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>
45 #include <linux/irqdomain.h>
47 #include <asm/ftrace.h>
49 #include <asm/ppc-opcode.h>
50 #include <asm/asm-prototypes.h>
51 #include <asm/archrandom.h>
52 #include <asm/debug.h>
53 #include <asm/disassemble.h>
54 #include <asm/cputable.h>
55 #include <asm/cacheflush.h>
56 #include <linux/uaccess.h>
57 #include <asm/interrupt.h>
59 #include <asm/kvm_ppc.h>
60 #include <asm/kvm_book3s.h>
61 #include <asm/mmu_context.h>
62 #include <asm/lppaca.h>
64 #include <asm/processor.h>
65 #include <asm/cputhreads.h>
67 #include <asm/hvcall.h>
68 #include <asm/switch_to.h>
70 #include <asm/dbell.h>
72 #include <asm/pnv-pci.h>
77 #include <asm/hw_breakpoint.h>
78 #include <asm/kvm_book3s_uvmem.h>
79 #include <asm/ultravisor.h>
81 #include <asm/plpar_wrappers.h>
84 #include "book3s_hv.h"
86 #define CREATE_TRACE_POINTS
89 /* #define EXIT_DEBUG */
90 /* #define EXIT_DEBUG_SIMPLE */
91 /* #define EXIT_DEBUG_INT */
93 /* Used to indicate that a guest page fault needs to be handled */
94 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
95 /* Used to indicate that a guest passthrough interrupt needs to be handled */
96 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
98 /* Used as a "null" value for timebase values */
99 #define TB_NIL (~(u64)0)
101 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
103 static int dynamic_mt_modes = 6;
104 module_param(dynamic_mt_modes, int, 0644);
105 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
106 static int target_smt_mode;
107 module_param(target_smt_mode, int, 0644);
108 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
110 static bool one_vm_per_core;
111 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
112 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
114 #ifdef CONFIG_KVM_XICS
115 static const struct kernel_param_ops module_param_ops = {
116 .set = param_set_int,
117 .get = param_get_int,
120 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
121 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
123 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
124 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
127 /* If set, guests are allowed to create and control nested guests */
128 static bool nested = true;
129 module_param(nested, bool, S_IRUGO | S_IWUSR);
130 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
132 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
135 * RWMR values for POWER8. These control the rate at which PURR
136 * and SPURR count and should be set according to the number of
137 * online threads in the vcore being run.
139 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
140 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
141 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
142 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
143 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
144 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
145 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
146 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
148 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
160 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
164 struct kvm_vcpu *vcpu;
166 while (++i < MAX_SMT_THREADS) {
167 vcpu = READ_ONCE(vc->runnable_threads[i]);
176 /* Used to traverse the list of runnable threads for a given vcore */
177 #define for_each_runnable_thread(i, vcpu, vc) \
178 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
180 static bool kvmppc_ipi_thread(int cpu)
182 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
184 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
185 if (kvmhv_on_pseries())
188 /* On POWER9 we can use msgsnd to IPI any cpu */
189 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
190 msg |= get_hard_smp_processor_id(cpu);
192 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
196 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
197 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
199 if (cpu_first_thread_sibling(cpu) ==
200 cpu_first_thread_sibling(smp_processor_id())) {
201 msg |= cpu_thread_in_core(cpu);
203 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
210 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
211 if (cpu >= 0 && cpu < nr_cpu_ids) {
212 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
216 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
224 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
227 struct rcuwait *waitp;
230 * rcuwait_wake_up contains smp_mb() which orders prior stores that
231 * create pending work vs below loads of cpu fields. The other side
232 * is the barrier in vcpu run that orders setting the cpu fields vs
233 * testing for pending work.
236 waitp = kvm_arch_vcpu_get_wait(vcpu);
237 if (rcuwait_wake_up(waitp))
238 ++vcpu->stat.generic.halt_wakeup;
240 cpu = READ_ONCE(vcpu->arch.thread_cpu);
241 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
244 /* CPU points to the first thread of the core */
246 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
247 smp_send_reschedule(cpu);
251 * We use the vcpu_load/put functions to measure stolen time.
252 * Stolen time is counted as time when either the vcpu is able to
253 * run as part of a virtual core, but the task running the vcore
254 * is preempted or sleeping, or when the vcpu needs something done
255 * in the kernel by the task running the vcpu, but that task is
256 * preempted or sleeping. Those two things have to be counted
257 * separately, since one of the vcpu tasks will take on the job
258 * of running the core, and the other vcpu tasks in the vcore will
259 * sleep waiting for it to do that, but that sleep shouldn't count
262 * Hence we accumulate stolen time when the vcpu can run as part of
263 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
264 * needs its task to do other things in the kernel (for example,
265 * service a page fault) in busy_stolen. We don't accumulate
266 * stolen time for a vcore when it is inactive, or for a vcpu
267 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
268 * a misnomer; it means that the vcpu task is not executing in
269 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
270 * the kernel. We don't have any way of dividing up that time
271 * between time that the vcpu is genuinely stopped, time that
272 * the task is actively working on behalf of the vcpu, and time
273 * that the task is preempted, so we don't count any of it as
276 * Updates to busy_stolen are protected by arch.tbacct_lock;
277 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
278 * lock. The stolen times are measured in units of timebase ticks.
279 * (Note that the != TB_NIL checks below are purely defensive;
280 * they should never fail.)
283 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb)
287 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
289 spin_lock_irqsave(&vc->stoltb_lock, flags);
291 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
294 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb)
298 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
300 spin_lock_irqsave(&vc->stoltb_lock, flags);
301 if (vc->preempt_tb != TB_NIL) {
302 vc->stolen_tb += tb - vc->preempt_tb;
303 vc->preempt_tb = TB_NIL;
305 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
308 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
310 struct kvmppc_vcore *vc = vcpu->arch.vcore;
314 if (cpu_has_feature(CPU_FTR_ARCH_300))
320 * We can test vc->runner without taking the vcore lock,
321 * because only this task ever sets vc->runner to this
322 * vcpu, and once it is set to this vcpu, only this task
323 * ever sets it to NULL.
325 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
326 kvmppc_core_end_stolen(vc, now);
328 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
329 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
330 vcpu->arch.busy_preempt != TB_NIL) {
331 vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt;
332 vcpu->arch.busy_preempt = TB_NIL;
334 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
337 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
339 struct kvmppc_vcore *vc = vcpu->arch.vcore;
343 if (cpu_has_feature(CPU_FTR_ARCH_300))
348 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
349 kvmppc_core_start_stolen(vc, now);
351 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
352 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
353 vcpu->arch.busy_preempt = now;
354 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
357 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
359 vcpu->arch.pvr = pvr;
362 /* Dummy value used in computing PCR value below */
363 #define PCR_ARCH_31 (PCR_ARCH_300 << 1)
365 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
367 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
368 struct kvmppc_vcore *vc = vcpu->arch.vcore;
370 /* We can (emulate) our own architecture version and anything older */
371 if (cpu_has_feature(CPU_FTR_ARCH_31))
372 host_pcr_bit = PCR_ARCH_31;
373 else if (cpu_has_feature(CPU_FTR_ARCH_300))
374 host_pcr_bit = PCR_ARCH_300;
375 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
376 host_pcr_bit = PCR_ARCH_207;
377 else if (cpu_has_feature(CPU_FTR_ARCH_206))
378 host_pcr_bit = PCR_ARCH_206;
380 host_pcr_bit = PCR_ARCH_205;
382 /* Determine lowest PCR bit needed to run guest in given PVR level */
383 guest_pcr_bit = host_pcr_bit;
385 switch (arch_compat) {
387 guest_pcr_bit = PCR_ARCH_205;
391 guest_pcr_bit = PCR_ARCH_206;
394 guest_pcr_bit = PCR_ARCH_207;
397 guest_pcr_bit = PCR_ARCH_300;
400 guest_pcr_bit = PCR_ARCH_31;
407 /* Check requested PCR bits don't exceed our capabilities */
408 if (guest_pcr_bit > host_pcr_bit)
411 spin_lock(&vc->lock);
412 vc->arch_compat = arch_compat;
414 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
415 * Also set all reserved PCR bits
417 vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
418 spin_unlock(&vc->lock);
423 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
427 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
428 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
429 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
430 for (r = 0; r < 16; ++r)
431 pr_err("r%2d = %.16lx r%d = %.16lx\n",
432 r, kvmppc_get_gpr(vcpu, r),
433 r+16, kvmppc_get_gpr(vcpu, r+16));
434 pr_err("ctr = %.16lx lr = %.16lx\n",
435 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
436 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
437 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
438 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
439 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
440 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
441 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
442 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
443 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
444 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
445 pr_err("fault dar = %.16lx dsisr = %.8x\n",
446 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
447 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
448 for (r = 0; r < vcpu->arch.slb_max; ++r)
449 pr_err(" ESID = %.16llx VSID = %.16llx\n",
450 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
451 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
452 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
453 vcpu->arch.last_inst);
456 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
458 return kvm_get_vcpu_by_id(kvm, id);
461 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
463 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
464 vpa->yield_count = cpu_to_be32(1);
467 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
468 unsigned long addr, unsigned long len)
470 /* check address is cacheline aligned */
471 if (addr & (L1_CACHE_BYTES - 1))
473 spin_lock(&vcpu->arch.vpa_update_lock);
474 if (v->next_gpa != addr || v->len != len) {
476 v->len = addr ? len : 0;
477 v->update_pending = 1;
479 spin_unlock(&vcpu->arch.vpa_update_lock);
483 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
492 static int vpa_is_registered(struct kvmppc_vpa *vpap)
494 if (vpap->update_pending)
495 return vpap->next_gpa != 0;
496 return vpap->pinned_addr != NULL;
499 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
501 unsigned long vcpuid, unsigned long vpa)
503 struct kvm *kvm = vcpu->kvm;
504 unsigned long len, nb;
506 struct kvm_vcpu *tvcpu;
509 struct kvmppc_vpa *vpap;
511 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
515 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
516 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
517 subfunc == H_VPA_REG_SLB) {
518 /* Registering new area - address must be cache-line aligned */
519 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
522 /* convert logical addr to kernel addr and read length */
523 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
526 if (subfunc == H_VPA_REG_VPA)
527 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
529 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
530 kvmppc_unpin_guest_page(kvm, va, vpa, false);
533 if (len > nb || len < sizeof(struct reg_vpa))
542 spin_lock(&tvcpu->arch.vpa_update_lock);
545 case H_VPA_REG_VPA: /* register VPA */
547 * The size of our lppaca is 1kB because of the way we align
548 * it for the guest to avoid crossing a 4kB boundary. We only
549 * use 640 bytes of the structure though, so we should accept
550 * clients that set a size of 640.
552 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
553 if (len < sizeof(struct lppaca))
555 vpap = &tvcpu->arch.vpa;
559 case H_VPA_REG_DTL: /* register DTL */
560 if (len < sizeof(struct dtl_entry))
562 len -= len % sizeof(struct dtl_entry);
564 /* Check that they have previously registered a VPA */
566 if (!vpa_is_registered(&tvcpu->arch.vpa))
569 vpap = &tvcpu->arch.dtl;
573 case H_VPA_REG_SLB: /* register SLB shadow buffer */
574 /* Check that they have previously registered a VPA */
576 if (!vpa_is_registered(&tvcpu->arch.vpa))
579 vpap = &tvcpu->arch.slb_shadow;
583 case H_VPA_DEREG_VPA: /* deregister VPA */
584 /* Check they don't still have a DTL or SLB buf registered */
586 if (vpa_is_registered(&tvcpu->arch.dtl) ||
587 vpa_is_registered(&tvcpu->arch.slb_shadow))
590 vpap = &tvcpu->arch.vpa;
594 case H_VPA_DEREG_DTL: /* deregister DTL */
595 vpap = &tvcpu->arch.dtl;
599 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
600 vpap = &tvcpu->arch.slb_shadow;
606 vpap->next_gpa = vpa;
608 vpap->update_pending = 1;
611 spin_unlock(&tvcpu->arch.vpa_update_lock);
616 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
618 struct kvm *kvm = vcpu->kvm;
624 * We need to pin the page pointed to by vpap->next_gpa,
625 * but we can't call kvmppc_pin_guest_page under the lock
626 * as it does get_user_pages() and down_read(). So we
627 * have to drop the lock, pin the page, then get the lock
628 * again and check that a new area didn't get registered
632 gpa = vpap->next_gpa;
633 spin_unlock(&vcpu->arch.vpa_update_lock);
637 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
638 spin_lock(&vcpu->arch.vpa_update_lock);
639 if (gpa == vpap->next_gpa)
641 /* sigh... unpin that one and try again */
643 kvmppc_unpin_guest_page(kvm, va, gpa, false);
646 vpap->update_pending = 0;
647 if (va && nb < vpap->len) {
649 * If it's now too short, it must be that userspace
650 * has changed the mappings underlying guest memory,
651 * so unregister the region.
