
2229 *val = get_reg_val(id, vcpu->arch.wort);
2231 case KVM_REG_PPC_TIDR:
2232 *val = get_reg_val(id, vcpu->arch.tid);
2234 case KVM_REG_PPC_PSSCR:
2235 *val = get_reg_val(id, vcpu->arch.psscr);
2237 case KVM_REG_PPC_VPA_ADDR:
2238 spin_lock(&vcpu->arch.vpa_update_lock);
2239 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2240 spin_unlock(&vcpu->arch.vpa_update_lock);
2242 case KVM_REG_PPC_VPA_SLB:
2243 spin_lock(&vcpu->arch.vpa_update_lock);
2244 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2245 val->vpaval.length = vcpu->arch.slb_shadow.len;
2246 spin_unlock(&vcpu->arch.vpa_update_lock);
2248 case KVM_REG_PPC_VPA_DTL:
2249 spin_lock(&vcpu->arch.vpa_update_lock);
2250 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2251 val->vpaval.length = vcpu->arch.dtl.len;
2252 spin_unlock(&vcpu->arch.vpa_update_lock);
2254 case KVM_REG_PPC_TB_OFFSET:
2255 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
2257 case KVM_REG_PPC_LPCR:
2258 case KVM_REG_PPC_LPCR_64:
2259 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
2261 case KVM_REG_PPC_PPR:
2262 *val = get_reg_val(id, vcpu->arch.ppr);
2264 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2265 case KVM_REG_PPC_TFHAR:
2266 *val = get_reg_val(id, vcpu->arch.tfhar);
2268 case KVM_REG_PPC_TFIAR:
2269 *val = get_reg_val(id, vcpu->arch.tfiar);
2271 case KVM_REG_PPC_TEXASR:
2272 *val = get_reg_val(id, vcpu->arch.texasr);
2274 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2275 i = id - KVM_REG_PPC_TM_GPR0;
2276 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2278 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2281 i = id - KVM_REG_PPC_TM_VSR0;
2283 for (j = 0; j < TS_FPRWIDTH; j++)
2284 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2286 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2287 val->vval = vcpu->arch.vr_tm.vr[i-32];
2293 case KVM_REG_PPC_TM_CR:
2294 *val = get_reg_val(id, vcpu->arch.cr_tm);
2296 case KVM_REG_PPC_TM_XER:
2297 *val = get_reg_val(id, vcpu->arch.xer_tm);
2299 case KVM_REG_PPC_TM_LR:
2300 *val = get_reg_val(id, vcpu->arch.lr_tm);
2302 case KVM_REG_PPC_TM_CTR:
2303 *val = get_reg_val(id, vcpu->arch.ctr_tm);
2305 case KVM_REG_PPC_TM_FPSCR:
2306 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2308 case KVM_REG_PPC_TM_AMR:
2309 *val = get_reg_val(id, vcpu->arch.amr_tm);
2311 case KVM_REG_PPC_TM_PPR:
2312 *val = get_reg_val(id, vcpu->arch.ppr_tm);
2314 case KVM_REG_PPC_TM_VRSAVE:
2315 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
2317 case KVM_REG_PPC_TM_VSCR:
2318 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2319 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2323 case KVM_REG_PPC_TM_DSCR:
2324 *val = get_reg_val(id, vcpu->arch.dscr_tm);
2326 case KVM_REG_PPC_TM_TAR:
2327 *val = get_reg_val(id, vcpu->arch.tar_tm);
2330 case KVM_REG_PPC_ARCH_COMPAT:
2331 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
2333 case KVM_REG_PPC_DEC_EXPIRY:
2334 *val = get_reg_val(id, vcpu->arch.dec_expires);
2336 case KVM_REG_PPC_ONLINE:
2337 *val = get_reg_val(id, vcpu->arch.online);
2339 case KVM_REG_PPC_PTCR:
2340 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2350 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2351 union kvmppc_one_reg *val)
2355 unsigned long addr, len;
2358 case KVM_REG_PPC_HIOR:
2359 /* Only allow this to be set to zero */
2360 if (set_reg_val(id, *val))
2363 case KVM_REG_PPC_DABR:
2364 vcpu->arch.dabr = set_reg_val(id, *val);
2366 case KVM_REG_PPC_DABRX:
2367 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2369 case KVM_REG_PPC_DSCR:
2370 vcpu->arch.dscr = set_reg_val(id, *val);
2372 case KVM_REG_PPC_PURR:
2373 vcpu->arch.purr = set_reg_val(id, *val);
2375 case KVM_REG_PPC_SPURR:
2376 vcpu->arch.spurr = set_reg_val(id, *val);
2378 case KVM_REG_PPC_AMR:
2379 vcpu->arch.amr = set_reg_val(id, *val);
2381 case KVM_REG_PPC_UAMOR:
2382 vcpu->arch.uamor = set_reg_val(id, *val);
2384 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2385 i = id - KVM_REG_PPC_MMCR0;
2386 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2388 case KVM_REG_PPC_MMCR2:
2389 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2391 case KVM_REG_PPC_MMCRA:
2392 vcpu->arch.mmcra = set_reg_val(id, *val);
2394 case KVM_REG_PPC_MMCRS:
2395 vcpu->arch.mmcrs = set_reg_val(id, *val);
2397 case KVM_REG_PPC_MMCR3:
2398 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2400 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2401 i = id - KVM_REG_PPC_PMC1;
2402 vcpu->arch.pmc[i] = set_reg_val(id, *val);
2404 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2405 i = id - KVM_REG_PPC_SPMC1;
2406 vcpu->arch.spmc[i] = set_reg_val(id, *val);
2408 case KVM_REG_PPC_SIAR:
2409 vcpu->arch.siar = set_reg_val(id, *val);
2411 case KVM_REG_PPC_SDAR:
2412 vcpu->arch.sdar = set_reg_val(id, *val);
2414 case KVM_REG_PPC_SIER:
2415 vcpu->arch.sier[0] = set_reg_val(id, *val);
2417 case KVM_REG_PPC_SIER2:
2418 vcpu->arch.sier[1] = set_reg_val(id, *val);
2420 case KVM_REG_PPC_SIER3:
2421 vcpu->arch.sier[2] = set_reg_val(id, *val);
2423 case KVM_REG_PPC_IAMR:
2424 vcpu->arch.iamr = set_reg_val(id, *val);
2426 case KVM_REG_PPC_PSPB:
2427 vcpu->arch.pspb = set_reg_val(id, *val);
2429 case KVM_REG_PPC_DPDES:
2430 if (cpu_has_feature(CPU_FTR_ARCH_300))
2431 vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2433 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2435 case KVM_REG_PPC_VTB:
2436 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2438 case KVM_REG_PPC_DAWR:
2439 vcpu->arch.dawr0 = set_reg_val(id, *val);
2441 case KVM_REG_PPC_DAWRX:
2442 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2444 case KVM_REG_PPC_DAWR1:
2445 vcpu->arch.dawr1 = set_reg_val(id, *val);
2447 case KVM_REG_PPC_DAWRX1:
2448 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2450 case KVM_REG_PPC_CIABR:
2451 vcpu->arch.ciabr = set_reg_val(id, *val);
2452 /* Don't allow setting breakpoints in hypervisor code */
2453 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2454 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
2456 case KVM_REG_PPC_CSIGR:
2457 vcpu->arch.csigr = set_reg_val(id, *val);
2459 case KVM_REG_PPC_TACR:
2460 vcpu->arch.tacr = set_reg_val(id, *val);
2462 case KVM_REG_PPC_TCSCR:
2463 vcpu->arch.tcscr = set_reg_val(id, *val);
2465 case KVM_REG_PPC_PID:
2466 vcpu->arch.pid = set_reg_val(id, *val);
2468 case KVM_REG_PPC_ACOP:
2469 vcpu->arch.acop = set_reg_val(id, *val);
2471 case KVM_REG_PPC_WORT:
2472 vcpu->arch.wort = set_reg_val(id, *val);
2474 case KVM_REG_PPC_TIDR:
2475 vcpu->arch.tid = set_reg_val(id, *val);
2477 case KVM_REG_PPC_PSSCR:
2478 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2480 case KVM_REG_PPC_VPA_ADDR:
2481 addr = set_reg_val(id, *val);
2483 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2484 vcpu->arch.dtl.next_gpa))
2486 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2488 case KVM_REG_PPC_VPA_SLB:
2489 addr = val->vpaval.addr;
2490 len = val->vpaval.length;
2492 if (addr && !vcpu->arch.vpa.next_gpa)
2494 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2496 case KVM_REG_PPC_VPA_DTL:
2497 addr = val->vpaval.addr;
2498 len = val->vpaval.length;
2500 if (addr && (len < sizeof(struct dtl_entry) ||
2501 !vcpu->arch.vpa.next_gpa))
2503 len -= len % sizeof(struct dtl_entry);
2504 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2506 case KVM_REG_PPC_TB_OFFSET:
2507 /* round up to multiple of 2^24 */
2508 vcpu->arch.vcore->tb_offset =
2509 ALIGN(set_reg_val(id, *val), 1UL << 24);
2511 case KVM_REG_PPC_LPCR:
2512 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2514 case KVM_REG_PPC_LPCR_64:
2515 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2517 case KVM_REG_PPC_PPR:
2518 vcpu->arch.ppr = set_reg_val(id, *val);
2520 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2521 case KVM_REG_PPC_TFHAR:
2522 vcpu->arch.tfhar = set_reg_val(id, *val);
2524 case KVM_REG_PPC_TFIAR:
2525 vcpu->arch.tfiar = set_reg_val(id, *val);
2527 case KVM_REG_PPC_TEXASR:
2528 vcpu->arch.texasr = set_reg_val(id, *val);
2530 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2531 i = id - KVM_REG_PPC_TM_GPR0;
2532 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2534 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2537 i = id - KVM_REG_PPC_TM_VSR0;
2539 for (j = 0; j < TS_FPRWIDTH; j++)
2540 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2542 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2543 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2548 case KVM_REG_PPC_TM_CR:
2549 vcpu->arch.cr_tm = set_reg_val(id, *val);
2551 case KVM_REG_PPC_TM_XER:
2552 vcpu->arch.xer_tm = set_reg_val(id, *val);
2554 case KVM_REG_PPC_TM_LR:
2555 vcpu->arch.lr_tm = set_reg_val(id, *val);
2557 case KVM_REG_PPC_TM_CTR:
2558 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2560 case KVM_REG_PPC_TM_FPSCR:
2561 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2563 case KVM_REG_PPC_TM_AMR:
2564 vcpu->arch.amr_tm = set_reg_val(id, *val);
2566 case KVM_REG_PPC_TM_PPR:
2567 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2569 case KVM_REG_PPC_TM_VRSAVE:
2570 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2572 case KVM_REG_PPC_TM_VSCR:
2573 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2574 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2578 case KVM_REG_PPC_TM_DSCR:
2579 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2581 case KVM_REG_PPC_TM_TAR:
2582 vcpu->arch.tar_tm = set_reg_val(id, *val);
2585 case KVM_REG_PPC_ARCH_COMPAT:
2586 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2588 case KVM_REG_PPC_DEC_EXPIRY:
2589 vcpu->arch.dec_expires = set_reg_val(id, *val);
2591 case KVM_REG_PPC_ONLINE:
2592 i = set_reg_val(id, *val);
2593 if (i && !vcpu->arch.online)
2594 atomic_inc(&vcpu->arch.vcore->online_count);
2595 else if (!i && vcpu->arch.online)
2596 atomic_dec(&vcpu->arch.vcore->online_count);
2597 vcpu->arch.online = i;
2599 case KVM_REG_PPC_PTCR:
2600 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2611 * On POWER9, threads are independent and can be in different partitions.
2612 * Therefore we consider each thread to be a subcore.
2613 * There is a restriction that all threads have to be in the same
2614 * MMU mode (radix or HPT), unfortunately, but since we only support
2615 * HPT guests on a HPT host so far, that isn't an impediment yet.