653 kvmppc_unpin_guest_page(kvm, va, gpa, false);
656 if (vpap->pinned_addr)
657 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
660 vpap->pinned_addr = va;
663 vpap->pinned_end = va + vpap->len;
666 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
668 if (!(vcpu->arch.vpa.update_pending ||
669 vcpu->arch.slb_shadow.update_pending ||
670 vcpu->arch.dtl.update_pending))
673 spin_lock(&vcpu->arch.vpa_update_lock);
674 if (vcpu->arch.vpa.update_pending) {
675 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
676 if (vcpu->arch.vpa.pinned_addr)
677 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
679 if (vcpu->arch.dtl.update_pending) {
680 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
681 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
682 vcpu->arch.dtl_index = 0;
684 if (vcpu->arch.slb_shadow.update_pending)
685 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
686 spin_unlock(&vcpu->arch.vpa_update_lock);
690 * Return the accumulated stolen time for the vcore up until `now'.
691 * The caller should hold the vcore lock.
693 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
698 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
700 spin_lock_irqsave(&vc->stoltb_lock, flags);
702 if (vc->vcore_state != VCORE_INACTIVE &&
703 vc->preempt_tb != TB_NIL)
704 p += now - vc->preempt_tb;
705 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
709 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
710 unsigned int pcpu, u64 now,
711 unsigned long stolen)
713 struct dtl_entry *dt;
716 dt = vcpu->arch.dtl_ptr;
717 vpa = vcpu->arch.vpa.pinned_addr;
722 dt->dispatch_reason = 7;
723 dt->preempt_reason = 0;
724 dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
725 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
726 dt->ready_to_enqueue_time = 0;
727 dt->waiting_to_ready_time = 0;
728 dt->timebase = cpu_to_be64(now);
730 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
731 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
734 if (dt == vcpu->arch.dtl.pinned_end)
735 dt = vcpu->arch.dtl.pinned_addr;
736 vcpu->arch.dtl_ptr = dt;
737 /* order writing *dt vs. writing vpa->dtl_idx */
739 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
740 vcpu->arch.dtl.dirty = true;
743 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
744 struct kvmppc_vcore *vc)
746 unsigned long stolen;
747 unsigned long core_stolen;
753 core_stolen = vcore_stolen_time(vc, now);
754 stolen = core_stolen - vcpu->arch.stolen_logged;
755 vcpu->arch.stolen_logged = core_stolen;
756 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
757 stolen += vcpu->arch.busy_stolen;
758 vcpu->arch.busy_stolen = 0;
759 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
761 __kvmppc_create_dtl_entry(vcpu, vc->pcpu, now + vc->tb_offset, stolen);
764 /* See if there is a doorbell interrupt pending for a vcpu */
765 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
768 struct kvmppc_vcore *vc;
770 if (vcpu->arch.doorbell_request)
772 if (cpu_has_feature(CPU_FTR_ARCH_300))
775 * Ensure that the read of vcore->dpdes comes after the read
776 * of vcpu->doorbell_request. This barrier matches the
777 * smp_wmb() in kvmppc_guest_entry_inject().
780 vc = vcpu->arch.vcore;
781 thr = vcpu->vcpu_id - vc->first_vcpuid;
782 return !!(vc->dpdes & (1 << thr));
785 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
787 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
789 if ((!vcpu->arch.vcore->arch_compat) &&
790 cpu_has_feature(CPU_FTR_ARCH_207S))
795 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
796 unsigned long resource, unsigned long value1,
797 unsigned long value2)
800 case H_SET_MODE_RESOURCE_SET_CIABR:
801 if (!kvmppc_power8_compatible(vcpu))
806 return H_UNSUPPORTED_FLAG_START;
807 /* Guests can't breakpoint the hypervisor */
808 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
810 vcpu->arch.ciabr = value1;
812 case H_SET_MODE_RESOURCE_SET_DAWR0:
813 if (!kvmppc_power8_compatible(vcpu))
815 if (!ppc_breakpoint_available())
818 return H_UNSUPPORTED_FLAG_START;
819 if (value2 & DABRX_HYP)
821 vcpu->arch.dawr0 = value1;
822 vcpu->arch.dawrx0 = value2;
824 case H_SET_MODE_RESOURCE_SET_DAWR1:
825 if (!kvmppc_power8_compatible(vcpu))
827 if (!ppc_breakpoint_available())
829 if (!cpu_has_feature(CPU_FTR_DAWR1))
831 if (!vcpu->kvm->arch.dawr1_enabled)
834 return H_UNSUPPORTED_FLAG_START;
835 if (value2 & DABRX_HYP)
837 vcpu->arch.dawr1 = value1;
838 vcpu->arch.dawrx1 = value2;
840 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
842 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
843 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
845 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
846 kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
847 return H_UNSUPPORTED_FLAG_START;
854 /* Copy guest memory in place - must reside within a single memslot */
855 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
858 struct kvm_memory_slot *to_memslot = NULL;
859 struct kvm_memory_slot *from_memslot = NULL;
860 unsigned long to_addr, from_addr;
863 /* Get HPA for from address */
864 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
867 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
870 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
871 if (kvm_is_error_hva(from_addr))
873 from_addr |= (from & (PAGE_SIZE - 1));
875 /* Get HPA for to address */
876 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
879 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
882 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
883 if (kvm_is_error_hva(to_addr))
885 to_addr |= (to & (PAGE_SIZE - 1));
888 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
892 mark_page_dirty(kvm, to >> PAGE_SHIFT);
896 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
897 unsigned long dest, unsigned long src)
899 u64 pg_sz = SZ_4K; /* 4K page size */
900 u64 pg_mask = SZ_4K - 1;
903 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
904 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
905 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
908 /* dest (and src if copy_page flag set) must be page aligned */
909 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
912 /* zero and/or copy the page as determined by the flags */
913 if (flags & H_COPY_PAGE) {
914 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
917 } else if (flags & H_ZERO_PAGE) {
918 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
923 /* We can ignore the remaining flags */
928 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
930 struct kvmppc_vcore *vcore = target->arch.vcore;
933 * We expect to have been called by the real mode handler
934 * (kvmppc_rm_h_confer()) which would have directly returned
935 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
936 * have useful work to do and should not confer) so we don't
939 * In the case of the P9 single vcpu per vcore case, the real
940 * mode handler is not called but no other threads are in the
943 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
944 spin_lock(&vcore->lock);
945 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
946 vcore->vcore_state != VCORE_INACTIVE &&
948 target = vcore->runner;
949 spin_unlock(&vcore->lock);
952 return kvm_vcpu_yield_to(target);
955 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
958 struct lppaca *lppaca;
960 spin_lock(&vcpu->arch.vpa_update_lock);
961 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
963 yield_count = be32_to_cpu(lppaca->yield_count);
964 spin_unlock(&vcpu->arch.vpa_update_lock);
969 * H_RPT_INVALIDATE hcall handler for nested guests.
971 * Handles only nested process-scoped invalidation requests in L0.
973 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
975 unsigned long type = kvmppc_get_gpr(vcpu, 6);
976 unsigned long pid, pg_sizes, start, end;
979 * The partition-scoped invalidations aren't handled here in L0.
981 if (type & H_RPTI_TYPE_NESTED)
984 pid = kvmppc_get_gpr(vcpu, 4);
985 pg_sizes = kvmppc_get_gpr(vcpu, 7);
986 start = kvmppc_get_gpr(vcpu, 8);
987 end = kvmppc_get_gpr(vcpu, 9);
989 do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
990 type, pg_sizes, start, end);
992 kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
996 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
997 unsigned long id, unsigned long target,
998 unsigned long type, unsigned long pg_sizes,
999 unsigned long start, unsigned long end)
1001 if (!kvm_is_radix(vcpu->kvm))
1002 return H_UNSUPPORTED;
1008 * Partition-scoped invalidation for nested guests.
1010 if (type & H_RPTI_TYPE_NESTED) {
1011 if (!nesting_enabled(vcpu->kvm))
1014 /* Support only cores as target */
1015 if (target != H_RPTI_TARGET_CMMU)
1018 return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1023 * Process-scoped invalidation for L1 guests.
1025 do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1026 type, pg_sizes, start, end);
1030 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1032 struct kvm *kvm = vcpu->kvm;
1033 unsigned long req = kvmppc_get_gpr(vcpu, 3);
1034 unsigned long target, ret = H_SUCCESS;
1036 struct kvm_vcpu *tvcpu;
1039 if (req <= MAX_HCALL_OPCODE &&
1040 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1045 ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1046 kvmppc_get_gpr(vcpu, 5),
1047 kvmppc_get_gpr(vcpu, 6));
1048 if (ret == H_TOO_HARD)
1052 ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1053 kvmppc_get_gpr(vcpu, 5),
1054 kvmppc_get_gpr(vcpu, 6),
1055 kvmppc_get_gpr(vcpu, 7));
1056 if (ret == H_TOO_HARD)
1060 ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1061 kvmppc_get_gpr(vcpu, 5));
1062 if (ret == H_TOO_HARD)
1066 ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1067 kvmppc_get_gpr(vcpu, 5));
1068 if (ret == H_TOO_HARD)
1072 ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1073 kvmppc_get_gpr(vcpu, 5));
1074 if (ret == H_TOO_HARD)
1078 ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1079 kvmppc_get_gpr(vcpu, 5),
1080 kvmppc_get_gpr(vcpu, 6));
1081 if (ret == H_TOO_HARD)
1085 ret = kvmppc_h_bulk_remove(vcpu);
1086 if (ret == H_TOO_HARD)
1093 target = kvmppc_get_gpr(vcpu, 4);
1094 tvcpu = kvmppc_find_vcpu(kvm, target);
1099 tvcpu->arch.prodded = 1;
1100 smp_mb(); /* This orders prodded store vs ceded load */
1101 if (tvcpu->arch.ceded)
1102 kvmppc_fast_vcpu_kick_hv(tvcpu);
1105 target = kvmppc_get_gpr(vcpu, 4);
1108 tvcpu = kvmppc_find_vcpu(kvm, target);
1113 yield_count = kvmppc_get_gpr(vcpu, 5);
1114 if (kvmppc_get_yield_count(tvcpu) != yield_count)
1116 kvm_arch_vcpu_yield_to(tvcpu);
1118 case H_REGISTER_VPA:
1119 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1120 kvmppc_get_gpr(vcpu, 5),
1121 kvmppc_get_gpr(vcpu, 6));
1124 if (list_empty(&kvm->arch.rtas_tokens))
1127 idx = srcu_read_lock(&kvm->srcu);
1128 rc = kvmppc_rtas_hcall(vcpu);
1129 srcu_read_unlock(&kvm->srcu, idx);
1136 /* Send the error out to userspace via KVM_RUN */
1138 case H_LOGICAL_CI_LOAD:
1139 ret = kvmppc_h_logical_ci_load(vcpu);
1140 if (ret == H_TOO_HARD)
1143 case H_LOGICAL_CI_STORE:
1144 ret = kvmppc_h_logical_ci_store(vcpu);
1145 if (ret == H_TOO_HARD)
1149 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1150 kvmppc_get_gpr(vcpu, 5),
1151 kvmppc_get_gpr(vcpu, 6),
1152 kvmppc_get_gpr(vcpu, 7));
1153 if (ret == H_TOO_HARD)
1162 if (kvmppc_xics_enabled(vcpu)) {
1163 if (xics_on_xive()) {
1164 ret = H_NOT_AVAILABLE;
1165 return RESUME_GUEST;
1167 ret = kvmppc_xics_hcall(vcpu, req);
1172 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1175 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1176 kvmppc_get_gpr(vcpu, 5));
1178 #ifdef CONFIG_SPAPR_TCE_IOMMU
1180 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1181 kvmppc_get_gpr(vcpu, 5));
1182 if (ret == H_TOO_HARD)
1186 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1187 kvmppc_get_gpr(vcpu, 5),
1188 kvmppc_get_gpr(vcpu, 6));
1189 if (ret == H_TOO_HARD)
1192 case H_PUT_TCE_INDIRECT:
1193 ret = kvmppc_h_put_tce_indirect(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)
1201 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1202 kvmppc_get_gpr(vcpu, 5),
1203 kvmppc_get_gpr(vcpu, 6),
1204 kvmppc_get_gpr(vcpu, 7));
1205 if (ret == H_TOO_HARD)
1210 if (!arch_get_random_seed_long(&vcpu->arch.regs.gpr[4]))
1213 case H_RPT_INVALIDATE:
1214 ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1215 kvmppc_get_gpr(vcpu, 5),
1216 kvmppc_get_gpr(vcpu, 6),
1217 kvmppc_get_gpr(vcpu, 7),
1218 kvmppc_get_gpr(vcpu, 8),
1219 kvmppc_get_gpr(vcpu, 9));
1222 case H_SET_PARTITION_TABLE:
1224 if (nesting_enabled(kvm))
1225 ret = kvmhv_set_partition_table(vcpu);
1227 case H_ENTER_NESTED:
1229 if (!nesting_enabled(kvm))
1231 ret = kvmhv_enter_nested_guest(vcpu);
1232 if (ret == H_INTERRUPT) {
1233 kvmppc_set_gpr(vcpu, 3, 0);
1234 vcpu->arch.hcall_needed = 0;
1236 } else if (ret == H_TOO_HARD) {
1237 kvmppc_set_gpr(vcpu, 3, 0);
1238 vcpu->arch.hcall_needed = 0;
1242 case H_TLB_INVALIDATE:
1244 if (nesting_enabled(kvm))
1245 ret = kvmhv_do_nested_tlbie(vcpu);
1247 case H_COPY_TOFROM_GUEST:
1249 if (nesting_enabled(kvm))
1250 ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1253 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1254 kvmppc_get_gpr(vcpu, 5),
1255 kvmppc_get_gpr(vcpu, 6));
1258 ret = H_UNSUPPORTED;
1259 if (kvmppc_get_srr1(vcpu) & MSR_S)
1260 ret = kvmppc_h_svm_page_in(kvm,
1261 kvmppc_get_gpr(vcpu, 4),
1262 kvmppc_get_gpr(vcpu, 5),
1263 kvmppc_get_gpr(vcpu, 6));
1265 case H_SVM_PAGE_OUT:
1266 ret = H_UNSUPPORTED;
1267 if (kvmppc_get_srr1(vcpu) & MSR_S)
1268 ret = kvmppc_h_svm_page_out(kvm,
1269 kvmppc_get_gpr(vcpu, 4),
1270 kvmppc_get_gpr(vcpu, 5),
1271 kvmppc_get_gpr(vcpu, 6));
1273 case H_SVM_INIT_START:
1274 ret = H_UNSUPPORTED;
1275 if (kvmppc_get_srr1(vcpu) & MSR_S)
1276 ret = kvmppc_h_svm_init_start(kvm);
1278 case H_SVM_INIT_DONE:
1279 ret = H_UNSUPPORTED;
1280 if (kvmppc_get_srr1(vcpu) & MSR_S)
1281 ret = kvmppc_h_svm_init_done(kvm);
1283 case H_SVM_INIT_ABORT:
1285 * Even if that call is made by the Ultravisor, the SSR1 value
1286 * is the guest context one, with the secure bit clear as it has
1287 * not yet been secured. So we can't check it here.
1288 * Instead the kvm->arch.secure_guest flag is checked inside
1289 * kvmppc_h_svm_init_abort().
1291 ret = kvmppc_h_svm_init_abort(kvm);
1297 WARN_ON_ONCE(ret == H_TOO_HARD);
1298 kvmppc_set_gpr(vcpu, 3, ret);
1299 vcpu->arch.hcall_needed = 0;
1300 return RESUME_GUEST;
1304 * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1305 * handlers in book3s_hv_rmhandlers.S.
1307 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1308 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1310 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1312 vcpu->arch.shregs.msr |= MSR_EE;
1313 vcpu->arch.ceded = 1;
1315 if (vcpu->arch.prodded) {
1316 vcpu->arch.prodded = 0;
1318 vcpu->arch.ceded = 0;
1322 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1328 case H_REGISTER_VPA:
1330 #ifdef CONFIG_SPAPR_TCE_IOMMU
1333 case H_PUT_TCE_INDIRECT:
1336 case H_LOGICAL_CI_LOAD:
1337 case H_LOGICAL_CI_STORE:
1338 #ifdef CONFIG_KVM_XICS
1347 case H_RPT_INVALIDATE:
1351 /* See if it's in the real-mode table */
1352 return kvmppc_hcall_impl_hv_realmode(cmd);
1355 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1359 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1362 * Fetch failed, so return to guest and
1363 * try executing it again.
1365 return RESUME_GUEST;
1368 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1369 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1370 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1373 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1374 return RESUME_GUEST;
1378 static void do_nothing(void *x)
1382 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1384 int thr, cpu, pcpu, nthreads;
1386 unsigned long dpdes;
1388 nthreads = vcpu->kvm->arch.emul_smt_mode;
1390 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1391 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1392 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1396 * If the vcpu is currently running on a physical cpu thread,
1397 * interrupt it in order to pull it out of the guest briefly,
1398 * which will update its vcore->dpdes value.
1400 pcpu = READ_ONCE(v->cpu);
1402 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1403 if (kvmppc_doorbell_pending(v))
1410 * On POWER9, emulate doorbell-related instructions in order to
1411 * give the guest the illusion of running on a multi-threaded core.
1412 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1415 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1419 struct kvm *kvm = vcpu->kvm;
1420 struct kvm_vcpu *tvcpu;
1422 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1423 return RESUME_GUEST;
1424 if (get_op(inst) != 31)
1425 return EMULATE_FAIL;
1427 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1428 switch (get_xop(inst)) {
1429 case OP_31_XOP_MSGSNDP:
1430 arg = kvmppc_get_gpr(vcpu, rb);
1431 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1434 if (arg >= kvm->arch.emul_smt_mode)
1436 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1439 if (!tvcpu->arch.doorbell_request) {
1440 tvcpu->arch.doorbell_request = 1;
1441 kvmppc_fast_vcpu_kick_hv(tvcpu);
1444 case OP_31_XOP_MSGCLRP:
1445 arg = kvmppc_get_gpr(vcpu, rb);
1446 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1448 vcpu->arch.vcore->dpdes = 0;
1449 vcpu->arch.doorbell_request = 0;
1451 case OP_31_XOP_MFSPR:
1452 switch (get_sprn(inst)) {
1457 arg = kvmppc_read_dpdes(vcpu);
1460 return EMULATE_FAIL;
1462 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1465 return EMULATE_FAIL;
1467 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1468 return RESUME_GUEST;
1472 * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1473 * HFSCR_PM is cleared for next entry. If the guest then tries to access
1474 * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1475 * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1476 * allow the guest access to continue.
1478 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1480 if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1481 return EMULATE_FAIL;
1483 vcpu->arch.hfscr |= HFSCR_PM;
1485 return RESUME_GUEST;
1488 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1490 if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1491 return EMULATE_FAIL;
1493 vcpu->arch.hfscr |= HFSCR_EBB;
1495 return RESUME_GUEST;
1498 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1500 if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1501 return EMULATE_FAIL;
1503 vcpu->arch.hfscr |= HFSCR_TM;
1505 return RESUME_GUEST;
1508 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1509 struct task_struct *tsk)
1511 struct kvm_run *run = vcpu->run;
1512 int r = RESUME_HOST;
1514 vcpu->stat.sum_exits++;
1517 * This can happen if an interrupt occurs in the last stages
1518 * of guest entry or the first stages of guest exit (i.e. after
1519 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1520 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1521 * That can happen due to a bug, or due to a machine check
1522 * occurring at just the wrong time.
1524 if (vcpu->arch.shregs.msr & MSR_HV) {
1525 printk(KERN_EMERG "KVM trap in HV mode!\n");
1526 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1527 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1528 vcpu->arch.shregs.msr);
1529 kvmppc_dump_regs(vcpu);
1530 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1531 run->hw.hardware_exit_reason = vcpu->arch.trap;
1534 run->exit_reason = KVM_EXIT_UNKNOWN;
1535 run->ready_for_interrupt_injection = 1;
1536 switch (vcpu->arch.trap) {
1537 /* We're good on these - the host merely wanted to get our attention */
1538 case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1539 WARN_ON_ONCE(1); /* Should never happen */
1540 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1542 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1543 vcpu->stat.dec_exits++;
1546 case BOOK3S_INTERRUPT_EXTERNAL:
1547 case BOOK3S_INTERRUPT_H_DOORBELL:
1548 case BOOK3S_INTERRUPT_H_VIRT:
1549 vcpu->stat.ext_intr_exits++;
1552 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1553 case BOOK3S_INTERRUPT_HMI:
1554 case BOOK3S_INTERRUPT_PERFMON:
1555 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1558 case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1559 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1560 DEFAULT_RATELIMIT_BURST);
1562 * Print the MCE event to host console. Ratelimit so the guest
1563 * can't flood the host log.
1565 if (__ratelimit(&rs))
1566 machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1569 * If the guest can do FWNMI, exit to userspace so it can
1570 * deliver a FWNMI to the guest.
1571 * Otherwise we synthesize a machine check for the guest
1572 * so that it knows that the machine check occurred.
1574 if (!vcpu->kvm->arch.fwnmi_enabled) {
1575 ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1576 kvmppc_core_queue_machine_check(vcpu, flags);
1581 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1582 run->exit_reason = KVM_EXIT_NMI;
1583 run->hw.hardware_exit_reason = vcpu->arch.trap;
1584 /* Clear out the old NMI status from run->flags */
1585 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1586 /* Now set the NMI status */
1587 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1588 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1590 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1595 case BOOK3S_INTERRUPT_PROGRAM:
1599 * Normally program interrupts are delivered directly
1600 * to the guest by the hardware, but we can get here
1601 * as a result of a hypervisor emulation interrupt
1602 * (e40) getting turned into a 700 by BML RTAS.
1604 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1605 kvmppc_core_queue_program(vcpu, flags);
1609 case BOOK3S_INTERRUPT_SYSCALL:
1613 if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1615 * Guest userspace executed sc 1. This can only be
1616 * reached by the P9 path because the old path
1617 * handles this case in realmode hcall handlers.
1619 if (!kvmhv_vcpu_is_radix(vcpu)) {
1621 * A guest could be running PR KVM, so this
1622 * may be a PR KVM hcall. It must be reflected
1623 * to the guest kernel as a sc interrupt.
1625 kvmppc_core_queue_syscall(vcpu);
1628 * Radix guests can not run PR KVM or nested HV
1629 * hash guests which might run PR KVM, so this
1630 * is always a privilege fault. Send a program
1631 * check to guest kernel.
1633 kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1640 * hcall - gather args and set exit_reason. This will next be
1641 * handled by kvmppc_pseries_do_hcall which may be able to deal
1642 * with it and resume guest, or may punt to userspace.
1644 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1645 for (i = 0; i < 9; ++i)
1646 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1647 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1648 vcpu->arch.hcall_needed = 1;
1653 * We get these next two if the guest accesses a page which it thinks
1654 * it has mapped but which is not actually present, either because
1655 * it is for an emulated I/O device or because the corresonding
1656 * host page has been paged out.
1658 * Any other HDSI/HISI interrupts have been handled already for P7/8
1659 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1660 * fault handling is done here.
1662 case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1666 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1667 unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1668 r = RESUME_GUEST; /* Just retry if it's the canary */
1672 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1674 * Radix doesn't require anything, and pre-ISAv3.0 hash
1675 * already attempted to handle this in rmhandlers. The
1676 * hash fault handling below is v3 only (it uses ASDR
1679 r = RESUME_PAGE_FAULT;
1683 if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1684 kvmppc_core_queue_data_storage(vcpu,
1685 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1690 if (!(vcpu->arch.shregs.msr & MSR_DR))
1691 vsid = vcpu->kvm->arch.vrma_slb_v;
1693 vsid = vcpu->arch.fault_gpa;
1695 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1696 vsid, vcpu->arch.fault_dsisr, true);
1699 } else if (err == -1 || err == -2) {
1700 r = RESUME_PAGE_FAULT;
1702 kvmppc_core_queue_data_storage(vcpu,
1703 vcpu->arch.fault_dar, err);
1708 case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1712 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1713 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1714 DSISR_SRR1_MATCH_64S;
1715 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1717 * Radix doesn't require anything, and pre-ISAv3.0 hash
1718 * already attempted to handle this in rmhandlers. The
1719 * hash fault handling below is v3 only (it uses ASDR
1722 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1723 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1724 r = RESUME_PAGE_FAULT;
1728 if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1729 kvmppc_core_queue_inst_storage(vcpu,
1730 vcpu->arch.fault_dsisr);
1735 if (!(vcpu->arch.shregs.msr & MSR_IR))
1736 vsid = vcpu->kvm->arch.vrma_slb_v;
1738 vsid = vcpu->arch.fault_gpa;
1740 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1741 vsid, vcpu->arch.fault_dsisr, false);
1744 } else if (err == -1) {
1745 r = RESUME_PAGE_FAULT;
1747 kvmppc_core_queue_inst_storage(vcpu, err);
1754 * This occurs if the guest executes an illegal instruction.
1755 * If the guest debug is disabled, generate a program interrupt
1756 * to the guest. If guest debug is enabled, we need to check
1757 * whether the instruction is a software breakpoint instruction.
1758 * Accordingly return to Guest or Host.
1760 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1761 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1762 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1763 swab32(vcpu->arch.emul_inst) :
1764 vcpu->arch.emul_inst;
1765 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1766 r = kvmppc_emulate_debug_inst(vcpu);
1768 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1773 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1774 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1776 * This occurs for various TM-related instructions that
1777 * we need to emulate on POWER9 DD2.2. We have already
1778 * handled the cases where the guest was in real-suspend
1779 * mode and was transitioning to transactional state.
1781 r = kvmhv_p9_tm_emulation(vcpu);
1784 fallthrough; /* go to facility unavailable handler */
1788 * This occurs if the guest (kernel or userspace), does something that
1789 * is prohibited by HFSCR.
1790 * On POWER9, this could be a doorbell instruction that we need
1792 * Otherwise, we just generate a program interrupt to the guest.
1794 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1795 u64 cause = vcpu->arch.hfscr >> 56;
1798 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1799 if (cause == FSCR_MSGP_LG)
1800 r = kvmppc_emulate_doorbell_instr(vcpu);
1801 if (cause == FSCR_PM_LG)
1802 r = kvmppc_pmu_unavailable(vcpu);
1803 if (cause == FSCR_EBB_LG)
1804 r = kvmppc_ebb_unavailable(vcpu);
1805 if (cause == FSCR_TM_LG)
1806 r = kvmppc_tm_unavailable(vcpu);
1808 if (r == EMULATE_FAIL) {
1809 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1815 case BOOK3S_INTERRUPT_HV_RM_HARD:
1816 r = RESUME_PASSTHROUGH;
1819 kvmppc_dump_regs(vcpu);
1820 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1821 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1822 vcpu->arch.shregs.msr);
1823 run->hw.hardware_exit_reason = vcpu->arch.trap;
1831 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1836 vcpu->stat.sum_exits++;
1839 * This can happen if an interrupt occurs in the last stages
1840 * of guest entry or the first stages of guest exit (i.e. after
1841 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1842 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1843 * That can happen due to a bug, or due to a machine check
1844 * occurring at just the wrong time.