2617 static int threads_per_vcore(struct kvm *kvm)
2619 if (cpu_has_feature(CPU_FTR_ARCH_300))
2621 return threads_per_subcore;
2624 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2626 struct kvmppc_vcore *vcore;
2628 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2633 spin_lock_init(&vcore->lock);
2634 spin_lock_init(&vcore->stoltb_lock);
2635 rcuwait_init(&vcore->wait);
2636 vcore->preempt_tb = TB_NIL;
2637 vcore->lpcr = kvm->arch.lpcr;
2638 vcore->first_vcpuid = id;
2640 INIT_LIST_HEAD(&vcore->preempt_list);
2645 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2646 static struct debugfs_timings_element {
2650 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2651 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2652 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2653 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2654 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2657 #define N_TIMINGS (ARRAY_SIZE(timings))
2659 struct debugfs_timings_state {
2660 struct kvm_vcpu *vcpu;
2661 unsigned int buflen;
2662 char buf[N_TIMINGS * 100];
2665 static int debugfs_timings_open(struct inode *inode, struct file *file)
2667 struct kvm_vcpu *vcpu = inode->i_private;
2668 struct debugfs_timings_state *p;
2670 p = kzalloc(sizeof(*p), GFP_KERNEL);
2674 kvm_get_kvm(vcpu->kvm);
2676 file->private_data = p;
2678 return nonseekable_open(inode, file);
2681 static int debugfs_timings_release(struct inode *inode, struct file *file)
2683 struct debugfs_timings_state *p = file->private_data;
2685 kvm_put_kvm(p->vcpu->kvm);
2690 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2691 size_t len, loff_t *ppos)
2693 struct debugfs_timings_state *p = file->private_data;
2694 struct kvm_vcpu *vcpu = p->vcpu;
2696 struct kvmhv_tb_accumulator tb;
2705 buf_end = s + sizeof(p->buf);
2706 for (i = 0; i < N_TIMINGS; ++i) {
2707 struct kvmhv_tb_accumulator *acc;
2709 acc = (struct kvmhv_tb_accumulator *)
2710 ((unsigned long)vcpu + timings[i].offset);
2712 for (loops = 0; loops < 1000; ++loops) {
2713 count = acc->seqcount;
2718 if (count == acc->seqcount) {
2726 snprintf(s, buf_end - s, "%s: stuck\n",
2729 snprintf(s, buf_end - s,
2730 "%s: %llu %llu %llu %llu\n",
2731 timings[i].name, count / 2,
2732 tb_to_ns(tb.tb_total),
2733 tb_to_ns(tb.tb_min),
2734 tb_to_ns(tb.tb_max));
2737 p->buflen = s - p->buf;
2741 if (pos >= p->buflen)
2743 if (len > p->buflen - pos)
2744 len = p->buflen - pos;
2745 n = copy_to_user(buf, p->buf + pos, len);
2755 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2756 size_t len, loff_t *ppos)
2761 static const struct file_operations debugfs_timings_ops = {
2762 .owner = THIS_MODULE,
2763 .open = debugfs_timings_open,
2764 .release = debugfs_timings_release,
2765 .read = debugfs_timings_read,
2766 .write = debugfs_timings_write,
2767 .llseek = generic_file_llseek,
2770 /* Create a debugfs directory for the vcpu */
2771 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2774 struct kvm *kvm = vcpu->kvm;
2776 snprintf(buf, sizeof(buf), "vcpu%u", id);
2777 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2778 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2779 &debugfs_timings_ops);
2782 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2783 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2786 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2788 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2792 struct kvmppc_vcore *vcore;
2799 vcpu->arch.shared = &vcpu->arch.shregs;
2800 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2802 * The shared struct is never shared on HV,
2803 * so we can always use host endianness
2805 #ifdef __BIG_ENDIAN__
2806 vcpu->arch.shared_big_endian = true;
2808 vcpu->arch.shared_big_endian = false;
2811 vcpu->arch.mmcr[0] = MMCR0_FC;
2812 if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2813 vcpu->arch.mmcr[0] |= MMCR0_PMCCEXT;
2814 vcpu->arch.mmcra = MMCRA_BHRB_DISABLE;
2817 vcpu->arch.ctrl = CTRL_RUNLATCH;
2818 /* default to host PVR, since we can't spoof it */
2819 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2820 spin_lock_init(&vcpu->arch.vpa_update_lock);
2821 spin_lock_init(&vcpu->arch.tbacct_lock);
2822 vcpu->arch.busy_preempt = TB_NIL;
2823 vcpu->arch.shregs.msr = MSR_ME;
2824 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2827 * Set the default HFSCR for the guest from the host value.
2828 * This value is only used on POWER9.
2829 * On POWER9, we want to virtualize the doorbell facility, so we
2830 * don't set the HFSCR_MSGP bit, and that causes those instructions
2831 * to trap and then we emulate them.
2833 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2834 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2835 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2836 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2837 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2838 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2839 vcpu->arch.hfscr |= HFSCR_TM;
2842 if (cpu_has_feature(CPU_FTR_TM_COMP))
2843 vcpu->arch.hfscr |= HFSCR_TM;
2845 vcpu->arch.hfscr_permitted = vcpu->arch.hfscr;
2848 * PM, EBB, TM are demand-faulted so start with it clear.
2850 vcpu->arch.hfscr &= ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM);
2852 kvmppc_mmu_book3s_hv_init(vcpu);
2854 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2856 init_waitqueue_head(&vcpu->arch.cpu_run);
2858 mutex_lock(&kvm->lock);
2861 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2862 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2863 pr_devel("KVM: VCPU ID too high\n");
2864 core = KVM_MAX_VCORES;
2866 BUG_ON(kvm->arch.smt_mode != 1);
2867 core = kvmppc_pack_vcpu_id(kvm, id);
2870 core = id / kvm->arch.smt_mode;
2872 if (core < KVM_MAX_VCORES) {
2873 vcore = kvm->arch.vcores[core];
2874 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2875 pr_devel("KVM: collision on id %u", id);
2877 } else if (!vcore) {
2879 * Take mmu_setup_lock for mutual exclusion
2880 * with kvmppc_update_lpcr().
2883 vcore = kvmppc_vcore_create(kvm,
2884 id & ~(kvm->arch.smt_mode - 1));
2885 mutex_lock(&kvm->arch.mmu_setup_lock);
2886 kvm->arch.vcores[core] = vcore;
2887 kvm->arch.online_vcores++;
2888 mutex_unlock(&kvm->arch.mmu_setup_lock);
2891 mutex_unlock(&kvm->lock);
2896 spin_lock(&vcore->lock);
2897 ++vcore->num_threads;
2898 spin_unlock(&vcore->lock);
2899 vcpu->arch.vcore = vcore;
2900 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2901 vcpu->arch.thread_cpu = -1;
2902 vcpu->arch.prev_cpu = -1;
2904 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2905 kvmppc_sanity_check(vcpu);
2907 debugfs_vcpu_init(vcpu, id);
2912 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2913 unsigned long flags)
2920 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2922 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2924 * On POWER8 (or POWER7), the threading mode is "strict",
2925 * so we pack smt_mode vcpus per vcore.
2927 if (smt_mode > threads_per_subcore)
2931 * On POWER9, the threading mode is "loose",
2932 * so each vcpu gets its own vcore.
2937 mutex_lock(&kvm->lock);
2939 if (!kvm->arch.online_vcores) {
2940 kvm->arch.smt_mode = smt_mode;
2941 kvm->arch.emul_smt_mode = esmt;
2944 mutex_unlock(&kvm->lock);
2949 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2951 if (vpa->pinned_addr)
2952 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2956 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2958 spin_lock(&vcpu->arch.vpa_update_lock);
2959 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2960 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2961 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2962 spin_unlock(&vcpu->arch.vpa_update_lock);
2965 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2967 /* Indicate we want to get back into the guest */
2971 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2973 unsigned long dec_nsec, now;
2976 if (now > kvmppc_dec_expires_host_tb(vcpu)) {
2977 /* decrementer has already gone negative */
2978 kvmppc_core_queue_dec(vcpu);
2979 kvmppc_core_prepare_to_enter(vcpu);
2982 dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
2983 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2984 vcpu->arch.timer_running = 1;
2987 extern int __kvmppc_vcore_entry(void);
2989 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2990 struct kvm_vcpu *vcpu, u64 tb)
2994 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2996 spin_lock_irq(&vcpu->arch.tbacct_lock);
2998 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2999 vcpu->arch.stolen_logged;
3000 vcpu->arch.busy_preempt = now;
3001 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3002 spin_unlock_irq(&vcpu->arch.tbacct_lock);
3004 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3007 static int kvmppc_grab_hwthread(int cpu)
3009 struct paca_struct *tpaca;
3010 long timeout = 10000;
3012 tpaca = paca_ptrs[cpu];
3014 /* Ensure the thread won't go into the kernel if it wakes */
3015 tpaca->kvm_hstate.kvm_vcpu = NULL;
3016 tpaca->kvm_hstate.kvm_vcore = NULL;
3017 tpaca->kvm_hstate.napping = 0;
3019 tpaca->kvm_hstate.hwthread_req = 1;
3022 * If the thread is already executing in the kernel (e.g. handling
3023 * a stray interrupt), wait for it to get back to nap mode.
3024 * The smp_mb() is to ensure that our setting of hwthread_req
3025 * is visible before we look at hwthread_state, so if this
3026 * races with the code at system_reset_pSeries and the thread
3027 * misses our setting of hwthread_req, we are sure to see its
3028 * setting of hwthread_state, and vice versa.
3031 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3032 if (--timeout <= 0) {
3033 pr_err("KVM: couldn't grab cpu %d\n", cpu);
3041 static void kvmppc_release_hwthread(int cpu)
3043 struct paca_struct *tpaca;
3045 tpaca = paca_ptrs[cpu];
3046 tpaca->kvm_hstate.hwthread_req = 0;
3047 tpaca->kvm_hstate.kvm_vcpu = NULL;
3048 tpaca->kvm_hstate.kvm_vcore = NULL;
3049 tpaca->kvm_hstate.kvm_split_mode = NULL;
3052 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3054 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3056 struct kvm_nested_guest *nested = vcpu->arch.nested;
3057 cpumask_t *need_tlb_flush;
3061 need_tlb_flush = &nested->need_tlb_flush;
3063 need_tlb_flush = &kvm->arch.need_tlb_flush;
3065 cpu = cpu_first_tlb_thread_sibling(cpu);
3066 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3067 i += cpu_tlb_thread_sibling_step())
3068 cpumask_set_cpu(i, need_tlb_flush);
3071 * Make sure setting of bit in need_tlb_flush precedes testing of
3072 * cpu_in_guest. The matching barrier on the other side is hwsync
3073 * when switching to guest MMU mode, which happens between
3074 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3079 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3080 i += cpu_tlb_thread_sibling_step()) {
3081 struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3084 smp_call_function_single(i, do_nothing, NULL, 1);
3088 static void do_migrate_away_vcpu(void *arg)
3090 struct kvm_vcpu *vcpu = arg;
3091 struct kvm *kvm = vcpu->kvm;
3094 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3095 * ptesync sequence on the old CPU before migrating to a new one, in
3096 * case we interrupted the guest between a tlbie ; eieio ;
3097 * tlbsync; ptesync sequence.
3099 * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3101 if (kvm->arch.lpcr & LPCR_GTSE)
3102 asm volatile("eieio; tlbsync; ptesync");
3104 asm volatile("ptesync");
3107 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3109 struct kvm_nested_guest *nested = vcpu->arch.nested;
3110 struct kvm *kvm = vcpu->kvm;
3113 if (!cpu_has_feature(CPU_FTR_HVMODE))
3117 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3119 prev_cpu = vcpu->arch.prev_cpu;
3122 * With radix, the guest can do TLB invalidations itself,
3123 * and it could choose to use the local form (tlbiel) if
3124 * it is invalidating a translation that has only ever been
3125 * used on one vcpu. However, that doesn't mean it has
3126 * only ever been used on one physical cpu, since vcpus
3127 * can move around between pcpus. To cope with this, when
3128 * a vcpu moves from one pcpu to another, we need to tell
3129 * any vcpus running on the same core as this vcpu previously
3130 * ran to flush the TLB.
3132 if (prev_cpu != pcpu) {
3133 if (prev_cpu >= 0) {
3134 if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3135 cpu_first_tlb_thread_sibling(pcpu))
3136 radix_flush_cpu(kvm, prev_cpu, vcpu);
3138 smp_call_function_single(prev_cpu,
3139 do_migrate_away_vcpu, vcpu, 1);
3142 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3144 vcpu->arch.prev_cpu = pcpu;
3148 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3151 struct paca_struct *tpaca;
3155 if (vcpu->arch.timer_running) {
3156 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3157 vcpu->arch.timer_running = 0;
3159 cpu += vcpu->arch.ptid;
3160 vcpu->cpu = vc->pcpu;
3161 vcpu->arch.thread_cpu = cpu;
3163 tpaca = paca_ptrs[cpu];
3164 tpaca->kvm_hstate.kvm_vcpu = vcpu;
3165 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3166 tpaca->kvm_hstate.fake_suspend = 0;
3167 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3169 tpaca->kvm_hstate.kvm_vcore = vc;
3170 if (cpu != smp_processor_id())
3171 kvmppc_ipi_thread(cpu);
3174 static void kvmppc_wait_for_nap(int n_threads)
3176 int cpu = smp_processor_id();
3181 for (loops = 0; loops < 1000000; ++loops) {
3183 * Check if all threads are finished.
3184 * We set the vcore pointer when starting a thread
3185 * and the thread clears it when finished, so we look
3186 * for any threads that still have a non-NULL vcore ptr.
3188 for (i = 1; i < n_threads; ++i)
3189 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3191 if (i == n_threads) {
3198 for (i = 1; i < n_threads; ++i)
3199 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3200 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3204 * Check that we are on thread 0 and that any other threads in
3205 * this core are off-line. Then grab the threads so they can't
3208 static int on_primary_thread(void)
3210 int cpu = smp_processor_id();
3213 /* Are we on a primary subcore? */
3214 if (cpu_thread_in_subcore(cpu))
3218 while (++thr < threads_per_subcore)
3219 if (cpu_online(cpu + thr))
3222 /* Grab all hw threads so they can't go into the kernel */
3223 for (thr = 1; thr < threads_per_subcore; ++thr) {
3224 if (kvmppc_grab_hwthread(cpu + thr)) {
3225 /* Couldn't grab one; let the others go */
3227 kvmppc_release_hwthread(cpu + thr);
3228 } while (--thr > 0);
3236 * A list of virtual cores for each physical CPU.
3237 * These are vcores that could run but their runner VCPU tasks are
3238 * (or may be) preempted.
3240 struct preempted_vcore_list {
3241 struct list_head list;
3245 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3247 static void init_vcore_lists(void)
3251 for_each_possible_cpu(cpu) {
3252 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3253 spin_lock_init(&lp->lock);
3254 INIT_LIST_HEAD(&lp->list);
3258 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3260 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3262 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3264 vc->vcore_state = VCORE_PREEMPT;
3265 vc->pcpu = smp_processor_id();
3266 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3267 spin_lock(&lp->lock);
3268 list_add_tail(&vc->preempt_list, &lp->list);
3269 spin_unlock(&lp->lock);
3272 /* Start accumulating stolen time */
3273 kvmppc_core_start_stolen(vc, mftb());
3276 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3278 struct preempted_vcore_list *lp;
3280 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3282 kvmppc_core_end_stolen(vc, mftb());
3283 if (!list_empty(&vc->preempt_list)) {
3284 lp = &per_cpu(preempted_vcores, vc->pcpu);
3285 spin_lock(&lp->lock);
3286 list_del_init(&vc->preempt_list);
3287 spin_unlock(&lp->lock);
3289 vc->vcore_state = VCORE_INACTIVE;
3293 * This stores information about the virtual cores currently
3294 * assigned to a physical core.
3298 int max_subcore_threads;
3300 int subcore_threads[MAX_SUBCORES];
3301 struct kvmppc_vcore *vc[MAX_SUBCORES];
3305 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3306 * respectively in 2-way micro-threading (split-core) mode on POWER8.
3308 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3310 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3312 memset(cip, 0, sizeof(*cip));
3313 cip->n_subcores = 1;
3314 cip->max_subcore_threads = vc->num_threads;
3315 cip->total_threads = vc->num_threads;
3316 cip->subcore_threads[0] = vc->num_threads;
3320 static bool subcore_config_ok(int n_subcores, int n_threads)
3323 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3324 * split-core mode, with one thread per subcore.
3326 if (cpu_has_feature(CPU_FTR_ARCH_300))
3327 return n_subcores <= 4 && n_threads == 1;
3329 /* On POWER8, can only dynamically split if unsplit to begin with */
3330 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3332 if (n_subcores > MAX_SUBCORES)
3334 if (n_subcores > 1) {
3335 if (!(dynamic_mt_modes & 2))
3337 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3341 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3344 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3346 vc->entry_exit_map = 0;
3348 vc->napping_threads = 0;
3349 vc->conferring_threads = 0;
3350 vc->tb_offset_applied = 0;
3353 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3355 int n_threads = vc->num_threads;
3358 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3361 /* In one_vm_per_core mode, require all vcores to be from the same vm */
3362 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3365 if (n_threads < cip->max_subcore_threads)
3366 n_threads = cip->max_subcore_threads;
3367 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3369 cip->max_subcore_threads = n_threads;
3371 sub = cip->n_subcores;
3373 cip->total_threads += vc->num_threads;
3374 cip->subcore_threads[sub] = vc->num_threads;
3376 init_vcore_to_run(vc);
3377 list_del_init(&vc->preempt_list);
3383 * Work out whether it is possible to piggyback the execution of
3384 * vcore *pvc onto the execution of the other vcores described in *cip.
3386 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3389 if (cip->total_threads + pvc->num_threads > target_threads)
3392 return can_dynamic_split(pvc, cip);
3395 static void prepare_threads(struct kvmppc_vcore *vc)
3398 struct kvm_vcpu *vcpu;
3400 for_each_runnable_thread(i, vcpu, vc) {
3401 if (signal_pending(vcpu->arch.run_task))
3402 vcpu->arch.ret = -EINTR;
3403 else if (vcpu->arch.vpa.update_pending ||
3404 vcpu->arch.slb_shadow.update_pending ||
3405 vcpu->arch.dtl.update_pending)
3406 vcpu->arch.ret = RESUME_GUEST;
3409 kvmppc_remove_runnable(vc, vcpu, mftb());
3410 wake_up(&vcpu->arch.cpu_run);
3414 static void collect_piggybacks(struct core_info *cip, int target_threads)
3416 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3417 struct kvmppc_vcore *pvc, *vcnext;
3419 spin_lock(&lp->lock);
3420 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3421 if (!spin_trylock(&pvc->lock))
3423 prepare_threads(pvc);
3424 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3425 list_del_init(&pvc->preempt_list);
3426 if (pvc->runner == NULL) {
3427 pvc->vcore_state = VCORE_INACTIVE;
3428 kvmppc_core_end_stolen(pvc, mftb());
3430 spin_unlock(&pvc->lock);
3433 if (!can_piggyback(pvc, cip, target_threads)) {
3434 spin_unlock(&pvc->lock);
3437 kvmppc_core_end_stolen(pvc, mftb());
3438 pvc->vcore_state = VCORE_PIGGYBACK;
3439 if (cip->total_threads >= target_threads)
3442 spin_unlock(&lp->lock);
3445 static bool recheck_signals_and_mmu(struct core_info *cip)
3448 struct kvm_vcpu *vcpu;
3449 struct kvmppc_vcore *vc;
3451 for (sub = 0; sub < cip->n_subcores; ++sub) {
3453 if (!vc->kvm->arch.mmu_ready)
3455 for_each_runnable_thread(i, vcpu, vc)
3456 if (signal_pending(vcpu->arch.run_task))
3462 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3464 int still_running = 0, i;
3467 struct kvm_vcpu *vcpu;
3469 spin_lock(&vc->lock);
3471 for_each_runnable_thread(i, vcpu, vc) {
3473 * It's safe to unlock the vcore in the loop here, because
3474 * for_each_runnable_thread() is safe against removal of
3475 * the vcpu, and the vcore state is VCORE_EXITING here,
3476 * so any vcpus becoming runnable will have their arch.trap
3477 * set to zero and can't actually run in the guest.
3479 spin_unlock(&vc->lock);
3480 /* cancel pending dec exception if dec is positive */
3481 if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3482 kvmppc_core_pending_dec(vcpu))
3483 kvmppc_core_dequeue_dec(vcpu);
3485 trace_kvm_guest_exit(vcpu);
3488 if (vcpu->arch.trap)
3489 ret = kvmppc_handle_exit_hv(vcpu,
3490 vcpu->arch.run_task);
3492 vcpu->arch.ret = ret;
3493 vcpu->arch.trap = 0;
3495 spin_lock(&vc->lock);
3496 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3497 if (vcpu->arch.pending_exceptions)
3498 kvmppc_core_prepare_to_enter(vcpu);
3499 if (vcpu->arch.ceded)
3500 kvmppc_set_timer(vcpu);
3504 kvmppc_remove_runnable(vc, vcpu, mftb());
3505 wake_up(&vcpu->arch.cpu_run);
3509 if (still_running > 0) {
3510 kvmppc_vcore_preempt(vc);
3511 } else if (vc->runner) {
3512 vc->vcore_state = VCORE_PREEMPT;
3513 kvmppc_core_start_stolen(vc, mftb());
3515 vc->vcore_state = VCORE_INACTIVE;
3517 if (vc->n_runnable > 0 && vc->runner == NULL) {
3518 /* make sure there's a candidate runner awake */
3520 vcpu = next_runnable_thread(vc, &i);
3521 wake_up(&vcpu->arch.cpu_run);
3524 spin_unlock(&vc->lock);
3528 * Clear core from the list of active host cores as we are about to
3529 * enter the guest. Only do this if it is the primary thread of the
3530 * core (not if a subcore) that is entering the guest.
3532 static inline int kvmppc_clear_host_core(unsigned int cpu)
3536 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3539 * Memory barrier can be omitted here as we will do a smp_wmb()
3540 * later in kvmppc_start_thread and we need ensure that state is
3541 * visible to other CPUs only after we enter guest.
3543 core = cpu >> threads_shift;
3544 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3549 * Advertise this core as an active host core since we exited the guest
3550 * Only need to do this if it is the primary thread of the core that is
3553 static inline int kvmppc_set_host_core(unsigned int cpu)
3557 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3561 * Memory barrier can be omitted here because we do a spin_unlock
3562 * immediately after this which provides the memory barrier.
3564 core = cpu >> threads_shift;
3565 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3569 static void set_irq_happened(int trap)
3572 case BOOK3S_INTERRUPT_EXTERNAL:
3573 local_paca->irq_happened |= PACA_IRQ_EE;
3575 case BOOK3S_INTERRUPT_H_DOORBELL:
3576 local_paca->irq_happened |= PACA_IRQ_DBELL;
3578 case BOOK3S_INTERRUPT_HMI:
3579 local_paca->irq_happened |= PACA_IRQ_HMI;
3581 case BOOK3S_INTERRUPT_SYSTEM_RESET:
3582 replay_system_reset();
3588 * Run a set of guest threads on a physical core.
3589 * Called with vc->lock held.
3591 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3593 struct kvm_vcpu *vcpu;
3596 struct core_info core_info;
3597 struct kvmppc_vcore *pvc;
3598 struct kvm_split_mode split_info, *sip;
3599 int split, subcore_size, active;
3602 unsigned long cmd_bit, stat_bit;
3605 int controlled_threads;
3609 if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3613 * Remove from the list any threads that have a signal pending
3614 * or need a VPA update done
3616 prepare_threads(vc);
3618 /* if the runner is no longer runnable, let the caller pick a new one */
3619 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3625 init_vcore_to_run(vc);
3626 vc->preempt_tb = TB_NIL;
3629 * Number of threads that we will be controlling: the same as
3630 * the number of threads per subcore, except on POWER9,
3631 * where it's 1 because the threads are (mostly) independent.
3633 controlled_threads = threads_per_vcore(vc->kvm);
3636 * Make sure we are running on primary threads, and that secondary
3637 * threads are offline. Also check if the number of threads in this
3638 * guest are greater than the current system threads per guest.