1846 if (vcpu->arch.shregs.msr & MSR_HV) {
1847 pr_emerg("KVM trap in HV mode while nested!\n");
1848 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1849 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1850 vcpu->arch.shregs.msr);
1851 kvmppc_dump_regs(vcpu);
1854 switch (vcpu->arch.trap) {
1855 /* We're good on these - the host merely wanted to get our attention */
1856 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1857 vcpu->stat.dec_exits++;
1860 case BOOK3S_INTERRUPT_EXTERNAL:
1861 vcpu->stat.ext_intr_exits++;
1864 case BOOK3S_INTERRUPT_H_DOORBELL:
1865 case BOOK3S_INTERRUPT_H_VIRT:
1866 vcpu->stat.ext_intr_exits++;
1869 /* These need to go to the nested HV */
1870 case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1871 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1872 vcpu->stat.dec_exits++;
1875 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1876 case BOOK3S_INTERRUPT_HMI:
1877 case BOOK3S_INTERRUPT_PERFMON:
1878 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1881 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1883 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1884 DEFAULT_RATELIMIT_BURST);
1885 /* Pass the machine check to the L1 guest */
1887 /* Print the MCE event to host console. */
1888 if (__ratelimit(&rs))
1889 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1893 * We get these next two if the guest accesses a page which it thinks
1894 * it has mapped but which is not actually present, either because
1895 * it is for an emulated I/O device or because the corresonding
1896 * host page has been paged out.
1898 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1899 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1900 r = kvmhv_nested_page_fault(vcpu);
1901 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1903 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1904 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1905 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1906 DSISR_SRR1_MATCH_64S;
1907 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1908 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1909 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1910 r = kvmhv_nested_page_fault(vcpu);
1911 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1914 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1915 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1917 * This occurs for various TM-related instructions that
1918 * we need to emulate on POWER9 DD2.2. We have already
1919 * handled the cases where the guest was in real-suspend
1920 * mode and was transitioning to transactional state.
1922 r = kvmhv_p9_tm_emulation(vcpu);
1925 fallthrough; /* go to facility unavailable handler */
1928 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1929 u64 cause = vcpu->arch.hfscr >> 56;
1932 * Only pass HFU interrupts to the L1 if the facility is
1933 * permitted but disabled by the L1's HFSCR, otherwise
1934 * the interrupt does not make sense to the L1 so turn
1937 if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) ||
1938 (vcpu->arch.nested_hfscr & (1UL << cause))) {
1939 vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST;
1942 * If the fetch failed, return to guest and
1943 * try executing it again.
1945 r = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1946 &vcpu->arch.emul_inst);
1947 if (r != EMULATE_DONE)
1958 case BOOK3S_INTERRUPT_HV_RM_HARD:
1959 vcpu->arch.trap = 0;
1961 if (!xics_on_xive())
1962 kvmppc_xics_rm_complete(vcpu, 0);
1964 case BOOK3S_INTERRUPT_SYSCALL:
1966 unsigned long req = kvmppc_get_gpr(vcpu, 3);
1969 * The H_RPT_INVALIDATE hcalls issued by nested
1970 * guests for process-scoped invalidations when
1971 * GTSE=0, are handled here in L0.
1973 if (req == H_RPT_INVALIDATE) {
1974 r = kvmppc_nested_h_rpt_invalidate(vcpu);
1989 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1990 struct kvm_sregs *sregs)
1994 memset(sregs, 0, sizeof(struct kvm_sregs));
1995 sregs->pvr = vcpu->arch.pvr;
1996 for (i = 0; i < vcpu->arch.slb_max; i++) {
1997 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1998 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
2004 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
2005 struct kvm_sregs *sregs)
2009 /* Only accept the same PVR as the host's, since we can't spoof it */
2010 if (sregs->pvr != vcpu->arch.pvr)
2014 for (i = 0; i < vcpu->arch.slb_nr; i++) {
2015 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2016 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2017 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2021 vcpu->arch.slb_max = j;
2027 * Enforce limits on guest LPCR values based on hardware availability,
2028 * guest configuration, and possibly hypervisor support and security
2031 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2033 /* LPCR_TC only applies to HPT guests */
2034 if (kvm_is_radix(kvm))
2037 /* On POWER8 and above, userspace can modify AIL */
2038 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2040 if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2041 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2043 * On some POWER9s we force AIL off for radix guests to prevent
2044 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2045 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2046 * be cached, which the host TLB management does not expect.
2048 if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2052 * On POWER9, allow userspace to enable large decrementer for the
2053 * guest, whether or not the host has it enabled.
2055 if (!cpu_has_feature(CPU_FTR_ARCH_300))
2061 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2063 if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2064 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2065 lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2069 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2070 bool preserve_top32)
2072 struct kvm *kvm = vcpu->kvm;
2073 struct kvmppc_vcore *vc = vcpu->arch.vcore;
2076 spin_lock(&vc->lock);
2079 * Userspace can only modify
2080 * DPFD (default prefetch depth), ILE (interrupt little-endian),
2081 * TC (translation control), AIL (alternate interrupt location),
2082 * LD (large decrementer).
2083 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2085 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2087 /* Broken 32-bit version of LPCR must not clear top bits */
2091 new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2092 (vc->lpcr & ~mask) | (new_lpcr & mask));
2095 * If ILE (interrupt little-endian) has changed, update the
2096 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2098 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2099 struct kvm_vcpu *vcpu;
2102 kvm_for_each_vcpu(i, vcpu, kvm) {
2103 if (vcpu->arch.vcore != vc)
2105 if (new_lpcr & LPCR_ILE)
2106 vcpu->arch.intr_msr |= MSR_LE;
2108 vcpu->arch.intr_msr &= ~MSR_LE;
2112 vc->lpcr = new_lpcr;
2114 spin_unlock(&vc->lock);
2117 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2118 union kvmppc_one_reg *val)
2124 case KVM_REG_PPC_DEBUG_INST:
2125 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2127 case KVM_REG_PPC_HIOR:
2128 *val = get_reg_val(id, 0);
2130 case KVM_REG_PPC_DABR:
2131 *val = get_reg_val(id, vcpu->arch.dabr);
2133 case KVM_REG_PPC_DABRX:
2134 *val = get_reg_val(id, vcpu->arch.dabrx);
2136 case KVM_REG_PPC_DSCR:
2137 *val = get_reg_val(id, vcpu->arch.dscr);
2139 case KVM_REG_PPC_PURR:
2140 *val = get_reg_val(id, vcpu->arch.purr);
2142 case KVM_REG_PPC_SPURR:
2143 *val = get_reg_val(id, vcpu->arch.spurr);
2145 case KVM_REG_PPC_AMR:
2146 *val = get_reg_val(id, vcpu->arch.amr);
2148 case KVM_REG_PPC_UAMOR:
2149 *val = get_reg_val(id, vcpu->arch.uamor);
2151 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2152 i = id - KVM_REG_PPC_MMCR0;
2153 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
2155 case KVM_REG_PPC_MMCR2:
2156 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
2158 case KVM_REG_PPC_MMCRA:
2159 *val = get_reg_val(id, vcpu->arch.mmcra);
2161 case KVM_REG_PPC_MMCRS:
2162 *val = get_reg_val(id, vcpu->arch.mmcrs);
2164 case KVM_REG_PPC_MMCR3:
2165 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2167 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2168 i = id - KVM_REG_PPC_PMC1;
2169 *val = get_reg_val(id, vcpu->arch.pmc[i]);
2171 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2172 i = id - KVM_REG_PPC_SPMC1;
2173 *val = get_reg_val(id, vcpu->arch.spmc[i]);
2175 case KVM_REG_PPC_SIAR:
2176 *val = get_reg_val(id, vcpu->arch.siar);
2178 case KVM_REG_PPC_SDAR:
2179 *val = get_reg_val(id, vcpu->arch.sdar);
2181 case KVM_REG_PPC_SIER:
2182 *val = get_reg_val(id, vcpu->arch.sier[0]);
2184 case KVM_REG_PPC_SIER2:
2185 *val = get_reg_val(id, vcpu->arch.sier[1]);
2187 case KVM_REG_PPC_SIER3:
2188 *val = get_reg_val(id, vcpu->arch.sier[2]);
2190 case KVM_REG_PPC_IAMR:
2191 *val = get_reg_val(id, vcpu->arch.iamr);
2193 case KVM_REG_PPC_PSPB:
2194 *val = get_reg_val(id, vcpu->arch.pspb);
2196 case KVM_REG_PPC_DPDES:
2198 * On POWER9, where we are emulating msgsndp etc.,
2199 * we return 1 bit for each vcpu, which can come from
2200 * either vcore->dpdes or doorbell_request.
2201 * On POWER8, doorbell_request is 0.
2203 if (cpu_has_feature(CPU_FTR_ARCH_300))
2204 *val = get_reg_val(id, vcpu->arch.doorbell_request);
2206 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2208 case KVM_REG_PPC_VTB:
2209 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
2211 case KVM_REG_PPC_DAWR:
2212 *val = get_reg_val(id, vcpu->arch.dawr0);
2214 case KVM_REG_PPC_DAWRX:
2215 *val = get_reg_val(id, vcpu->arch.dawrx0);
2217 case KVM_REG_PPC_DAWR1:
2218 *val = get_reg_val(id, vcpu->arch.dawr1);
2220 case KVM_REG_PPC_DAWRX1:
2221 *val = get_reg_val(id, vcpu->arch.dawrx1);
2223 case KVM_REG_PPC_CIABR:
2224 *val = get_reg_val(id, vcpu->arch.ciabr);
2226 case KVM_REG_PPC_CSIGR:
2227 *val = get_reg_val(id, vcpu->arch.csigr);
2229 case KVM_REG_PPC_TACR:
2230 *val = get_reg_val(id, vcpu->arch.tacr);
2232 case KVM_REG_PPC_TCSCR:
2233 *val = get_reg_val(id, vcpu->arch.tcscr);
2235 case KVM_REG_PPC_PID:
2236 *val = get_reg_val(id, vcpu->arch.pid);
2238 case KVM_REG_PPC_ACOP:
2239 *val = get_reg_val(id, vcpu->arch.acop);
2241 case KVM_REG_PPC_WORT:
2242 *val = get_reg_val(id, vcpu->arch.wort);
2244 case KVM_REG_PPC_TIDR:
2245 *val = get_reg_val(id, vcpu->arch.tid);
2247 case KVM_REG_PPC_PSSCR:
2248 *val = get_reg_val(id, vcpu->arch.psscr);
2250 case KVM_REG_PPC_VPA_ADDR:
2251 spin_lock(&vcpu->arch.vpa_update_lock);
2252 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2253 spin_unlock(&vcpu->arch.vpa_update_lock);
2255 case KVM_REG_PPC_VPA_SLB:
2256 spin_lock(&vcpu->arch.vpa_update_lock);
2257 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2258 val->vpaval.length = vcpu->arch.slb_shadow.len;
2259 spin_unlock(&vcpu->arch.vpa_update_lock);
2261 case KVM_REG_PPC_VPA_DTL:
2262 spin_lock(&vcpu->arch.vpa_update_lock);
2263 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2264 val->vpaval.length = vcpu->arch.dtl.len;
2265 spin_unlock(&vcpu->arch.vpa_update_lock);
2267 case KVM_REG_PPC_TB_OFFSET:
2268 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
2270 case KVM_REG_PPC_LPCR:
2271 case KVM_REG_PPC_LPCR_64:
2272 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
2274 case KVM_REG_PPC_PPR:
2275 *val = get_reg_val(id, vcpu->arch.ppr);
2277 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2278 case KVM_REG_PPC_TFHAR:
2279 *val = get_reg_val(id, vcpu->arch.tfhar);
2281 case KVM_REG_PPC_TFIAR:
2282 *val = get_reg_val(id, vcpu->arch.tfiar);
2284 case KVM_REG_PPC_TEXASR:
2285 *val = get_reg_val(id, vcpu->arch.texasr);
2287 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2288 i = id - KVM_REG_PPC_TM_GPR0;
2289 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2291 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2294 i = id - KVM_REG_PPC_TM_VSR0;
2296 for (j = 0; j < TS_FPRWIDTH; j++)
2297 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2299 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2300 val->vval = vcpu->arch.vr_tm.vr[i-32];
2306 case KVM_REG_PPC_TM_CR:
2307 *val = get_reg_val(id, vcpu->arch.cr_tm);
2309 case KVM_REG_PPC_TM_XER:
2310 *val = get_reg_val(id, vcpu->arch.xer_tm);
2312 case KVM_REG_PPC_TM_LR:
2313 *val = get_reg_val(id, vcpu->arch.lr_tm);
2315 case KVM_REG_PPC_TM_CTR:
2316 *val = get_reg_val(id, vcpu->arch.ctr_tm);
2318 case KVM_REG_PPC_TM_FPSCR:
2319 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2321 case KVM_REG_PPC_TM_AMR:
2322 *val = get_reg_val(id, vcpu->arch.amr_tm);
2324 case KVM_REG_PPC_TM_PPR:
2325 *val = get_reg_val(id, vcpu->arch.ppr_tm);
2327 case KVM_REG_PPC_TM_VRSAVE:
2328 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
2330 case KVM_REG_PPC_TM_VSCR:
2331 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2332 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2336 case KVM_REG_PPC_TM_DSCR:
2337 *val = get_reg_val(id, vcpu->arch.dscr_tm);
2339 case KVM_REG_PPC_TM_TAR:
2340 *val = get_reg_val(id, vcpu->arch.tar_tm);
2343 case KVM_REG_PPC_ARCH_COMPAT:
2344 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
2346 case KVM_REG_PPC_DEC_EXPIRY:
2347 *val = get_reg_val(id, vcpu->arch.dec_expires);
2349 case KVM_REG_PPC_ONLINE:
2350 *val = get_reg_val(id, vcpu->arch.online);
2352 case KVM_REG_PPC_PTCR:
2353 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2363 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2364 union kvmppc_one_reg *val)
2368 unsigned long addr, len;
2371 case KVM_REG_PPC_HIOR:
2372 /* Only allow this to be set to zero */
2373 if (set_reg_val(id, *val))
2376 case KVM_REG_PPC_DABR:
2377 vcpu->arch.dabr = set_reg_val(id, *val);
2379 case KVM_REG_PPC_DABRX:
2380 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2382 case KVM_REG_PPC_DSCR:
2383 vcpu->arch.dscr = set_reg_val(id, *val);
2385 case KVM_REG_PPC_PURR:
2386 vcpu->arch.purr = set_reg_val(id, *val);
2388 case KVM_REG_PPC_SPURR:
2389 vcpu->arch.spurr = set_reg_val(id, *val);
2391 case KVM_REG_PPC_AMR:
2392 vcpu->arch.amr = set_reg_val(id, *val);
2394 case KVM_REG_PPC_UAMOR:
2395 vcpu->arch.uamor = set_reg_val(id, *val);
2397 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2398 i = id - KVM_REG_PPC_MMCR0;
2399 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2401 case KVM_REG_PPC_MMCR2:
2402 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2404 case KVM_REG_PPC_MMCRA:
2405 vcpu->arch.mmcra = set_reg_val(id, *val);
2407 case KVM_REG_PPC_MMCRS:
2408 vcpu->arch.mmcrs = set_reg_val(id, *val);
2410 case KVM_REG_PPC_MMCR3:
2411 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2413 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2414 i = id - KVM_REG_PPC_PMC1;
2415 vcpu->arch.pmc[i] = set_reg_val(id, *val);
2417 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2418 i = id - KVM_REG_PPC_SPMC1;
2419 vcpu->arch.spmc[i] = set_reg_val(id, *val);
2421 case KVM_REG_PPC_SIAR:
2422 vcpu->arch.siar = set_reg_val(id, *val);
2424 case KVM_REG_PPC_SDAR:
2425 vcpu->arch.sdar = set_reg_val(id, *val);
2427 case KVM_REG_PPC_SIER:
2428 vcpu->arch.sier[0] = set_reg_val(id, *val);
2430 case KVM_REG_PPC_SIER2:
2431 vcpu->arch.sier[1] = set_reg_val(id, *val);
2433 case KVM_REG_PPC_SIER3:
2434 vcpu->arch.sier[2] = set_reg_val(id, *val);
2436 case KVM_REG_PPC_IAMR:
2437 vcpu->arch.iamr = set_reg_val(id, *val);
2439 case KVM_REG_PPC_PSPB:
2440 vcpu->arch.pspb = set_reg_val(id, *val);
2442 case KVM_REG_PPC_DPDES:
2443 if (cpu_has_feature(CPU_FTR_ARCH_300))
2444 vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2446 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2448 case KVM_REG_PPC_VTB:
2449 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2451 case KVM_REG_PPC_DAWR:
2452 vcpu->arch.dawr0 = set_reg_val(id, *val);
2454 case KVM_REG_PPC_DAWRX:
2455 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2457 case KVM_REG_PPC_DAWR1:
2458 vcpu->arch.dawr1 = set_reg_val(id, *val);
2460 case KVM_REG_PPC_DAWRX1:
2461 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2463 case KVM_REG_PPC_CIABR:
2464 vcpu->arch.ciabr = set_reg_val(id, *val);
2465 /* Don't allow setting breakpoints in hypervisor code */
2466 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2467 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
2469 case KVM_REG_PPC_CSIGR:
2470 vcpu->arch.csigr = set_reg_val(id, *val);
2472 case KVM_REG_PPC_TACR:
2473 vcpu->arch.tacr = set_reg_val(id, *val);
2475 case KVM_REG_PPC_TCSCR:
2476 vcpu->arch.tcscr = set_reg_val(id, *val);
2478 case KVM_REG_PPC_PID:
2479 vcpu->arch.pid = set_reg_val(id, *val);
2481 case KVM_REG_PPC_ACOP:
2482 vcpu->arch.acop = set_reg_val(id, *val);
2484 case KVM_REG_PPC_WORT:
2485 vcpu->arch.wort = set_reg_val(id, *val);
2487 case KVM_REG_PPC_TIDR:
2488 vcpu->arch.tid = set_reg_val(id, *val);
2490 case KVM_REG_PPC_PSSCR:
2491 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2493 case KVM_REG_PPC_VPA_ADDR:
2494 addr = set_reg_val(id, *val);
2496 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2497 vcpu->arch.dtl.next_gpa))
2499 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2501 case KVM_REG_PPC_VPA_SLB:
2502 addr = val->vpaval.addr;
2503 len = val->vpaval.length;
2505 if (addr && !vcpu->arch.vpa.next_gpa)
2507 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2509 case KVM_REG_PPC_VPA_DTL:
2510 addr = val->vpaval.addr;
2511 len = val->vpaval.length;
2513 if (addr && (len < sizeof(struct dtl_entry) ||
2514 !vcpu->arch.vpa.next_gpa))
2516 len -= len % sizeof(struct dtl_entry);
2517 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2519 case KVM_REG_PPC_TB_OFFSET:
2520 /* round up to multiple of 2^24 */
2521 vcpu->arch.vcore->tb_offset =
2522 ALIGN(set_reg_val(id, *val), 1UL << 24);
2524 case KVM_REG_PPC_LPCR:
2525 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2527 case KVM_REG_PPC_LPCR_64:
2528 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2530 case KVM_REG_PPC_PPR:
2531 vcpu->arch.ppr = set_reg_val(id, *val);
2533 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2534 case KVM_REG_PPC_TFHAR:
2535 vcpu->arch.tfhar = set_reg_val(id, *val);
2537 case KVM_REG_PPC_TFIAR:
2538 vcpu->arch.tfiar = set_reg_val(id, *val);
2540 case KVM_REG_PPC_TEXASR:
2541 vcpu->arch.texasr = set_reg_val(id, *val);
2543 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2544 i = id - KVM_REG_PPC_TM_GPR0;
2545 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2547 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2550 i = id - KVM_REG_PPC_TM_VSR0;
2552 for (j = 0; j < TS_FPRWIDTH; j++)
2553 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2555 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2556 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2561 case KVM_REG_PPC_TM_CR:
2562 vcpu->arch.cr_tm = set_reg_val(id, *val);
2564 case KVM_REG_PPC_TM_XER:
2565 vcpu->arch.xer_tm = set_reg_val(id, *val);
2567 case KVM_REG_PPC_TM_LR:
2568 vcpu->arch.lr_tm = set_reg_val(id, *val);
2570 case KVM_REG_PPC_TM_CTR:
2571 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2573 case KVM_REG_PPC_TM_FPSCR:
2574 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2576 case KVM_REG_PPC_TM_AMR:
2577 vcpu->arch.amr_tm = set_reg_val(id, *val);
2579 case KVM_REG_PPC_TM_PPR:
2580 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2582 case KVM_REG_PPC_TM_VRSAVE:
2583 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2585 case KVM_REG_PPC_TM_VSCR:
2586 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2587 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2591 case KVM_REG_PPC_TM_DSCR:
2592 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2594 case KVM_REG_PPC_TM_TAR:
2595 vcpu->arch.tar_tm = set_reg_val(id, *val);
2598 case KVM_REG_PPC_ARCH_COMPAT:
2599 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2601 case KVM_REG_PPC_DEC_EXPIRY:
2602 vcpu->arch.dec_expires = set_reg_val(id, *val);
2604 case KVM_REG_PPC_ONLINE:
2605 i = set_reg_val(id, *val);
2606 if (i && !vcpu->arch.online)
2607 atomic_inc(&vcpu->arch.vcore->online_count);
2608 else if (!i && vcpu->arch.online)
2609 atomic_dec(&vcpu->arch.vcore->online_count);
2610 vcpu->arch.online = i;
2612 case KVM_REG_PPC_PTCR:
2613 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2624 * On POWER9, threads are independent and can be in different partitions.
2625 * Therefore we consider each thread to be a subcore.
2626 * There is a restriction that all threads have to be in the same
2627 * MMU mode (radix or HPT), unfortunately, but since we only support
2628 * HPT guests on a HPT host so far, that isn't an impediment yet.
2630 static int threads_per_vcore(struct kvm *kvm)
2632 if (cpu_has_feature(CPU_FTR_ARCH_300))
2634 return threads_per_subcore;
2637 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2639 struct kvmppc_vcore *vcore;
2641 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2646 spin_lock_init(&vcore->lock);
2647 spin_lock_init(&vcore->stoltb_lock);
2648 rcuwait_init(&vcore->wait);
2649 vcore->preempt_tb = TB_NIL;
2650 vcore->lpcr = kvm->arch.lpcr;
2651 vcore->first_vcpuid = id;
2653 INIT_LIST_HEAD(&vcore->preempt_list);
2658 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2659 static struct debugfs_timings_element {
2663 #ifdef CONFIG_KVM_BOOK3S_HV_P9_TIMING
2664 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2665 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2666 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2667 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2668 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2670 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2671 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2672 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2673 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2674 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2678 #define N_TIMINGS (ARRAY_SIZE(timings))
2680 struct debugfs_timings_state {
2681 struct kvm_vcpu *vcpu;
2682 unsigned int buflen;
2683 char buf[N_TIMINGS * 100];
2686 static int debugfs_timings_open(struct inode *inode, struct file *file)
2688 struct kvm_vcpu *vcpu = inode->i_private;
2689 struct debugfs_timings_state *p;
2691 p = kzalloc(sizeof(*p), GFP_KERNEL);
2695 kvm_get_kvm(vcpu->kvm);
2697 file->private_data = p;
2699 return nonseekable_open(inode, file);
2702 static int debugfs_timings_release(struct inode *inode, struct file *file)
2704 struct debugfs_timings_state *p = file->private_data;
2706 kvm_put_kvm(p->vcpu->kvm);
2711 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2712 size_t len, loff_t *ppos)
2714 struct debugfs_timings_state *p = file->private_data;
2715 struct kvm_vcpu *vcpu = p->vcpu;
2717 struct kvmhv_tb_accumulator tb;
2726 buf_end = s + sizeof(p->buf);
2727 for (i = 0; i < N_TIMINGS; ++i) {
2728 struct kvmhv_tb_accumulator *acc;
2730 acc = (struct kvmhv_tb_accumulator *)
2731 ((unsigned long)vcpu + timings[i].offset);
2733 for (loops = 0; loops < 1000; ++loops) {
2734 count = acc->seqcount;
2739 if (count == acc->seqcount) {
2747 snprintf(s, buf_end - s, "%s: stuck\n",
2750 snprintf(s, buf_end - s,
2751 "%s: %llu %llu %llu %llu\n",
2752 timings[i].name, count / 2,
2753 tb_to_ns(tb.tb_total),
2754 tb_to_ns(tb.tb_min),
2755 tb_to_ns(tb.tb_max));
2758 p->buflen = s - p->buf;
2762 if (pos >= p->buflen)
2764 if (len > p->buflen - pos)
2765 len = p->buflen - pos;
2766 n = copy_to_user(buf, p->buf + pos, len);
2776 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2777 size_t len, loff_t *ppos)
2782 static const struct file_operations debugfs_timings_ops = {
2783 .owner = THIS_MODULE,
2784 .open = debugfs_timings_open,
2785 .release = debugfs_timings_release,
2786 .read = debugfs_timings_read,
2787 .write = debugfs_timings_write,
2788 .llseek = generic_file_llseek,
2791 /* Create a debugfs directory for the vcpu */
2792 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2794 if (cpu_has_feature(CPU_FTR_ARCH_300) == IS_ENABLED(CONFIG_KVM_BOOK3S_HV_P9_TIMING))
2795 debugfs_create_file("timings", 0444, debugfs_dentry, vcpu,
2796 &debugfs_timings_ops);
2800 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2801 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2805 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2807 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2811 struct kvmppc_vcore *vcore;
2818 vcpu->arch.shared = &vcpu->arch.shregs;
2819 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2821 * The shared struct is never shared on HV,
2822 * so we can always use host endianness
2824 #ifdef __BIG_ENDIAN__
2825 vcpu->arch.shared_big_endian = true;
2827 vcpu->arch.shared_big_endian = false;
2830 vcpu->arch.mmcr[0] = MMCR0_FC;
2831 if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2832 vcpu->arch.mmcr[0] |= MMCR0_PMCCEXT;
2833 vcpu->arch.mmcra = MMCRA_BHRB_DISABLE;
2836 vcpu->arch.ctrl = CTRL_RUNLATCH;
2837 /* default to host PVR, since we can't spoof it */
2838 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2839 spin_lock_init(&vcpu->arch.vpa_update_lock);
2840 spin_lock_init(&vcpu->arch.tbacct_lock);
2841 vcpu->arch.busy_preempt = TB_NIL;
2842 vcpu->arch.shregs.msr = MSR_ME;
2843 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2846 * Set the default HFSCR for the guest from the host value.