3640 if ((controlled_threads > 1) &&
3641 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3642 for_each_runnable_thread(i, vcpu, vc) {
3643 vcpu->arch.ret = -EBUSY;
3644 kvmppc_remove_runnable(vc, vcpu, mftb());
3645 wake_up(&vcpu->arch.cpu_run);
3651 * See if we could run any other vcores on the physical core
3652 * along with this one.
3654 init_core_info(&core_info, vc);
3655 pcpu = smp_processor_id();
3656 target_threads = controlled_threads;
3657 if (target_smt_mode && target_smt_mode < target_threads)
3658 target_threads = target_smt_mode;
3659 if (vc->num_threads < target_threads)
3660 collect_piggybacks(&core_info, target_threads);
3663 * Hard-disable interrupts, and check resched flag and signals.
3664 * If we need to reschedule or deliver a signal, clean up
3665 * and return without going into the guest(s).
3666 * If the mmu_ready flag has been cleared, don't go into the
3667 * guest because that means a HPT resize operation is in progress.
3669 local_irq_disable();
3671 if (lazy_irq_pending() || need_resched() ||
3672 recheck_signals_and_mmu(&core_info)) {
3674 vc->vcore_state = VCORE_INACTIVE;
3675 /* Unlock all except the primary vcore */
3676 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3677 pvc = core_info.vc[sub];
3678 /* Put back on to the preempted vcores list */
3679 kvmppc_vcore_preempt(pvc);
3680 spin_unlock(&pvc->lock);
3682 for (i = 0; i < controlled_threads; ++i)
3683 kvmppc_release_hwthread(pcpu + i);
3687 kvmppc_clear_host_core(pcpu);
3689 /* Decide on micro-threading (split-core) mode */
3690 subcore_size = threads_per_subcore;
3691 cmd_bit = stat_bit = 0;
3692 split = core_info.n_subcores;
3694 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3698 memset(&split_info, 0, sizeof(split_info));
3699 for (sub = 0; sub < core_info.n_subcores; ++sub)
3700 split_info.vc[sub] = core_info.vc[sub];
3703 if (split == 2 && (dynamic_mt_modes & 2)) {
3704 cmd_bit = HID0_POWER8_1TO2LPAR;
3705 stat_bit = HID0_POWER8_2LPARMODE;
3708 cmd_bit = HID0_POWER8_1TO4LPAR;
3709 stat_bit = HID0_POWER8_4LPARMODE;
3711 subcore_size = MAX_SMT_THREADS / split;
3712 split_info.rpr = mfspr(SPRN_RPR);
3713 split_info.pmmar = mfspr(SPRN_PMMAR);
3714 split_info.ldbar = mfspr(SPRN_LDBAR);
3715 split_info.subcore_size = subcore_size;
3717 split_info.subcore_size = 1;
3720 /* order writes to split_info before kvm_split_mode pointer */
3724 for (thr = 0; thr < controlled_threads; ++thr) {
3725 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3727 paca->kvm_hstate.napping = 0;
3728 paca->kvm_hstate.kvm_split_mode = sip;
3731 /* Initiate micro-threading (split-core) on POWER8 if required */
3733 unsigned long hid0 = mfspr(SPRN_HID0);
3735 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3737 mtspr(SPRN_HID0, hid0);
3740 hid0 = mfspr(SPRN_HID0);
3741 if (hid0 & stat_bit)
3748 * On POWER8, set RWMR register.
3749 * Since it only affects PURR and SPURR, it doesn't affect
3750 * the host, so we don't save/restore the host value.
3753 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3754 int n_online = atomic_read(&vc->online_count);
3757 * Use the 8-thread value if we're doing split-core
3758 * or if the vcore's online count looks bogus.
3760 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3761 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3762 rwmr_val = p8_rwmr_values[n_online];
3763 mtspr(SPRN_RWMR, rwmr_val);
3766 /* Start all the threads */
3768 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3769 thr = is_power8 ? subcore_thread_map[sub] : sub;
3772 pvc = core_info.vc[sub];
3773 pvc->pcpu = pcpu + thr;
3774 for_each_runnable_thread(i, vcpu, pvc) {
3775 kvmppc_start_thread(vcpu, pvc);
3776 kvmppc_create_dtl_entry(vcpu, pvc);
3777 trace_kvm_guest_enter(vcpu);
3778 if (!vcpu->arch.ptid)
3780 active |= 1 << (thr + vcpu->arch.ptid);
3783 * We need to start the first thread of each subcore
3784 * even if it doesn't have a vcpu.
3787 kvmppc_start_thread(NULL, pvc);
3791 * Ensure that split_info.do_nap is set after setting
3792 * the vcore pointer in the PACA of the secondaries.
3797 * When doing micro-threading, poke the inactive threads as well.
3798 * This gets them to the nap instruction after kvm_do_nap,
3799 * which reduces the time taken to unsplit later.
3802 split_info.do_nap = 1; /* ask secondaries to nap when done */
3803 for (thr = 1; thr < threads_per_subcore; ++thr)
3804 if (!(active & (1 << thr)))
3805 kvmppc_ipi_thread(pcpu + thr);
3808 vc->vcore_state = VCORE_RUNNING;
3811 trace_kvmppc_run_core(vc, 0);
3813 for (sub = 0; sub < core_info.n_subcores; ++sub)
3814 spin_unlock(&core_info.vc[sub]->lock);
3816 guest_enter_irqoff();
3818 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3820 this_cpu_disable_ftrace();
3823 * Interrupts will be enabled once we get into the guest,
3824 * so tell lockdep that we're about to enable interrupts.
3826 trace_hardirqs_on();
3828 trap = __kvmppc_vcore_entry();
3830 trace_hardirqs_off();
3832 this_cpu_enable_ftrace();
3834 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3836 set_irq_happened(trap);
3838 spin_lock(&vc->lock);
3839 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3840 vc->vcore_state = VCORE_EXITING;
3842 /* wait for secondary threads to finish writing their state to memory */
3843 kvmppc_wait_for_nap(controlled_threads);
3845 /* Return to whole-core mode if we split the core earlier */
3847 unsigned long hid0 = mfspr(SPRN_HID0);
3848 unsigned long loops = 0;
3850 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3851 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3853 mtspr(SPRN_HID0, hid0);
3856 hid0 = mfspr(SPRN_HID0);
3857 if (!(hid0 & stat_bit))
3862 split_info.do_nap = 0;
3865 kvmppc_set_host_core(pcpu);
3867 context_tracking_guest_exit();
3868 if (!vtime_accounting_enabled_this_cpu()) {
3871 * Service IRQs here before vtime_account_guest_exit() so any
3872 * ticks that occurred while running the guest are accounted to
3873 * the guest. If vtime accounting is enabled, accounting uses
3874 * TB rather than ticks, so it can be done without enabling
3875 * interrupts here, which has the problem that it accounts
3876 * interrupt processing overhead to the host.
3878 local_irq_disable();
3880 vtime_account_guest_exit();
3884 /* Let secondaries go back to the offline loop */
3885 for (i = 0; i < controlled_threads; ++i) {
3886 kvmppc_release_hwthread(pcpu + i);
3887 if (sip && sip->napped[i])
3888 kvmppc_ipi_thread(pcpu + i);
3891 spin_unlock(&vc->lock);
3893 /* make sure updates to secondary vcpu structs are visible now */
3898 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3899 pvc = core_info.vc[sub];
3900 post_guest_process(pvc, pvc == vc);
3903 spin_lock(&vc->lock);
3906 vc->vcore_state = VCORE_INACTIVE;
3907 trace_kvmppc_run_core(vc, 1);
3910 static inline bool hcall_is_xics(unsigned long req)
3912 return req == H_EOI || req == H_CPPR || req == H_IPI ||
3913 req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
3916 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu)
3918 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3920 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3921 lp->yield_count = cpu_to_be32(yield_count);
3922 vcpu->arch.vpa.dirty = 1;
3926 /* call our hypervisor to load up HV regs and go */
3927 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
3929 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3930 unsigned long host_psscr;
3932 struct hv_guest_state hvregs;
3933 struct p9_host_os_sprs host_os_sprs;
3939 save_p9_host_os_sprs(&host_os_sprs);
3942 * We need to save and restore the guest visible part of the
3943 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3944 * doesn't do this for us. Note only required if pseries since
3945 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
3947 host_psscr = mfspr(SPRN_PSSCR_PR);
3949 kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
3950 if (lazy_irq_pending())
3953 if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
3954 msr = mfmsr(); /* TM restore can update msr */
3956 if (vcpu->arch.psscr != host_psscr)
3957 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3959 kvmhv_save_hv_regs(vcpu, &hvregs);
3961 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3962 hvregs.version = HV_GUEST_STATE_VERSION;
3963 if (vcpu->arch.nested) {
3964 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3965 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3967 hvregs.lpid = vcpu->kvm->arch.lpid;
3968 hvregs.vcpu_token = vcpu->vcpu_id;
3970 hvregs.hdec_expiry = time_limit;
3973 * When setting DEC, we must always deal with irq_work_raise
3974 * via NMI vs setting DEC. The problem occurs right as we
3975 * switch into guest mode if a NMI hits and sets pending work
3976 * and sets DEC, then that will apply to the guest and not
3977 * bring us back to the host.
3979 * irq_work_raise could check a flag (or possibly LPCR[HDICE]
3980 * for example) and set HDEC to 1? That wouldn't solve the
3981 * nested hv case which needs to abort the hcall or zero the
3984 * XXX: Another day's problem.
3986 mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
3988 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3989 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3990 switch_pmu_to_guest(vcpu, &host_os_sprs);
3991 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3992 __pa(&vcpu->arch.regs));
3993 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3994 switch_pmu_to_host(vcpu, &host_os_sprs);
3995 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3996 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3997 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3998 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
4000 store_vcpu_state(vcpu);
4002 dec = mfspr(SPRN_DEC);
4003 if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4006 vcpu->arch.dec_expires = dec + (*tb + vc->tb_offset);
4008 timer_rearm_host_dec(*tb);
4010 restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4011 if (vcpu->arch.psscr != host_psscr)
4012 mtspr(SPRN_PSSCR_PR, host_psscr);
4018 * Guest entry for POWER9 and later CPUs.
4020 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4021 unsigned long lpcr, u64 *tb)
4026 next_timer = timer_get_next_tb();
4027 if (*tb >= next_timer)
4028 return BOOK3S_INTERRUPT_HV_DECREMENTER;
4029 if (next_timer < time_limit)
4030 time_limit = next_timer;
4031 else if (*tb >= time_limit) /* nested time limit */
4032 return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4034 vcpu->arch.ceded = 0;
4036 vcpu_vpa_increment_dispatch(vcpu);
4038 if (kvmhv_on_pseries()) {
4039 trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4041 /* H_CEDE has to be handled now, not later */
4042 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
4043 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4045 kvmppc_set_gpr(vcpu, 3, 0);
4050 struct kvm *kvm = vcpu->kvm;
4052 kvmppc_xive_push_vcpu(vcpu);
4054 __this_cpu_write(cpu_in_guest, kvm);
4055 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4056 __this_cpu_write(cpu_in_guest, NULL);
4058 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
4059 !(vcpu->arch.shregs.msr & MSR_PR)) {
4060 unsigned long req = kvmppc_get_gpr(vcpu, 3);
4062 /* H_CEDE has to be handled now, not later */
4063 if (req == H_CEDE) {
4065 kvmppc_xive_rearm_escalation(vcpu); /* may un-cede */
4066 kvmppc_set_gpr(vcpu, 3, 0);
4069 /* XICS hcalls must be handled before xive is pulled */
4070 } else if (hcall_is_xics(req)) {
4073 ret = kvmppc_xive_xics_hcall(vcpu, req);
4074 if (ret != H_TOO_HARD) {
4075 kvmppc_set_gpr(vcpu, 3, ret);
4080 kvmppc_xive_pull_vcpu(vcpu);
4082 if (kvm_is_radix(kvm))
4083 vcpu->arch.slb_max = 0;
4086 vcpu_vpa_increment_dispatch(vcpu);
4092 * Wait for some other vcpu thread to execute us, and
4093 * wake us up when we need to handle something in the host.