2847 * This value is only used on POWER9.
2848 * On POWER9, we want to virtualize the doorbell facility, so we
2849 * don't set the HFSCR_MSGP bit, and that causes those instructions
2850 * to trap and then we emulate them.
2852 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2853 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2854 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2855 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2856 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2857 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2858 vcpu->arch.hfscr |= HFSCR_TM;
2861 if (cpu_has_feature(CPU_FTR_TM_COMP))
2862 vcpu->arch.hfscr |= HFSCR_TM;
2864 vcpu->arch.hfscr_permitted = vcpu->arch.hfscr;
2867 * PM, EBB, TM are demand-faulted so start with it clear.
2869 vcpu->arch.hfscr &= ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM);
2871 kvmppc_mmu_book3s_hv_init(vcpu);
2873 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2875 init_waitqueue_head(&vcpu->arch.cpu_run);
2877 mutex_lock(&kvm->lock);
2880 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2881 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2882 pr_devel("KVM: VCPU ID too high\n");
2883 core = KVM_MAX_VCORES;
2885 BUG_ON(kvm->arch.smt_mode != 1);
2886 core = kvmppc_pack_vcpu_id(kvm, id);
2889 core = id / kvm->arch.smt_mode;
2891 if (core < KVM_MAX_VCORES) {
2892 vcore = kvm->arch.vcores[core];
2893 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2894 pr_devel("KVM: collision on id %u", id);
2896 } else if (!vcore) {
2898 * Take mmu_setup_lock for mutual exclusion
2899 * with kvmppc_update_lpcr().
2902 vcore = kvmppc_vcore_create(kvm,
2903 id & ~(kvm->arch.smt_mode - 1));
2904 mutex_lock(&kvm->arch.mmu_setup_lock);
2905 kvm->arch.vcores[core] = vcore;
2906 kvm->arch.online_vcores++;
2907 mutex_unlock(&kvm->arch.mmu_setup_lock);
2910 mutex_unlock(&kvm->lock);
2915 spin_lock(&vcore->lock);
2916 ++vcore->num_threads;
2917 spin_unlock(&vcore->lock);
2918 vcpu->arch.vcore = vcore;
2919 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2920 vcpu->arch.thread_cpu = -1;
2921 vcpu->arch.prev_cpu = -1;
2923 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2924 kvmppc_sanity_check(vcpu);
2929 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2930 unsigned long flags)
2937 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2939 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2941 * On POWER8 (or POWER7), the threading mode is "strict",
2942 * so we pack smt_mode vcpus per vcore.
2944 if (smt_mode > threads_per_subcore)
2948 * On POWER9, the threading mode is "loose",
2949 * so each vcpu gets its own vcore.
2954 mutex_lock(&kvm->lock);
2956 if (!kvm->arch.online_vcores) {
2957 kvm->arch.smt_mode = smt_mode;
2958 kvm->arch.emul_smt_mode = esmt;
2961 mutex_unlock(&kvm->lock);
2966 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2968 if (vpa->pinned_addr)
2969 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2973 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2975 spin_lock(&vcpu->arch.vpa_update_lock);
2976 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2977 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2978 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2979 spin_unlock(&vcpu->arch.vpa_update_lock);
2982 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2984 /* Indicate we want to get back into the guest */
2988 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2990 unsigned long dec_nsec, now;
2993 if (now > kvmppc_dec_expires_host_tb(vcpu)) {
2994 /* decrementer has already gone negative */
2995 kvmppc_core_queue_dec(vcpu);
2996 kvmppc_core_prepare_to_enter(vcpu);
2999 dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
3000 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
3001 vcpu->arch.timer_running = 1;
3004 extern int __kvmppc_vcore_entry(void);
3006 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
3007 struct kvm_vcpu *vcpu, u64 tb)
3011 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3013 spin_lock_irq(&vcpu->arch.tbacct_lock);
3015 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
3016 vcpu->arch.stolen_logged;
3017 vcpu->arch.busy_preempt = now;
3018 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3019 spin_unlock_irq(&vcpu->arch.tbacct_lock);
3021 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3024 static int kvmppc_grab_hwthread(int cpu)
3026 struct paca_struct *tpaca;
3027 long timeout = 10000;
3029 tpaca = paca_ptrs[cpu];
3031 /* Ensure the thread won't go into the kernel if it wakes */
3032 tpaca->kvm_hstate.kvm_vcpu = NULL;
3033 tpaca->kvm_hstate.kvm_vcore = NULL;
3034 tpaca->kvm_hstate.napping = 0;
3036 tpaca->kvm_hstate.hwthread_req = 1;
3039 * If the thread is already executing in the kernel (e.g. handling
3040 * a stray interrupt), wait for it to get back to nap mode.
3041 * The smp_mb() is to ensure that our setting of hwthread_req
3042 * is visible before we look at hwthread_state, so if this
3043 * races with the code at system_reset_pSeries and the thread
3044 * misses our setting of hwthread_req, we are sure to see its
3045 * setting of hwthread_state, and vice versa.
3048 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3049 if (--timeout <= 0) {
3050 pr_err("KVM: couldn't grab cpu %d\n", cpu);
3058 static void kvmppc_release_hwthread(int cpu)
3060 struct paca_struct *tpaca;
3062 tpaca = paca_ptrs[cpu];
3063 tpaca->kvm_hstate.hwthread_req = 0;
3064 tpaca->kvm_hstate.kvm_vcpu = NULL;
3065 tpaca->kvm_hstate.kvm_vcore = NULL;
3066 tpaca->kvm_hstate.kvm_split_mode = NULL;
3069 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3071 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3073 struct kvm_nested_guest *nested = vcpu->arch.nested;
3074 cpumask_t *need_tlb_flush;
3078 need_tlb_flush = &nested->need_tlb_flush;
3080 need_tlb_flush = &kvm->arch.need_tlb_flush;
3082 cpu = cpu_first_tlb_thread_sibling(cpu);
3083 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3084 i += cpu_tlb_thread_sibling_step())
3085 cpumask_set_cpu(i, need_tlb_flush);
3088 * Make sure setting of bit in need_tlb_flush precedes testing of
3089 * cpu_in_guest. The matching barrier on the other side is hwsync
3090 * when switching to guest MMU mode, which happens between
3091 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3096 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3097 i += cpu_tlb_thread_sibling_step()) {
3098 struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3101 smp_call_function_single(i, do_nothing, NULL, 1);
3105 static void do_migrate_away_vcpu(void *arg)
3107 struct kvm_vcpu *vcpu = arg;
3108 struct kvm *kvm = vcpu->kvm;
3111 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3112 * ptesync sequence on the old CPU before migrating to a new one, in
3113 * case we interrupted the guest between a tlbie ; eieio ;
3114 * tlbsync; ptesync sequence.
3116 * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3118 if (kvm->arch.lpcr & LPCR_GTSE)
3119 asm volatile("eieio; tlbsync; ptesync");
3121 asm volatile("ptesync");
3124 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3126 struct kvm_nested_guest *nested = vcpu->arch.nested;
3127 struct kvm *kvm = vcpu->kvm;
3130 if (!cpu_has_feature(CPU_FTR_HVMODE))
3134 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3136 prev_cpu = vcpu->arch.prev_cpu;
3139 * With radix, the guest can do TLB invalidations itself,
3140 * and it could choose to use the local form (tlbiel) if
3141 * it is invalidating a translation that has only ever been
3142 * used on one vcpu. However, that doesn't mean it has
3143 * only ever been used on one physical cpu, since vcpus
3144 * can move around between pcpus. To cope with this, when
3145 * a vcpu moves from one pcpu to another, we need to tell
3146 * any vcpus running on the same core as this vcpu previously
3147 * ran to flush the TLB.
3149 if (prev_cpu != pcpu) {
3150 if (prev_cpu >= 0) {
3151 if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3152 cpu_first_tlb_thread_sibling(pcpu))
3153 radix_flush_cpu(kvm, prev_cpu, vcpu);
3155 smp_call_function_single(prev_cpu,
3156 do_migrate_away_vcpu, vcpu, 1);
3159 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3161 vcpu->arch.prev_cpu = pcpu;
3165 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3168 struct paca_struct *tpaca;
3172 if (vcpu->arch.timer_running) {
3173 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3174 vcpu->arch.timer_running = 0;
3176 cpu += vcpu->arch.ptid;
3177 vcpu->cpu = vc->pcpu;
3178 vcpu->arch.thread_cpu = cpu;
3180 tpaca = paca_ptrs[cpu];
3181 tpaca->kvm_hstate.kvm_vcpu = vcpu;
3182 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3183 tpaca->kvm_hstate.fake_suspend = 0;
3184 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3186 tpaca->kvm_hstate.kvm_vcore = vc;
3187 if (cpu != smp_processor_id())
3188 kvmppc_ipi_thread(cpu);
3191 static void kvmppc_wait_for_nap(int n_threads)
3193 int cpu = smp_processor_id();
3198 for (loops = 0; loops < 1000000; ++loops) {
3200 * Check if all threads are finished.
3201 * We set the vcore pointer when starting a thread
3202 * and the thread clears it when finished, so we look
3203 * for any threads that still have a non-NULL vcore ptr.
3205 for (i = 1; i < n_threads; ++i)
3206 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3208 if (i == n_threads) {
3215 for (i = 1; i < n_threads; ++i)
3216 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3217 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3221 * Check that we are on thread 0 and that any other threads in
3222 * this core are off-line. Then grab the threads so they can't
3225 static int on_primary_thread(void)
3227 int cpu = smp_processor_id();
3230 /* Are we on a primary subcore? */
3231 if (cpu_thread_in_subcore(cpu))
3235 while (++thr < threads_per_subcore)
3236 if (cpu_online(cpu + thr))
3239 /* Grab all hw threads so they can't go into the kernel */
3240 for (thr = 1; thr < threads_per_subcore; ++thr) {
3241 if (kvmppc_grab_hwthread(cpu + thr)) {
3242 /* Couldn't grab one; let the others go */
3244 kvmppc_release_hwthread(cpu + thr);
3245 } while (--thr > 0);
3253 * A list of virtual cores for each physical CPU.
3254 * These are vcores that could run but their runner VCPU tasks are
3255 * (or may be) preempted.
3257 struct preempted_vcore_list {
3258 struct list_head list;
3262 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3264 static void init_vcore_lists(void)
3268 for_each_possible_cpu(cpu) {
3269 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3270 spin_lock_init(&lp->lock);
3271 INIT_LIST_HEAD(&lp->list);
3275 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3277 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3279 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3281 vc->vcore_state = VCORE_PREEMPT;
3282 vc->pcpu = smp_processor_id();
3283 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3284 spin_lock(&lp->lock);
3285 list_add_tail(&vc->preempt_list, &lp->list);
3286 spin_unlock(&lp->lock);
3289 /* Start accumulating stolen time */
3290 kvmppc_core_start_stolen(vc, mftb());
3293 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3295 struct preempted_vcore_list *lp;
3297 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3299 kvmppc_core_end_stolen(vc, mftb());
3300 if (!list_empty(&vc->preempt_list)) {
3301 lp = &per_cpu(preempted_vcores, vc->pcpu);
3302 spin_lock(&lp->lock);
3303 list_del_init(&vc->preempt_list);
3304 spin_unlock(&lp->lock);
3306 vc->vcore_state = VCORE_INACTIVE;
3310 * This stores information about the virtual cores currently
3311 * assigned to a physical core.
3315 int max_subcore_threads;
3317 int subcore_threads[MAX_SUBCORES];
3318 struct kvmppc_vcore *vc[MAX_SUBCORES];
3322 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3323 * respectively in 2-way micro-threading (split-core) mode on POWER8.
3325 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3327 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3329 memset(cip, 0, sizeof(*cip));
3330 cip->n_subcores = 1;
3331 cip->max_subcore_threads = vc->num_threads;
3332 cip->total_threads = vc->num_threads;
3333 cip->subcore_threads[0] = vc->num_threads;
3337 static bool subcore_config_ok(int n_subcores, int n_threads)
3340 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3341 * split-core mode, with one thread per subcore.
3343 if (cpu_has_feature(CPU_FTR_ARCH_300))
3344 return n_subcores <= 4 && n_threads == 1;
3346 /* On POWER8, can only dynamically split if unsplit to begin with */
3347 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3349 if (n_subcores > MAX_SUBCORES)
3351 if (n_subcores > 1) {
3352 if (!(dynamic_mt_modes & 2))
3354 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3358 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3361 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3363 vc->entry_exit_map = 0;
3365 vc->napping_threads = 0;
3366 vc->conferring_threads = 0;
3367 vc->tb_offset_applied = 0;
3370 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3372 int n_threads = vc->num_threads;
3375 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3378 /* In one_vm_per_core mode, require all vcores to be from the same vm */
3379 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3382 if (n_threads < cip->max_subcore_threads)
3383 n_threads = cip->max_subcore_threads;
3384 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3386 cip->max_subcore_threads = n_threads;
3388 sub = cip->n_subcores;
3390 cip->total_threads += vc->num_threads;
3391 cip->subcore_threads[sub] = vc->num_threads;
3393 init_vcore_to_run(vc);
3394 list_del_init(&vc->preempt_list);
3400 * Work out whether it is possible to piggyback the execution of
3401 * vcore *pvc onto the execution of the other vcores described in *cip.
3403 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3406 if (cip->total_threads + pvc->num_threads > target_threads)
3409 return can_dynamic_split(pvc, cip);
3412 static void prepare_threads(struct kvmppc_vcore *vc)
3415 struct kvm_vcpu *vcpu;
3417 for_each_runnable_thread(i, vcpu, vc) {
3418 if (signal_pending(vcpu->arch.run_task))
3419 vcpu->arch.ret = -EINTR;
3420 else if (vcpu->arch.vpa.update_pending ||
3421 vcpu->arch.slb_shadow.update_pending ||
3422 vcpu->arch.dtl.update_pending)
3423 vcpu->arch.ret = RESUME_GUEST;
3426 kvmppc_remove_runnable(vc, vcpu, mftb());
3427 wake_up(&vcpu->arch.cpu_run);
3431 static void collect_piggybacks(struct core_info *cip, int target_threads)
3433 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3434 struct kvmppc_vcore *pvc, *vcnext;
3436 spin_lock(&lp->lock);
3437 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3438 if (!spin_trylock(&pvc->lock))
3440 prepare_threads(pvc);
3441 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3442 list_del_init(&pvc->preempt_list);
3443 if (pvc->runner == NULL) {
3444 pvc->vcore_state = VCORE_INACTIVE;
3445 kvmppc_core_end_stolen(pvc, mftb());
3447 spin_unlock(&pvc->lock);
3450 if (!can_piggyback(pvc, cip, target_threads)) {
3451 spin_unlock(&pvc->lock);
3454 kvmppc_core_end_stolen(pvc, mftb());
3455 pvc->vcore_state = VCORE_PIGGYBACK;
3456 if (cip->total_threads >= target_threads)
3459 spin_unlock(&lp->lock);
3462 static bool recheck_signals_and_mmu(struct core_info *cip)
3465 struct kvm_vcpu *vcpu;
3466 struct kvmppc_vcore *vc;
3468 for (sub = 0; sub < cip->n_subcores; ++sub) {
3470 if (!vc->kvm->arch.mmu_ready)
3472 for_each_runnable_thread(i, vcpu, vc)
3473 if (signal_pending(vcpu->arch.run_task))
3479 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3481 int still_running = 0, i;
3484 struct kvm_vcpu *vcpu;
3486 spin_lock(&vc->lock);
3488 for_each_runnable_thread(i, vcpu, vc) {
3490 * It's safe to unlock the vcore in the loop here, because
3491 * for_each_runnable_thread() is safe against removal of
3492 * the vcpu, and the vcore state is VCORE_EXITING here,
3493 * so any vcpus becoming runnable will have their arch.trap
3494 * set to zero and can't actually run in the guest.
3496 spin_unlock(&vc->lock);
3497 /* cancel pending dec exception if dec is positive */
3498 if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3499 kvmppc_core_pending_dec(vcpu))
3500 kvmppc_core_dequeue_dec(vcpu);
3502 trace_kvm_guest_exit(vcpu);
3505 if (vcpu->arch.trap)
3506 ret = kvmppc_handle_exit_hv(vcpu,
3507 vcpu->arch.run_task);
3509 vcpu->arch.ret = ret;
3510 vcpu->arch.trap = 0;
3512 spin_lock(&vc->lock);
3513 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3514 if (vcpu->arch.pending_exceptions)
3515 kvmppc_core_prepare_to_enter(vcpu);
3516 if (vcpu->arch.ceded)
3517 kvmppc_set_timer(vcpu);
3521 kvmppc_remove_runnable(vc, vcpu, mftb());
3522 wake_up(&vcpu->arch.cpu_run);
3526 if (still_running > 0) {
3527 kvmppc_vcore_preempt(vc);
3528 } else if (vc->runner) {
3529 vc->vcore_state = VCORE_PREEMPT;
3530 kvmppc_core_start_stolen(vc, mftb());
3532 vc->vcore_state = VCORE_INACTIVE;
3534 if (vc->n_runnable > 0 && vc->runner == NULL) {
3535 /* make sure there's a candidate runner awake */
3537 vcpu = next_runnable_thread(vc, &i);
3538 wake_up(&vcpu->arch.cpu_run);
3541 spin_unlock(&vc->lock);
3545 * Clear core from the list of active host cores as we are about to
3546 * enter the guest. Only do this if it is the primary thread of the
3547 * core (not if a subcore) that is entering the guest.
3549 static inline int kvmppc_clear_host_core(unsigned int cpu)
3553 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3556 * Memory barrier can be omitted here as we will do a smp_wmb()
3557 * later in kvmppc_start_thread and we need ensure that state is
3558 * visible to other CPUs only after we enter guest.
3560 core = cpu >> threads_shift;
3561 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3566 * Advertise this core as an active host core since we exited the guest
3567 * Only need to do this if it is the primary thread of the core that is
3570 static inline int kvmppc_set_host_core(unsigned int cpu)
3574 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3578 * Memory barrier can be omitted here because we do a spin_unlock
3579 * immediately after this which provides the memory barrier.
3581 core = cpu >> threads_shift;
3582 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3586 static void set_irq_happened(int trap)
3589 case BOOK3S_INTERRUPT_EXTERNAL:
3590 local_paca->irq_happened |= PACA_IRQ_EE;
3592 case BOOK3S_INTERRUPT_H_DOORBELL:
3593 local_paca->irq_happened |= PACA_IRQ_DBELL;
3595 case BOOK3S_INTERRUPT_HMI:
3596 local_paca->irq_happened |= PACA_IRQ_HMI;
3598 case BOOK3S_INTERRUPT_SYSTEM_RESET:
3599 replay_system_reset();
3605 * Run a set of guest threads on a physical core.
3606 * Called with vc->lock held.
3608 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3610 struct kvm_vcpu *vcpu;
3613 struct core_info core_info;
3614 struct kvmppc_vcore *pvc;
3615 struct kvm_split_mode split_info, *sip;
3616 int split, subcore_size, active;
3619 unsigned long cmd_bit, stat_bit;
3622 int controlled_threads;
3626 if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3630 * Remove from the list any threads that have a signal pending
3631 * or need a VPA update done
3633 prepare_threads(vc);
3635 /* if the runner is no longer runnable, let the caller pick a new one */
3636 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3642 init_vcore_to_run(vc);
3643 vc->preempt_tb = TB_NIL;
3646 * Number of threads that we will be controlling: the same as
3647 * the number of threads per subcore, except on POWER9,
3648 * where it's 1 because the threads are (mostly) independent.
3650 controlled_threads = threads_per_vcore(vc->kvm);
3653 * Make sure we are running on primary threads, and that secondary
3654 * threads are offline. Also check if the number of threads in this
3655 * guest are greater than the current system threads per guest.
3657 if ((controlled_threads > 1) &&
3658 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3659 for_each_runnable_thread(i, vcpu, vc) {
3660 vcpu->arch.ret = -EBUSY;
3661 kvmppc_remove_runnable(vc, vcpu, mftb());
3662 wake_up(&vcpu->arch.cpu_run);
3668 * See if we could run any other vcores on the physical core
3669 * along with this one.
3671 init_core_info(&core_info, vc);
3672 pcpu = smp_processor_id();
3673 target_threads = controlled_threads;
3674 if (target_smt_mode && target_smt_mode < target_threads)
3675 target_threads = target_smt_mode;
3676 if (vc->num_threads < target_threads)
3677 collect_piggybacks(&core_info, target_threads);
3680 * Hard-disable interrupts, and check resched flag and signals.
3681 * If we need to reschedule or deliver a signal, clean up
3682 * and return without going into the guest(s).
3683 * If the mmu_ready flag has been cleared, don't go into the
3684 * guest because that means a HPT resize operation is in progress.
3686 local_irq_disable();
3688 if (lazy_irq_pending() || need_resched() ||
3689 recheck_signals_and_mmu(&core_info)) {
3691 vc->vcore_state = VCORE_INACTIVE;
3692 /* Unlock all except the primary vcore */
3693 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3694 pvc = core_info.vc[sub];
3695 /* Put back on to the preempted vcores list */
3696 kvmppc_vcore_preempt(pvc);
3697 spin_unlock(&pvc->lock);
3699 for (i = 0; i < controlled_threads; ++i)
3700 kvmppc_release_hwthread(pcpu + i);
3704 kvmppc_clear_host_core(pcpu);
3706 /* Decide on micro-threading (split-core) mode */
3707 subcore_size = threads_per_subcore;
3708 cmd_bit = stat_bit = 0;
3709 split = core_info.n_subcores;
3711 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3715 memset(&split_info, 0, sizeof(split_info));
3716 for (sub = 0; sub < core_info.n_subcores; ++sub)
3717 split_info.vc[sub] = core_info.vc[sub];
3720 if (split == 2 && (dynamic_mt_modes & 2)) {
3721 cmd_bit = HID0_POWER8_1TO2LPAR;
3722 stat_bit = HID0_POWER8_2LPARMODE;
3725 cmd_bit = HID0_POWER8_1TO4LPAR;
3726 stat_bit = HID0_POWER8_4LPARMODE;
3728 subcore_size = MAX_SMT_THREADS / split;
3729 split_info.rpr = mfspr(SPRN_RPR);
3730 split_info.pmmar = mfspr(SPRN_PMMAR);
3731 split_info.ldbar = mfspr(SPRN_LDBAR);
3732 split_info.subcore_size = subcore_size;
3734 split_info.subcore_size = 1;
3737 /* order writes to split_info before kvm_split_mode pointer */
3741 for (thr = 0; thr < controlled_threads; ++thr) {
3742 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3744 paca->kvm_hstate.napping = 0;
3745 paca->kvm_hstate.kvm_split_mode = sip;
3748 /* Initiate micro-threading (split-core) on POWER8 if required */
3750 unsigned long hid0 = mfspr(SPRN_HID0);
3752 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3754 mtspr(SPRN_HID0, hid0);
3757 hid0 = mfspr(SPRN_HID0);
3758 if (hid0 & stat_bit)
3765 * On POWER8, set RWMR register.
3766 * Since it only affects PURR and SPURR, it doesn't affect
3767 * the host, so we don't save/restore the host value.
3770 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3771 int n_online = atomic_read(&vc->online_count);
3774 * Use the 8-thread value if we're doing split-core
3775 * or if the vcore's online count looks bogus.
3777 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3778 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3779 rwmr_val = p8_rwmr_values[n_online];
3780 mtspr(SPRN_RWMR, rwmr_val);
3783 /* Start all the threads */
3785 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3786 thr = is_power8 ? subcore_thread_map[sub] : sub;
3789 pvc = core_info.vc[sub];
3790 pvc->pcpu = pcpu + thr;
3791 for_each_runnable_thread(i, vcpu, pvc) {
3793 * XXX: is kvmppc_start_thread called too late here?
3794 * It updates vcpu->cpu and vcpu->arch.thread_cpu
3795 * which are used by kvmppc_fast_vcpu_kick_hv(), but
3796 * kick is called after new exceptions become available
3797 * and exceptions are checked earlier than here, by
3798 * kvmppc_core_prepare_to_enter.
3800 kvmppc_start_thread(vcpu, pvc);
3801 kvmppc_create_dtl_entry(vcpu, pvc);
3802 trace_kvm_guest_enter(vcpu);
3803 if (!vcpu->arch.ptid)
3805 active |= 1 << (thr + vcpu->arch.ptid);
3808 * We need to start the first thread of each subcore
3809 * even if it doesn't have a vcpu.
3812 kvmppc_start_thread(NULL, pvc);
3816 * Ensure that split_info.do_nap is set after setting
3817 * the vcore pointer in the PACA of the secondaries.
3822 * When doing micro-threading, poke the inactive threads as well.
3823 * This gets them to the nap instruction after kvm_do_nap,
3824 * which reduces the time taken to unsplit later.
3827 split_info.do_nap = 1; /* ask secondaries to nap when done */
3828 for (thr = 1; thr < threads_per_subcore; ++thr)
3829 if (!(active & (1 << thr)))
3830 kvmppc_ipi_thread(pcpu + thr);
3833 vc->vcore_state = VCORE_RUNNING;
3836 trace_kvmppc_run_core(vc, 0);
3838 for (sub = 0; sub < core_info.n_subcores; ++sub)
3839 spin_unlock(&core_info.vc[sub]->lock);
3841 guest_enter_irqoff();
3843 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3845 this_cpu_disable_ftrace();
3848 * Interrupts will be enabled once we get into the guest,
3849 * so tell lockdep that we're about to enable interrupts.