4095 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4096 struct kvm_vcpu *vcpu, int wait_state)
4100 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4101 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4102 spin_unlock(&vc->lock);
4104 spin_lock(&vc->lock);
4106 finish_wait(&vcpu->arch.cpu_run, &wait);
4109 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4111 if (!halt_poll_ns_grow)
4114 vc->halt_poll_ns *= halt_poll_ns_grow;
4115 if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4116 vc->halt_poll_ns = halt_poll_ns_grow_start;
4119 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4121 if (halt_poll_ns_shrink == 0)
4122 vc->halt_poll_ns = 0;
4124 vc->halt_poll_ns /= halt_poll_ns_shrink;
4127 #ifdef CONFIG_KVM_XICS
4128 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4130 if (!xics_on_xive())
4132 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4133 vcpu->arch.xive_saved_state.cppr;
4136 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4140 #endif /* CONFIG_KVM_XICS */
4142 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4144 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4145 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4151 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4153 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4159 * Check to see if any of the runnable vcpus on the vcore have pending
4160 * exceptions or are no longer ceded
4162 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4164 struct kvm_vcpu *vcpu;
4167 for_each_runnable_thread(i, vcpu, vc) {
4168 if (kvmppc_vcpu_check_block(vcpu))
4176 * All the vcpus in this vcore are idle, so wait for a decrementer
4177 * or external interrupt to one of the vcpus. vc->lock is held.
4179 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4181 ktime_t cur, start_poll, start_wait;
4185 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4187 /* Poll for pending exceptions and ceded state */
4188 cur = start_poll = ktime_get();
4189 if (vc->halt_poll_ns) {
4190 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4191 ++vc->runner->stat.generic.halt_attempted_poll;
4193 vc->vcore_state = VCORE_POLLING;
4194 spin_unlock(&vc->lock);
4197 if (kvmppc_vcore_check_block(vc)) {
4202 } while (kvm_vcpu_can_poll(cur, stop));
4204 spin_lock(&vc->lock);
4205 vc->vcore_state = VCORE_INACTIVE;
4208 ++vc->runner->stat.generic.halt_successful_poll;
4213 prepare_to_rcuwait(&vc->wait);
4214 set_current_state(TASK_INTERRUPTIBLE);
4215 if (kvmppc_vcore_check_block(vc)) {
4216 finish_rcuwait(&vc->wait);
4218 /* If we polled, count this as a successful poll */
4219 if (vc->halt_poll_ns)
4220 ++vc->runner->stat.generic.halt_successful_poll;
4224 start_wait = ktime_get();
4226 vc->vcore_state = VCORE_SLEEPING;
4227 trace_kvmppc_vcore_blocked(vc, 0);
4228 spin_unlock(&vc->lock);
4230 finish_rcuwait(&vc->wait);
4231 spin_lock(&vc->lock);
4232 vc->vcore_state = VCORE_INACTIVE;
4233 trace_kvmppc_vcore_blocked(vc, 1);
4234 ++vc->runner->stat.halt_successful_wait;
4239 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4241 /* Attribute wait time */
4243 vc->runner->stat.generic.halt_wait_ns +=
4244 ktime_to_ns(cur) - ktime_to_ns(start_wait);
4245 KVM_STATS_LOG_HIST_UPDATE(
4246 vc->runner->stat.generic.halt_wait_hist,
4247 ktime_to_ns(cur) - ktime_to_ns(start_wait));
4248 /* Attribute failed poll time */
4249 if (vc->halt_poll_ns) {
4250 vc->runner->stat.generic.halt_poll_fail_ns +=
4251 ktime_to_ns(start_wait) -
4252 ktime_to_ns(start_poll);
4253 KVM_STATS_LOG_HIST_UPDATE(
4254 vc->runner->stat.generic.halt_poll_fail_hist,
4255 ktime_to_ns(start_wait) -
4256 ktime_to_ns(start_poll));
4259 /* Attribute successful poll time */
4260 if (vc->halt_poll_ns) {
4261 vc->runner->stat.generic.halt_poll_success_ns +=
4263 ktime_to_ns(start_poll);
4264 KVM_STATS_LOG_HIST_UPDATE(
4265 vc->runner->stat.generic.halt_poll_success_hist,
4266 ktime_to_ns(cur) - ktime_to_ns(start_poll));
4270 /* Adjust poll time */
4272 if (block_ns <= vc->halt_poll_ns)
4274 /* We slept and blocked for longer than the max halt time */
4275 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4276 shrink_halt_poll_ns(vc);
4277 /* We slept and our poll time is too small */
4278 else if (vc->halt_poll_ns < halt_poll_ns &&
4279 block_ns < halt_poll_ns)
4280 grow_halt_poll_ns(vc);
4281 if (vc->halt_poll_ns > halt_poll_ns)
4282 vc->halt_poll_ns = halt_poll_ns;
4284 vc->halt_poll_ns = 0;
4286 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4290 * This never fails for a radix guest, as none of the operations it does
4291 * for a radix guest can fail or have a way to report failure.
4293 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4296 struct kvm *kvm = vcpu->kvm;
4298 mutex_lock(&kvm->arch.mmu_setup_lock);
4299 if (!kvm->arch.mmu_ready) {
4300 if (!kvm_is_radix(kvm))
4301 r = kvmppc_hv_setup_htab_rma(vcpu);
4303 if (cpu_has_feature(CPU_FTR_ARCH_300))
4304 kvmppc_setup_partition_table(kvm);
4305 kvm->arch.mmu_ready = 1;
4308 mutex_unlock(&kvm->arch.mmu_setup_lock);
4312 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4314 struct kvm_run *run = vcpu->run;
4316 struct kvmppc_vcore *vc;
4319 trace_kvmppc_run_vcpu_enter(vcpu);
4321 run->exit_reason = 0;
4322 vcpu->arch.ret = RESUME_GUEST;
4323 vcpu->arch.trap = 0;
4324 kvmppc_update_vpas(vcpu);
4327 * Synchronize with other threads in this virtual core
4329 vc = vcpu->arch.vcore;
4330 spin_lock(&vc->lock);
4331 vcpu->arch.ceded = 0;
4332 vcpu->arch.run_task = current;
4333 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4334 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4335 vcpu->arch.busy_preempt = TB_NIL;
4336 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4340 * This happens the first time this is called for a vcpu.
4341 * If the vcore is already running, we may be able to start
4342 * this thread straight away and have it join in.
4344 if (!signal_pending(current)) {
4345 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4346 vc->vcore_state == VCORE_RUNNING) &&
4347 !VCORE_IS_EXITING(vc)) {
4348 kvmppc_create_dtl_entry(vcpu, vc);
4349 kvmppc_start_thread(vcpu, vc);
4350 trace_kvm_guest_enter(vcpu);
4351 } else if (vc->vcore_state == VCORE_SLEEPING) {
4352 rcuwait_wake_up(&vc->wait);
4357 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4358 !signal_pending(current)) {
4359 /* See if the MMU is ready to go */
4360 if (!vcpu->kvm->arch.mmu_ready) {
4361 spin_unlock(&vc->lock);
4362 r = kvmhv_setup_mmu(vcpu);
4363 spin_lock(&vc->lock);
4365 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4367 hardware_entry_failure_reason = 0;
4373 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4374 kvmppc_vcore_end_preempt(vc);
4376 if (vc->vcore_state != VCORE_INACTIVE) {
4377 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4380 for_each_runnable_thread(i, v, vc) {
4381 kvmppc_core_prepare_to_enter(v);
4382 if (signal_pending(v->arch.run_task)) {
4383 kvmppc_remove_runnable(vc, v, mftb());
4384 v->stat.signal_exits++;
4385 v->run->exit_reason = KVM_EXIT_INTR;
4386 v->arch.ret = -EINTR;
4387 wake_up(&v->arch.cpu_run);
4390 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4393 for_each_runnable_thread(i, v, vc) {
4394 if (!kvmppc_vcpu_woken(v))
4395 n_ceded += v->arch.ceded;
4400 if (n_ceded == vc->n_runnable) {
4401 kvmppc_vcore_blocked(vc);
4402 } else if (need_resched()) {
4403 kvmppc_vcore_preempt(vc);
4404 /* Let something else run */
4405 cond_resched_lock(&vc->lock);
4406 if (vc->vcore_state == VCORE_PREEMPT)
4407 kvmppc_vcore_end_preempt(vc);
4409 kvmppc_run_core(vc);
4414 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4415 (vc->vcore_state == VCORE_RUNNING ||
4416 vc->vcore_state == VCORE_EXITING ||
4417 vc->vcore_state == VCORE_PIGGYBACK))
4418 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4420 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4421 kvmppc_vcore_end_preempt(vc);
4423 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4424 kvmppc_remove_runnable(vc, vcpu, mftb());
4425 vcpu->stat.signal_exits++;
4426 run->exit_reason = KVM_EXIT_INTR;
4427 vcpu->arch.ret = -EINTR;
4430 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4431 /* Wake up some vcpu to run the core */
4433 v = next_runnable_thread(vc, &i);
4434 wake_up(&v->arch.cpu_run);
4437 trace_kvmppc_run_vcpu_exit(vcpu);
4438 spin_unlock(&vc->lock);
4439 return vcpu->arch.ret;
4442 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4445 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4446 struct kvm_run *run = vcpu->run;
4449 struct kvmppc_vcore *vc;
4450 struct kvm *kvm = vcpu->kvm;
4451 struct kvm_nested_guest *nested = vcpu->arch.nested;
4452 unsigned long flags;
4455 trace_kvmppc_run_vcpu_enter(vcpu);
4457 run->exit_reason = 0;
4458 vcpu->arch.ret = RESUME_GUEST;
4459 vcpu->arch.trap = 0;
4461 vc = vcpu->arch.vcore;
4462 vcpu->arch.ceded = 0;
4463 vcpu->arch.run_task = current;
4464 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4465 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4467 /* See if the MMU is ready to go */
4468 if (unlikely(!kvm->arch.mmu_ready)) {
4469 r = kvmhv_setup_mmu(vcpu);
4471 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4472 run->fail_entry.hardware_entry_failure_reason = 0;
4481 kvmppc_update_vpas(vcpu);
4484 pcpu = smp_processor_id();
4485 if (kvm_is_radix(kvm))
4486 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4488 /* flags save not required, but irq_pmu has no disable/enable API */
4489 powerpc_local_irq_pmu_save(flags);
4491 if (signal_pending(current))
4493 if (need_resched() || !kvm->arch.mmu_ready)
4497 kvmppc_core_prepare_to_enter(vcpu);
4498 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4499 &vcpu->arch.pending_exceptions))
4501 } else if (vcpu->arch.pending_exceptions ||
4502 vcpu->arch.doorbell_request ||
4503 xive_interrupt_pending(vcpu)) {
4504 vcpu->arch.ret = RESUME_HOST;
4508 if (vcpu->arch.timer_running) {
4509 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4510 vcpu->arch.timer_running = 0;
4516 vcpu->arch.thread_cpu = pcpu;
4518 local_paca->kvm_hstate.kvm_vcpu = vcpu;
4519 local_paca->kvm_hstate.ptid = 0;
4520 local_paca->kvm_hstate.fake_suspend = 0;
4522 __kvmppc_create_dtl_entry(vcpu, pcpu, tb + vc->tb_offset, 0);
4524 trace_kvm_guest_enter(vcpu);
4526 guest_enter_irqoff();
4528 srcu_idx = srcu_read_lock(&kvm->srcu);
4530 this_cpu_disable_ftrace();
4532 /* Tell lockdep that we're about to enable interrupts */
4533 trace_hardirqs_on();
4535 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4536 vcpu->arch.trap = trap;
4538 trace_hardirqs_off();
4540 this_cpu_enable_ftrace();
4542 srcu_read_unlock(&kvm->srcu, srcu_idx);
4544 set_irq_happened(trap);
4546 context_tracking_guest_exit();
4547 if (!vtime_accounting_enabled_this_cpu()) {
4550 * Service IRQs here before vtime_account_guest_exit() so any
4551 * ticks that occurred while running the guest are accounted to
4552 * the guest. If vtime accounting is enabled, accounting uses
4553 * TB rather than ticks, so it can be done without enabling
4554 * interrupts here, which has the problem that it accounts
4555 * interrupt processing overhead to the host.
4557 local_irq_disable();
4559 vtime_account_guest_exit();
4562 vcpu->arch.thread_cpu = -1;
4564 powerpc_local_irq_pmu_restore(flags);
4569 * cancel pending decrementer exception if DEC is now positive, or if
4570 * entering a nested guest in which case the decrementer is now owned
4571 * by L2 and the L1 decrementer is provided in hdec_expires
4573 if (kvmppc_core_pending_dec(vcpu) &&
4574 ((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
4575 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4576 kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4577 kvmppc_core_dequeue_dec(vcpu);
4579 trace_kvm_guest_exit(vcpu);
4583 r = kvmppc_handle_exit_hv(vcpu, current);
4585 r = kvmppc_handle_nested_exit(vcpu);
4589 if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
4590 kvmppc_set_timer(vcpu);
4592 prepare_to_rcuwait(wait);
4594 set_current_state(TASK_INTERRUPTIBLE);
4595 if (signal_pending(current)) {
4596 vcpu->stat.signal_exits++;
4597 run->exit_reason = KVM_EXIT_INTR;
4598 vcpu->arch.ret = -EINTR;
4602 if (kvmppc_vcpu_check_block(vcpu))
4605 trace_kvmppc_vcore_blocked(vc, 0);
4607 trace_kvmppc_vcore_blocked(vc, 1);
4609 finish_rcuwait(wait);
4611 vcpu->arch.ceded = 0;
4614 trace_kvmppc_run_vcpu_exit(vcpu);
4616 return vcpu->arch.ret;
4619 vcpu->stat.signal_exits++;
4620 run->exit_reason = KVM_EXIT_INTR;
4621 vcpu->arch.ret = -EINTR;
4623 powerpc_local_irq_pmu_restore(flags);
4628 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4630 struct kvm_run *run = vcpu->run;
4636 if (!vcpu->arch.sane) {
4637 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4641 /* No need to go into the guest when all we'll do is come back out */
4642 if (signal_pending(current)) {
4643 run->exit_reason = KVM_EXIT_INTR;
4647 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4649 * Don't allow entry with a suspended transaction, because
4650 * the guest entry/exit code will lose it.
4652 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4653 (current->thread.regs->msr & MSR_TM)) {
4654 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4655 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4656 run->fail_entry.hardware_entry_failure_reason = 0;
4663 * Force online to 1 for the sake of old userspace which doesn't
4666 if (!vcpu->arch.online) {
4667 atomic_inc(&vcpu->arch.vcore->online_count);
4668 vcpu->arch.online = 1;
4671 kvmppc_core_prepare_to_enter(vcpu);
4674 atomic_inc(&kvm->arch.vcpus_running);
4675 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4679 if (IS_ENABLED(CONFIG_PPC_FPU))
4681 if (cpu_has_feature(CPU_FTR_ALTIVEC))
4683 if (cpu_has_feature(CPU_FTR_VSX))
4685 if ((cpu_has_feature(CPU_FTR_TM) ||
4686 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
4687 (vcpu->arch.hfscr & HFSCR_TM))
4689 msr = msr_check_and_set(msr);
4691 kvmppc_save_user_regs();
4693 kvmppc_save_current_sprs();
4695 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4696 vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4697 vcpu->arch.pgdir = kvm->mm->pgd;
4698 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4701 if (cpu_has_feature(CPU_FTR_ARCH_300))
4702 r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4703 vcpu->arch.vcore->lpcr);
4705 r = kvmppc_run_vcpu(vcpu);
4707 if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4708 if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4710 * These should have been caught reflected
4711 * into the guest by now. Final sanity check:
4712 * don't allow userspace to execute hcalls in
4718 trace_kvm_hcall_enter(vcpu);
4719 r = kvmppc_pseries_do_hcall(vcpu);
4720 trace_kvm_hcall_exit(vcpu, r);
4721 kvmppc_core_prepare_to_enter(vcpu);
4722 } else if (r == RESUME_PAGE_FAULT) {
4723 srcu_idx = srcu_read_lock(&kvm->srcu);
4724 r = kvmppc_book3s_hv_page_fault(vcpu,
4725 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4726 srcu_read_unlock(&kvm->srcu, srcu_idx);
4727 } else if (r == RESUME_PASSTHROUGH) {
4728 if (WARN_ON(xics_on_xive()))
4731 r = kvmppc_xics_rm_complete(vcpu, 0);
4733 } while (is_kvmppc_resume_guest(r));
4735 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4736 atomic_dec(&kvm->arch.vcpus_running);
4738 srr_regs_clobbered();
4743 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4744 int shift, int sllp)
4746 (*sps)->page_shift = shift;
4747 (*sps)->slb_enc = sllp;
4748 (*sps)->enc[0].page_shift = shift;
4749 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4751 * Add 16MB MPSS support (may get filtered out by userspace)
4754 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4756 (*sps)->enc[1].page_shift = 24;
4757 (*sps)->enc[1].pte_enc = penc;
4763 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4764 struct kvm_ppc_smmu_info *info)
4766 struct kvm_ppc_one_seg_page_size *sps;
4769 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4770 * POWER7 doesn't support keys for instruction accesses,
4771 * POWER8 and POWER9 do.
4773 info->data_keys = 32;
4774 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4776 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4777 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4778 info->slb_size = 32;
4780 /* We only support these sizes for now, and no muti-size segments */
4781 sps = &info->sps[0];
4782 kvmppc_add_seg_page_size(&sps, 12, 0);
4783 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4784 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4786 /* If running as a nested hypervisor, we don't support HPT guests */
4787 if (kvmhv_on_pseries())
4788 info->flags |= KVM_PPC_NO_HASH;
4794 * Get (and clear) the dirty memory log for a memory slot.
4796 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4797 struct kvm_dirty_log *log)
4799 struct kvm_memslots *slots;
4800 struct kvm_memory_slot *memslot;
4803 unsigned long *buf, *p;
4804 struct kvm_vcpu *vcpu;
4806 mutex_lock(&kvm->slots_lock);
4809 if (log->slot >= KVM_USER_MEM_SLOTS)
4812 slots = kvm_memslots(kvm);
4813 memslot = id_to_memslot(slots, log->slot);
4815 if (!memslot || !memslot->dirty_bitmap)
4819 * Use second half of bitmap area because both HPT and radix
4820 * accumulate bits in the first half.
4822 n = kvm_dirty_bitmap_bytes(memslot);
4823 buf = memslot->dirty_bitmap + n / sizeof(long);
4826 if (kvm_is_radix(kvm))
4827 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4829 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4834 * We accumulate dirty bits in the first half of the
4835 * memslot's dirty_bitmap area, for when pages are paged
4836 * out or modified by the host directly. Pick up these
4837 * bits and add them to the map.
4839 p = memslot->dirty_bitmap;
4840 for (i = 0; i < n / sizeof(long); ++i)
4841 buf[i] |= xchg(&p[i], 0);
4843 /* Harvest dirty bits from VPA and DTL updates */
4844 /* Note: we never modify the SLB shadow buffer areas */
4845 kvm_for_each_vcpu(i, vcpu, kvm) {
4846 spin_lock(&vcpu->arch.vpa_update_lock);
4847 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4848 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4849 spin_unlock(&vcpu->arch.vpa_update_lock);
4853 if (copy_to_user(log->dirty_bitmap, buf, n))
4858 mutex_unlock(&kvm->slots_lock);
4862 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4864 vfree(slot->arch.rmap);
4865 slot->arch.rmap = NULL;
4868 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4869 const struct kvm_memory_slot *old,
4870 struct kvm_memory_slot *new,
4871 enum kvm_mr_change change)
4873 if (change == KVM_MR_CREATE) {
4874 unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
4876 if ((size >> PAGE_SHIFT) > totalram_pages())
4879 new->arch.rmap = vzalloc(size);
4880 if (!new->arch.rmap)
4882 } else if (change != KVM_MR_DELETE) {
4883 new->arch.rmap = old->arch.rmap;
4889 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4890 struct kvm_memory_slot *old,
4891 const struct kvm_memory_slot *new,
4892 enum kvm_mr_change change)
4895 * If we are creating or modifying a memslot, it might make
4896 * some address that was previously cached as emulated
4897 * MMIO be no longer emulated MMIO, so invalidate
4898 * all the caches of emulated MMIO translations.
4900 if (change != KVM_MR_DELETE)
4901 atomic64_inc(&kvm->arch.mmio_update);
4904 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4905 * have already called kvm_arch_flush_shadow_memslot() to
4906 * flush shadow mappings. For KVM_MR_CREATE we have no
4907 * previous mappings. So the only case to handle is
4908 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4910 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4911 * to get rid of any THP PTEs in the partition-scoped page tables
4912 * so we can track dirtiness at the page level; we flush when
4913 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4916 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4917 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4918 kvmppc_radix_flush_memslot(kvm, old);
4920 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4922 if (!kvm->arch.secure_guest)
4928 * @TODO kvmppc_uvmem_memslot_create() can fail and
4929 * return error. Fix this.
4931 kvmppc_uvmem_memslot_create(kvm, new);
4934 kvmppc_uvmem_memslot_delete(kvm, old);
4937 /* TODO: Handle KVM_MR_MOVE */
4943 * Update LPCR values in kvm->arch and in vcores.
4944 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4945 * of kvm->arch.lpcr update).
4947 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4952 if ((kvm->arch.lpcr & mask) == lpcr)
4955 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4957 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4958 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4962 spin_lock(&vc->lock);
4963 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4964 verify_lpcr(kvm, vc->lpcr);
4965 spin_unlock(&vc->lock);
4966 if (++cores_done >= kvm->arch.online_vcores)
4971 void kvmppc_setup_partition_table(struct kvm *kvm)
4973 unsigned long dw0, dw1;
4975 if (!kvm_is_radix(kvm)) {
4976 /* PS field - page size for VRMA */
4977 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4978 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4979 /* HTABSIZE and HTABORG fields */
4980 dw0 |= kvm->arch.sdr1;
4982 /* Second dword as set by userspace */
4983 dw1 = kvm->arch.process_table;
4985 dw0 = PATB_HR | radix__get_tree_size() |
4986 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4987 dw1 = PATB_GR | kvm->arch.process_table;
4989 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4993 * Set up HPT (hashed page table) and RMA (real-mode area).
4994 * Must be called with kvm->arch.mmu_setup_lock held.
4996 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4999 struct kvm *kvm = vcpu->kvm;
5001 struct kvm_memory_slot *memslot;
5002 struct vm_area_struct *vma;
5003 unsigned long lpcr = 0, senc;
5004 unsigned long psize, porder;
5007 /* Allocate hashed page table (if not done already) and reset it */
5008 if (!kvm->arch.hpt.virt) {
5009 int order = KVM_DEFAULT_HPT_ORDER;
5010 struct kvm_hpt_info info;
5012 err = kvmppc_allocate_hpt(&info, order);
5013 /* If we get here, it means userspace didn't specify a
5014 * size explicitly. So, try successively smaller
5015 * sizes if the default failed. */
5016 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5017 err = kvmppc_allocate_hpt(&info, order);
5020 pr_err("KVM: Couldn't alloc HPT\n");
5024 kvmppc_set_hpt(kvm, &info);
5027 /* Look up the memslot for guest physical address 0 */
5028 srcu_idx = srcu_read_lock(&kvm->srcu);
5029 memslot = gfn_to_memslot(kvm, 0);
5031 /* We must have some memory at 0 by now */
5033 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5036 /* Look up the VMA for the start of this memory slot */
5037 hva = memslot->userspace_addr;
5038 mmap_read_lock(kvm->mm);
5039 vma = vma_lookup(kvm->mm, hva);
5040 if (!vma || (vma->vm_flags & VM_IO))
5043 psize = vma_kernel_pagesize(vma);
5045 mmap_read_unlock(kvm->mm);
5047 /* We can handle 4k, 64k or 16M pages in the VRMA */
5048 if (psize >= 0x1000000)
5050 else if (psize >= 0x10000)
5054 porder = __ilog2(psize);
5056 senc = slb_pgsize_encoding(psize);
5057 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5058 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5059 /* Create HPTEs in the hash page table for the VRMA */
5060 kvmppc_map_vrma(vcpu, memslot, porder);
5062 /* Update VRMASD field in the LPCR */
5063 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5064 /* the -4 is to account for senc values starting at 0x10 */
5065 lpcr = senc << (LPCR_VRMASD_SH - 4);
5066 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5069 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5073 srcu_read_unlock(&kvm->srcu, srcu_idx);
5078 mmap_read_unlock(kvm->mm);
5083 * Must be called with kvm->arch.mmu_setup_lock held and
5084 * mmu_ready = 0 and no vcpus running.
5086 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5088 unsigned long lpcr, lpcr_mask;
5090 if (nesting_enabled(kvm))
5091 kvmhv_release_all_nested(kvm);
5092 kvmppc_rmap_reset(kvm);
5093 kvm->arch.process_table = 0;
5094 /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5095 spin_lock(&kvm->mmu_lock);
5096 kvm->arch.radix = 0;
5097 spin_unlock(&kvm->mmu_lock);
5098 kvmppc_free_radix(kvm);
5101 lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5102 if (cpu_has_feature(CPU_FTR_ARCH_31))
5103 lpcr_mask |= LPCR_HAIL;
5104 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5110 * Must be called with kvm->arch.mmu_setup_lock held and
5111 * mmu_ready = 0 and no vcpus running.
5113 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5115 unsigned long lpcr, lpcr_mask;
5118 err = kvmppc_init_vm_radix(kvm);
5121 kvmppc_rmap_reset(kvm);
5122 /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5123 spin_lock(&kvm->mmu_lock);
5124 kvm->arch.radix = 1;
5125 spin_unlock(&kvm->mmu_lock);
5126 kvmppc_free_hpt(&kvm->arch.hpt);
5128 lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5129 lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5130 if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5131 lpcr_mask |= LPCR_HAIL;
5132 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5133 (kvm->arch.host_lpcr & LPCR_HAIL))
5136 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5141 #ifdef CONFIG_KVM_XICS
5143 * Allocate a per-core structure for managing state about which cores are
5144 * running in the host versus the guest and for exchanging data between
5145 * real mode KVM and CPU running in the host.
5146 * This is only done for the first VM.
5147 * The allocated structure stays even if all VMs have stopped.
5148 * It is only freed when the kvm-hv module is unloaded.
5149 * It's OK for this routine to fail, we just don't support host
5150 * core operations like redirecting H_IPI wakeups.
5152 void kvmppc_alloc_host_rm_ops(void)
5154 struct kvmppc_host_rm_ops *ops;
5155 unsigned long l_ops;
5159 if (cpu_has_feature(CPU_FTR_ARCH_300))
5162 /* Not the first time here ? */
5163 if (kvmppc_host_rm_ops_hv != NULL)
5166 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5170 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5171 ops->rm_core = kzalloc(size, GFP_KERNEL);
5173 if (!ops->rm_core) {
5180 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5181 if (!cpu_online(cpu))
5184 core = cpu >> threads_shift;
5185 ops->rm_core[core].rm_state.in_host = 1;
5188 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5191 * Make the contents of the kvmppc_host_rm_ops structure visible
5192 * to other CPUs before we assign it to the global variable.
5193 * Do an atomic assignment (no locks used here), but if someone
5194 * beats us to it, just free our copy and return.
5197 l_ops = (unsigned long) ops;
5199 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5201 kfree(ops->rm_core);
5206 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5207 "ppc/kvm_book3s:prepare",
5208 kvmppc_set_host_core,
5209 kvmppc_clear_host_core);
5213 void kvmppc_free_host_rm_ops(void)
5215 if (kvmppc_host_rm_ops_hv) {
5216 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5217 kfree(kvmppc_host_rm_ops_hv->rm_core);
5218 kfree(kvmppc_host_rm_ops_hv);
5219 kvmppc_host_rm_ops_hv = NULL;
5224 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5226 unsigned long lpcr, lpid;
5230 mutex_init(&kvm->arch.uvmem_lock);
5231 INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5232 mutex_init(&kvm->arch.mmu_setup_lock);
5234 /* Allocate the guest's logical partition ID */
5236 lpid = kvmppc_alloc_lpid();
5239 kvm->arch.lpid = lpid;
5241 kvmppc_alloc_host_rm_ops();
5243 kvmhv_vm_nested_init(kvm);
5246 * Since we don't flush the TLB when tearing down a VM,
5247 * and this lpid might have previously been used,
5248 * make sure we flush on each core before running the new VM.
5249 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5250 * does this flush for us.
5252 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5253 cpumask_setall(&kvm->arch.need_tlb_flush);
5255 /* Start out with the default set of hcalls enabled */
5256 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5257 sizeof(kvm->arch.enabled_hcalls));
5259 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5260 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5262 /* Init LPCR for virtual RMA mode */
5263 if (cpu_has_feature(CPU_FTR_HVMODE)) {
5264 kvm->arch.host_lpid = mfspr(SPRN_LPID);
5265 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5266 lpcr &= LPCR_PECE | LPCR_LPES;
5270 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5271 LPCR_VPM0 | LPCR_VPM1;
5272 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5273 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5274 /* On POWER8 turn on online bit to enable PURR/SPURR */
5275 if (cpu_has_feature(CPU_FTR_ARCH_207S))
5278 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5279 * Set HVICE bit to enable hypervisor virtualization interrupts.
5280 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5281 * be unnecessary but better safe than sorry in case we re-enable
5282 * EE in HV mode with this LPCR still set)
5284 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5286 lpcr |= LPCR_HVICE | LPCR_HEIC;
5289 * If xive is enabled, we route 0x500 interrupts directly
5297 * If the host uses radix, the guest starts out as radix.
5299 if (radix_enabled()) {
5300 kvm->arch.radix = 1;
5301 kvm->arch.mmu_ready = 1;
5303 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5304 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5305 cpu_has_feature(CPU_FTR_ARCH_31) &&
5306 (kvm->arch.host_lpcr & LPCR_HAIL))
5308 ret = kvmppc_init_vm_radix(kvm);
5310 kvmppc_free_lpid(kvm->arch.lpid);
5313 kvmppc_setup_partition_table(kvm);
5316 verify_lpcr(kvm, lpcr);
5317 kvm->arch.lpcr = lpcr;
5319 /* Initialization for future HPT resizes */
5320 kvm->arch.resize_hpt = NULL;
5323 * Work out how many sets the TLB has, for the use of
5324 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5326 if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5328 * P10 will flush all the congruence class with a single tlbiel
5330 kvm->arch.tlb_sets = 1;
5331 } else if (radix_enabled())
5332 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
5333 else if (cpu_has_feature(CPU_FTR_ARCH_300))
5334 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
5335 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5336 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
5338 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
5341 * Track that we now have a HV mode VM active. This blocks secondary
5342 * CPU threads from coming online.
5344 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5345 kvm_hv_vm_activated();
5348 * Initialize smt_mode depending on processor.
5349 * POWER8 and earlier have to use "strict" threading, where
5350 * all vCPUs in a vcore have to run on the same (sub)core,
5351 * whereas on POWER9 the threads can each run a different
5354 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5355 kvm->arch.smt_mode = threads_per_subcore;
5357 kvm->arch.smt_mode = 1;
5358 kvm->arch.emul_smt_mode = 1;
5361 * Create a debugfs directory for the VM
5363 snprintf(buf, sizeof(buf), "vm%d", current->pid);
5364 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5365 kvmppc_mmu_debugfs_init(kvm);
5366 if (radix_enabled())
5367 kvmhv_radix_debugfs_init(kvm);
5372 static void kvmppc_free_vcores(struct kvm *kvm)
5376 for (i = 0; i < KVM_MAX_VCORES; ++i)
5377 kfree(kvm->arch.vcores[i]);
5378 kvm->arch.online_vcores = 0;
5381 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5383 debugfs_remove_recursive(kvm->arch.debugfs_dir);
5385 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5386 kvm_hv_vm_deactivated();
5388 kvmppc_free_vcores(kvm);
5391 if (kvm_is_radix(kvm))
5392 kvmppc_free_radix(kvm);
5394 kvmppc_free_hpt(&kvm->arch.hpt);
5396 /* Perform global invalidation and return lpid to the pool */
5397 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5398 if (nesting_enabled(kvm))
5399 kvmhv_release_all_nested(kvm);
5400 kvm->arch.process_table = 0;
5401 if (kvm->arch.secure_guest)
5402 uv_svm_terminate(kvm->arch.lpid);
5403 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5406 kvmppc_free_lpid(kvm->arch.lpid);
5408 kvmppc_free_pimap(kvm);
5411 /* We don't need to emulate any privileged instructions or dcbz */
5412 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5413 unsigned int inst, int *advance)
5415 return EMULATE_FAIL;
5418 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5421 return EMULATE_FAIL;
5424 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5427 return EMULATE_FAIL;
5430 static int kvmppc_core_check_processor_compat_hv(void)
5432 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5433 cpu_has_feature(CPU_FTR_ARCH_206))
5436 /* POWER9 in radix mode is capable of being a nested hypervisor. */
5437 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5443 #ifdef CONFIG_KVM_XICS
5445 void kvmppc_free_pimap(struct kvm *kvm)
5447 kfree(kvm->arch.pimap);
5450 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5452 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5455 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5457 struct irq_desc *desc;
5458 struct kvmppc_irq_map *irq_map;
5459 struct kvmppc_passthru_irqmap *pimap;
5460 struct irq_chip *chip;
5462 struct irq_data *host_data;
5464 if (!kvm_irq_bypass)
5467 desc = irq_to_desc(host_irq);
5471 mutex_lock(&kvm->lock);
5473 pimap = kvm->arch.pimap;
5474 if (pimap == NULL) {
5475 /* First call, allocate structure to hold IRQ map */
5476 pimap = kvmppc_alloc_pimap();
5477 if (pimap == NULL) {
5478 mutex_unlock(&kvm->lock);
5481 kvm->arch.pimap = pimap;
5485 * For now, we only support interrupts for which the EOI operation
5486 * is an OPAL call followed by a write to XIRR, since that's
5487 * what our real-mode EOI code does, or a XIVE interrupt
5489 chip = irq_data_get_irq_chip(&desc->irq_data);
5490 if (!chip || !is_pnv_opal_msi(chip)) {
5491 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5492 host_irq, guest_gsi);
5493 mutex_unlock(&kvm->lock);
5498 * See if we already have an entry for this guest IRQ number.
5499 * If it's mapped to a hardware IRQ number, that's an error,
5500 * otherwise re-use this entry.
5502 for (i = 0; i < pimap->n_mapped; i++) {
5503 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5504 if (pimap->mapped[i].r_hwirq) {
5505 mutex_unlock(&kvm->lock);
5512 if (i == KVMPPC_PIRQ_MAPPED) {
5513 mutex_unlock(&kvm->lock);
5514 return -EAGAIN; /* table is full */
5517 irq_map = &pimap->mapped[i];
5519 irq_map->v_hwirq = guest_gsi;
5520 irq_map->desc = desc;
5523 * Order the above two stores before the next to serialize with
5524 * the KVM real mode handler.
5529 * The 'host_irq' number is mapped in the PCI-MSI domain but
5530 * the underlying calls, which will EOI the interrupt in real
5531 * mode, need an HW IRQ number mapped in the XICS IRQ domain.
5533 host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
5534 irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
5536 if (i == pimap->n_mapped)
5540 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
5542 kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
5544 irq_map->r_hwirq = 0;
5546 mutex_unlock(&kvm->lock);
5551 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5553 struct irq_desc *desc;
5554 struct kvmppc_passthru_irqmap *pimap;
5557 if (!kvm_irq_bypass)
5560 desc = irq_to_desc(host_irq);
5564 mutex_lock(&kvm->lock);
5565 if (!kvm->arch.pimap)
5568 pimap = kvm->arch.pimap;
5570 for (i = 0; i < pimap->n_mapped; i++) {
5571 if (guest_gsi == pimap->mapped[i].v_hwirq)
5575 if (i == pimap->n_mapped) {
5576 mutex_unlock(&kvm->lock);
5581 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
5583 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5585 /* invalidate the entry (what do do on error from the above ?) */
5586 pimap->mapped[i].r_hwirq = 0;
5589 * We don't free this structure even when the count goes to
5590 * zero. The structure is freed when we destroy the VM.
5593 mutex_unlock(&kvm->lock);
5597 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5598 struct irq_bypass_producer *prod)
5601 struct kvm_kernel_irqfd *irqfd =
5602 container_of(cons, struct kvm_kernel_irqfd, consumer);
5604 irqfd->producer = prod;
5606 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5608 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5609 prod->irq, irqfd->gsi, ret);
5614 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5615 struct irq_bypass_producer *prod)
5618 struct kvm_kernel_irqfd *irqfd =
5619 container_of(cons, struct kvm_kernel_irqfd, consumer);
5621 irqfd->producer = NULL;
5624 * When producer of consumer is unregistered, we change back to
5625 * default external interrupt handling mode - KVM real mode
5626 * will switch back to host.
5628 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5630 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5631 prod->irq, irqfd->gsi, ret);
5635 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5636 unsigned int ioctl, unsigned long arg)
5638 struct kvm *kvm __maybe_unused = filp->private_data;
5639 void __user *argp = (void __user *)arg;
5644 case KVM_PPC_ALLOCATE_HTAB: {
5647 /* If we're a nested hypervisor, we currently only support radix */
5648 if (kvmhv_on_pseries()) {
5654 if (get_user(htab_order, (u32 __user *)argp))
5656 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5663 case KVM_PPC_GET_HTAB_FD: {
5664 struct kvm_get_htab_fd ghf;
5667 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5669 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5673 case KVM_PPC_RESIZE_HPT_PREPARE: {
5674 struct kvm_ppc_resize_hpt rhpt;
5677 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5680 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5684 case KVM_PPC_RESIZE_HPT_COMMIT: {
5685 struct kvm_ppc_resize_hpt rhpt;
5688 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5691 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5703 * List of hcall numbers to enable by default.
5704 * For compatibility with old userspace, we enable by default
5705 * all hcalls that were implemented before the hcall-enabling
5706 * facility was added. Note this list should not include H_RTAS.
5708 static unsigned int default_hcall_list[] = {
5714 #ifdef CONFIG_SPAPR_TCE_IOMMU
5724 #ifdef CONFIG_KVM_XICS
5735 static void init_default_hcalls(void)
5740 for (i = 0; default_hcall_list[i]; ++i) {
5741 hcall = default_hcall_list[i];
5742 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5743 __set_bit(hcall / 4, default_enabled_hcalls);
5747 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5753 /* If not on a POWER9, reject it */
5754 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5757 /* If any unknown flags set, reject it */
5758 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5761 /* GR (guest radix) bit in process_table field must match */
5762 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5763 if (!!(cfg->process_table & PATB_GR) != radix)
5766 /* Process table size field must be reasonable, i.e. <= 24 */
5767 if ((cfg->process_table & PRTS_MASK) > 24)
5770 /* We can change a guest to/from radix now, if the host is radix */
5771 if (radix && !radix_enabled())
5774 /* If we're a nested hypervisor, we currently only support radix */
5775 if (kvmhv_on_pseries() && !radix)
5778 mutex_lock(&kvm->arch.mmu_setup_lock);
5779 if (radix != kvm_is_radix(kvm)) {
5780 if (kvm->arch.mmu_ready) {
5781 kvm->arch.mmu_ready = 0;
5782 /* order mmu_ready vs. vcpus_running */
5784 if (atomic_read(&kvm->arch.vcpus_running)) {
5785 kvm->arch.mmu_ready = 1;
5791 err = kvmppc_switch_mmu_to_radix(kvm);
5793 err = kvmppc_switch_mmu_to_hpt(kvm);
5798 kvm->arch.process_table = cfg->process_table;
5799 kvmppc_setup_partition_table(kvm);
5801 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5802 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5806 mutex_unlock(&kvm->arch.mmu_setup_lock);
5810 static int kvmhv_enable_nested(struct kvm *kvm)
5814 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5816 if (!radix_enabled())
5819 /* kvm == NULL means the caller is testing if the capability exists */
5821 kvm->arch.nested_enable = true;
5825 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5830 if (kvmhv_vcpu_is_radix(vcpu)) {
5831 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5837 /* For now quadrants are the only way to access nested guest memory */
5838 if (rc && vcpu->arch.nested)
5844 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5849 if (kvmhv_vcpu_is_radix(vcpu)) {
5850 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5856 /* For now quadrants are the only way to access nested guest memory */
5857 if (rc && vcpu->arch.nested)
5863 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5865 unpin_vpa(kvm, vpa);
5867 vpa->pinned_addr = NULL;
5869 vpa->update_pending = 0;
5873 * Enable a guest to become a secure VM, or test whether
5874 * that could be enabled.
5875 * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5876 * tested (kvm == NULL) or enabled (kvm != NULL).
5878 static int kvmhv_enable_svm(struct kvm *kvm)
5880 if (!kvmppc_uvmem_available())
5883 kvm->arch.svm_enabled = 1;
5888 * IOCTL handler to turn off secure mode of guest
5890 * - Release all device pages
5891 * - Issue ucall to terminate the guest on the UV side
5892 * - Unpin the VPA pages.
5893 * - Reinit the partition scoped page tables
5895 static int kvmhv_svm_off(struct kvm *kvm)
5897 struct kvm_vcpu *vcpu;
5903 if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5906 mutex_lock(&kvm->arch.mmu_setup_lock);
5907 mmu_was_ready = kvm->arch.mmu_ready;
5908 if (kvm->arch.mmu_ready) {
5909 kvm->arch.mmu_ready = 0;
5910 /* order mmu_ready vs. vcpus_running */
5912 if (atomic_read(&kvm->arch.vcpus_running)) {
5913 kvm->arch.mmu_ready = 1;
5919 srcu_idx = srcu_read_lock(&kvm->srcu);
5920 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5921 struct kvm_memory_slot *memslot;
5922 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5928 kvm_for_each_memslot(memslot, bkt, slots) {
5929 kvmppc_uvmem_drop_pages(memslot, kvm, true);
5930 uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5933 srcu_read_unlock(&kvm->srcu, srcu_idx);
5935 ret = uv_svm_terminate(kvm->arch.lpid);
5936 if (ret != U_SUCCESS) {
5942 * When secure guest is reset, all the guest pages are sent
5943 * to UV via UV_PAGE_IN before the non-boot vcpus get a
5944 * chance to run and unpin their VPA pages. Unpinning of all
5945 * VPA pages is done here explicitly so that VPA pages
5946 * can be migrated to the secure side.
5948 * This is required to for the secure SMP guest to reboot
5951 kvm_for_each_vcpu(i, vcpu, kvm) {
5952 spin_lock(&vcpu->arch.vpa_update_lock);
5953 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5954 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5955 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5956 spin_unlock(&vcpu->arch.vpa_update_lock);
5959 kvmppc_setup_partition_table(kvm);
5960 kvm->arch.secure_guest = 0;
5961 kvm->arch.mmu_ready = mmu_was_ready;
5963 mutex_unlock(&kvm->arch.mmu_setup_lock);
5967 static int kvmhv_enable_dawr1(struct kvm *kvm)
5969 if (!cpu_has_feature(CPU_FTR_DAWR1))
5972 /* kvm == NULL means the caller is testing if the capability exists */
5974 kvm->arch.dawr1_enabled = true;
5978 static bool kvmppc_hash_v3_possible(void)
5980 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5983 if (!cpu_has_feature(CPU_FTR_HVMODE))
5987 * POWER9 chips before version 2.02 can't have some threads in
5988 * HPT mode and some in radix mode on the same core.
5990 if (radix_enabled()) {
5991 unsigned int pvr = mfspr(SPRN_PVR);
5992 if ((pvr >> 16) == PVR_POWER9 &&
5993 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5994 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
6001 static struct kvmppc_ops kvm_ops_hv = {
6002 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6003 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6004 .get_one_reg = kvmppc_get_one_reg_hv,
6005 .set_one_reg = kvmppc_set_one_reg_hv,
6006 .vcpu_load = kvmppc_core_vcpu_load_hv,
6007 .vcpu_put = kvmppc_core_vcpu_put_hv,
6008 .inject_interrupt = kvmppc_inject_interrupt_hv,
6009 .set_msr = kvmppc_set_msr_hv,
6010 .vcpu_run = kvmppc_vcpu_run_hv,
6011 .vcpu_create = kvmppc_core_vcpu_create_hv,
6012 .vcpu_free = kvmppc_core_vcpu_free_hv,
6013 .check_requests = kvmppc_core_check_requests_hv,
6014 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
6015 .flush_memslot = kvmppc_core_flush_memslot_hv,
6016 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6017 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
6018 .unmap_gfn_range = kvm_unmap_gfn_range_hv,
6019 .age_gfn = kvm_age_gfn_hv,
6020 .test_age_gfn = kvm_test_age_gfn_hv,
6021 .set_spte_gfn = kvm_set_spte_gfn_hv,
6022 .free_memslot = kvmppc_core_free_memslot_hv,
6023 .init_vm = kvmppc_core_init_vm_hv,
6024 .destroy_vm = kvmppc_core_destroy_vm_hv,
6025 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6026 .emulate_op = kvmppc_core_emulate_op_hv,
6027 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6028 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6029 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6030 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
6031 .hcall_implemented = kvmppc_hcall_impl_hv,
6032 #ifdef CONFIG_KVM_XICS
6033 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6034 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6036 .configure_mmu = kvmhv_configure_mmu,
6037 .get_rmmu_info = kvmhv_get_rmmu_info,
6038 .set_smt_mode = kvmhv_set_smt_mode,
6039 .enable_nested = kvmhv_enable_nested,
6040 .load_from_eaddr = kvmhv_load_from_eaddr,
6041 .store_to_eaddr = kvmhv_store_to_eaddr,
6042 .enable_svm = kvmhv_enable_svm,
6043 .svm_off = kvmhv_svm_off,
6044 .enable_dawr1 = kvmhv_enable_dawr1,
6045 .hash_v3_possible = kvmppc_hash_v3_possible,
6048 static int kvm_init_subcore_bitmap(void)
6051 int nr_cores = cpu_nr_cores();
6052 struct sibling_subcore_state *sibling_subcore_state;
6054 for (i = 0; i < nr_cores; i++) {
6055 int first_cpu = i * threads_per_core;
6056 int node = cpu_to_node(first_cpu);
6058 /* Ignore if it is already allocated. */
6059 if (paca_ptrs[first_cpu]->sibling_subcore_state)
6062 sibling_subcore_state =
6063 kzalloc_node(sizeof(struct sibling_subcore_state),
6065 if (!sibling_subcore_state)
6069 for (j = 0; j < threads_per_core; j++) {
6070 int cpu = first_cpu + j;
6072 paca_ptrs[cpu]->sibling_subcore_state =
6073 sibling_subcore_state;
6079 static int kvmppc_radix_possible(void)
6081 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6084 static int kvmppc_book3s_init_hv(void)
6088 if (!tlbie_capable) {
6089 pr_err("KVM-HV: Host does not support TLBIE\n");
6094 * FIXME!! Do we need to check on all cpus ?
6096 r = kvmppc_core_check_processor_compat_hv();
6100 r = kvmhv_nested_init();
6104 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6105 r = kvm_init_subcore_bitmap();
6111 * We need a way of accessing the XICS interrupt controller,
6112 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6113 * indirectly, via OPAL.
6116 if (!xics_on_xive() && !kvmhv_on_pseries() &&
6117 !local_paca->kvm_hstate.xics_phys) {
6118 struct device_node *np;
6120 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6122 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6125 /* presence of intc confirmed - node can be dropped again */
6130 kvm_ops_hv.owner = THIS_MODULE;
6131 kvmppc_hv_ops = &kvm_ops_hv;
6133 init_default_hcalls();
6137 r = kvmppc_mmu_hv_init();
6141 if (kvmppc_radix_possible())
6142 r = kvmppc_radix_init();
6144 r = kvmppc_uvmem_init();
6146 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6151 static void kvmppc_book3s_exit_hv(void)
6153 kvmppc_uvmem_free();
6154 kvmppc_free_host_rm_ops();
6155 if (kvmppc_radix_possible())
6156 kvmppc_radix_exit();
6157 kvmppc_hv_ops = NULL;
6158 kvmhv_nested_exit();
6161 module_init(kvmppc_book3s_init_hv);
6162 module_exit(kvmppc_book3s_exit_hv);
6163 MODULE_LICENSE("GPL");
6164 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6165 MODULE_ALIAS("devname:kvm");