3851 trace_hardirqs_on();
3853 trap = __kvmppc_vcore_entry();
3855 trace_hardirqs_off();
3857 this_cpu_enable_ftrace();
3859 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3861 set_irq_happened(trap);
3863 spin_lock(&vc->lock);
3864 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3865 vc->vcore_state = VCORE_EXITING;
3867 /* wait for secondary threads to finish writing their state to memory */
3868 kvmppc_wait_for_nap(controlled_threads);
3870 /* Return to whole-core mode if we split the core earlier */
3872 unsigned long hid0 = mfspr(SPRN_HID0);
3873 unsigned long loops = 0;
3875 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3876 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3878 mtspr(SPRN_HID0, hid0);
3881 hid0 = mfspr(SPRN_HID0);
3882 if (!(hid0 & stat_bit))
3887 split_info.do_nap = 0;
3890 kvmppc_set_host_core(pcpu);
3892 context_tracking_guest_exit();
3893 if (!vtime_accounting_enabled_this_cpu()) {
3896 * Service IRQs here before vtime_account_guest_exit() so any
3897 * ticks that occurred while running the guest are accounted to
3898 * the guest. If vtime accounting is enabled, accounting uses
3899 * TB rather than ticks, so it can be done without enabling
3900 * interrupts here, which has the problem that it accounts
3901 * interrupt processing overhead to the host.
3903 local_irq_disable();
3905 vtime_account_guest_exit();
3909 /* Let secondaries go back to the offline loop */
3910 for (i = 0; i < controlled_threads; ++i) {
3911 kvmppc_release_hwthread(pcpu + i);
3912 if (sip && sip->napped[i])
3913 kvmppc_ipi_thread(pcpu + i);
3916 spin_unlock(&vc->lock);
3918 /* make sure updates to secondary vcpu structs are visible now */
3923 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3924 pvc = core_info.vc[sub];
3925 post_guest_process(pvc, pvc == vc);
3928 spin_lock(&vc->lock);
3931 vc->vcore_state = VCORE_INACTIVE;
3932 trace_kvmppc_run_core(vc, 1);
3935 static inline bool hcall_is_xics(unsigned long req)
3937 return req == H_EOI || req == H_CPPR || req == H_IPI ||
3938 req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
3941 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu)
3943 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3945 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3946 lp->yield_count = cpu_to_be32(yield_count);
3947 vcpu->arch.vpa.dirty = 1;
3951 /* call our hypervisor to load up HV regs and go */
3952 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
3954 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3955 unsigned long host_psscr;
3957 struct hv_guest_state hvregs;
3958 struct p9_host_os_sprs host_os_sprs;
3964 save_p9_host_os_sprs(&host_os_sprs);
3967 * We need to save and restore the guest visible part of the
3968 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3969 * doesn't do this for us. Note only required if pseries since
3970 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
3972 host_psscr = mfspr(SPRN_PSSCR_PR);
3974 kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
3975 if (lazy_irq_pending())
3978 if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
3979 msr = mfmsr(); /* TM restore can update msr */
3981 if (vcpu->arch.psscr != host_psscr)
3982 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3984 kvmhv_save_hv_regs(vcpu, &hvregs);
3987 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3988 hvregs.version = HV_GUEST_STATE_VERSION;
3989 if (vcpu->arch.nested) {
3990 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3991 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3993 hvregs.lpid = vcpu->kvm->arch.lpid;
3994 hvregs.vcpu_token = vcpu->vcpu_id;
3996 hvregs.hdec_expiry = time_limit;
3999 * When setting DEC, we must always deal with irq_work_raise
4000 * via NMI vs setting DEC. The problem occurs right as we
4001 * switch into guest mode if a NMI hits and sets pending work
4002 * and sets DEC, then that will apply to the guest and not
4003 * bring us back to the host.
4005 * irq_work_raise could check a flag (or possibly LPCR[HDICE]
4006 * for example) and set HDEC to 1? That wouldn't solve the
4007 * nested hv case which needs to abort the hcall or zero the
4010 * XXX: Another day's problem.
4012 mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
4014 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
4015 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
4016 switch_pmu_to_guest(vcpu, &host_os_sprs);
4017 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
4018 __pa(&vcpu->arch.regs));
4019 kvmhv_restore_hv_return_state(vcpu, &hvregs);
4020 switch_pmu_to_host(vcpu, &host_os_sprs);
4021 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
4022 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
4023 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
4024 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
4026 store_vcpu_state(vcpu);
4028 dec = mfspr(SPRN_DEC);
4029 if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4032 vcpu->arch.dec_expires = dec + (*tb + vc->tb_offset);
4034 timer_rearm_host_dec(*tb);
4036 restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4037 if (vcpu->arch.psscr != host_psscr)
4038 mtspr(SPRN_PSSCR_PR, host_psscr);
4044 * Guest entry for POWER9 and later CPUs.
4046 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4047 unsigned long lpcr, u64 *tb)
4049 struct kvm *kvm = vcpu->kvm;
4050 struct kvm_nested_guest *nested = vcpu->arch.nested;
4054 next_timer = timer_get_next_tb();
4055 if (*tb >= next_timer)
4056 return BOOK3S_INTERRUPT_HV_DECREMENTER;
4057 if (next_timer < time_limit)
4058 time_limit = next_timer;
4059 else if (*tb >= time_limit) /* nested time limit */
4060 return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4062 vcpu->arch.ceded = 0;
4064 vcpu_vpa_increment_dispatch(vcpu);
4066 if (kvmhv_on_pseries()) {
4067 trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4069 /* H_CEDE has to be handled now, not later */
4070 if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested &&
4071 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4073 kvmppc_set_gpr(vcpu, 3, 0);
4077 } else if (nested) {
4078 __this_cpu_write(cpu_in_guest, kvm);
4079 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4080 __this_cpu_write(cpu_in_guest, NULL);
4083 kvmppc_xive_push_vcpu(vcpu);
4085 __this_cpu_write(cpu_in_guest, kvm);
4086 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4087 __this_cpu_write(cpu_in_guest, NULL);
4089 if (trap == BOOK3S_INTERRUPT_SYSCALL &&
4090 !(vcpu->arch.shregs.msr & MSR_PR)) {
4091 unsigned long req = kvmppc_get_gpr(vcpu, 3);
4094 * XIVE rearm and XICS hcalls must be handled
4095 * before xive context is pulled (is this
4098 if (req == H_CEDE) {
4099 /* H_CEDE has to be handled now */
4101 if (!kvmppc_xive_rearm_escalation(vcpu)) {
4103 * Pending escalation so abort
4106 vcpu->arch.ceded = 0;
4108 kvmppc_set_gpr(vcpu, 3, 0);
4111 } else if (req == H_ENTER_NESTED) {
4113 * L2 should not run with the L1
4114 * context so rearm and pull it.
4116 if (!kvmppc_xive_rearm_escalation(vcpu)) {
4118 * Pending escalation so abort
4121 kvmppc_set_gpr(vcpu, 3, 0);
4125 } else if (hcall_is_xics(req)) {
4128 ret = kvmppc_xive_xics_hcall(vcpu, req);
4129 if (ret != H_TOO_HARD) {
4130 kvmppc_set_gpr(vcpu, 3, ret);
4135 kvmppc_xive_pull_vcpu(vcpu);
4137 if (kvm_is_radix(kvm))
4138 vcpu->arch.slb_max = 0;
4141 vcpu_vpa_increment_dispatch(vcpu);
4147 * Wait for some other vcpu thread to execute us, and
4148 * wake us up when we need to handle something in the host.
4150 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4151 struct kvm_vcpu *vcpu, int wait_state)
4155 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4156 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4157 spin_unlock(&vc->lock);
4159 spin_lock(&vc->lock);
4161 finish_wait(&vcpu->arch.cpu_run, &wait);
4164 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4166 if (!halt_poll_ns_grow)
4169 vc->halt_poll_ns *= halt_poll_ns_grow;
4170 if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4171 vc->halt_poll_ns = halt_poll_ns_grow_start;
4174 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4176 if (halt_poll_ns_shrink == 0)
4177 vc->halt_poll_ns = 0;
4179 vc->halt_poll_ns /= halt_poll_ns_shrink;
4182 #ifdef CONFIG_KVM_XICS
4183 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4185 if (!xics_on_xive())
4187 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4188 vcpu->arch.xive_saved_state.cppr;
4191 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4195 #endif /* CONFIG_KVM_XICS */
4197 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4199 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4200 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4206 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4208 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4214 * Check to see if any of the runnable vcpus on the vcore have pending
4215 * exceptions or are no longer ceded
4217 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4219 struct kvm_vcpu *vcpu;
4222 for_each_runnable_thread(i, vcpu, vc) {
4223 if (kvmppc_vcpu_check_block(vcpu))
4231 * All the vcpus in this vcore are idle, so wait for a decrementer
4232 * or external interrupt to one of the vcpus. vc->lock is held.
4234 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4236 ktime_t cur, start_poll, start_wait;
4240 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4242 /* Poll for pending exceptions and ceded state */
4243 cur = start_poll = ktime_get();
4244 if (vc->halt_poll_ns) {
4245 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4246 ++vc->runner->stat.generic.halt_attempted_poll;
4248 vc->vcore_state = VCORE_POLLING;
4249 spin_unlock(&vc->lock);
4252 if (kvmppc_vcore_check_block(vc)) {
4257 } while (kvm_vcpu_can_poll(cur, stop));
4259 spin_lock(&vc->lock);
4260 vc->vcore_state = VCORE_INACTIVE;
4263 ++vc->runner->stat.generic.halt_successful_poll;
4268 prepare_to_rcuwait(&vc->wait);
4269 set_current_state(TASK_INTERRUPTIBLE);
4270 if (kvmppc_vcore_check_block(vc)) {
4271 finish_rcuwait(&vc->wait);
4273 /* If we polled, count this as a successful poll */
4274 if (vc->halt_poll_ns)
4275 ++vc->runner->stat.generic.halt_successful_poll;
4279 start_wait = ktime_get();
4281 vc->vcore_state = VCORE_SLEEPING;
4282 trace_kvmppc_vcore_blocked(vc->runner, 0);
4283 spin_unlock(&vc->lock);
4285 finish_rcuwait(&vc->wait);
4286 spin_lock(&vc->lock);
4287 vc->vcore_state = VCORE_INACTIVE;
4288 trace_kvmppc_vcore_blocked(vc->runner, 1);
4289 ++vc->runner->stat.halt_successful_wait;
4294 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4296 /* Attribute wait time */
4298 vc->runner->stat.generic.halt_wait_ns +=
4299 ktime_to_ns(cur) - ktime_to_ns(start_wait);
4300 KVM_STATS_LOG_HIST_UPDATE(
4301 vc->runner->stat.generic.halt_wait_hist,
4302 ktime_to_ns(cur) - ktime_to_ns(start_wait));
4303 /* Attribute failed poll time */
4304 if (vc->halt_poll_ns) {
4305 vc->runner->stat.generic.halt_poll_fail_ns +=
4306 ktime_to_ns(start_wait) -
4307 ktime_to_ns(start_poll);
4308 KVM_STATS_LOG_HIST_UPDATE(
4309 vc->runner->stat.generic.halt_poll_fail_hist,
4310 ktime_to_ns(start_wait) -
4311 ktime_to_ns(start_poll));
4314 /* Attribute successful poll time */
4315 if (vc->halt_poll_ns) {
4316 vc->runner->stat.generic.halt_poll_success_ns +=
4318 ktime_to_ns(start_poll);
4319 KVM_STATS_LOG_HIST_UPDATE(
4320 vc->runner->stat.generic.halt_poll_success_hist,
4321 ktime_to_ns(cur) - ktime_to_ns(start_poll));
4325 /* Adjust poll time */
4327 if (block_ns <= vc->halt_poll_ns)
4329 /* We slept and blocked for longer than the max halt time */
4330 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4331 shrink_halt_poll_ns(vc);
4332 /* We slept and our poll time is too small */
4333 else if (vc->halt_poll_ns < halt_poll_ns &&
4334 block_ns < halt_poll_ns)
4335 grow_halt_poll_ns(vc);
4336 if (vc->halt_poll_ns > halt_poll_ns)
4337 vc->halt_poll_ns = halt_poll_ns;
4339 vc->halt_poll_ns = 0;
4341 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4345 * This never fails for a radix guest, as none of the operations it does
4346 * for a radix guest can fail or have a way to report failure.
4348 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4351 struct kvm *kvm = vcpu->kvm;
4353 mutex_lock(&kvm->arch.mmu_setup_lock);
4354 if (!kvm->arch.mmu_ready) {
4355 if (!kvm_is_radix(kvm))
4356 r = kvmppc_hv_setup_htab_rma(vcpu);
4358 if (cpu_has_feature(CPU_FTR_ARCH_300))
4359 kvmppc_setup_partition_table(kvm);
4360 kvm->arch.mmu_ready = 1;
4363 mutex_unlock(&kvm->arch.mmu_setup_lock);
4367 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4369 struct kvm_run *run = vcpu->run;
4371 struct kvmppc_vcore *vc;
4374 trace_kvmppc_run_vcpu_enter(vcpu);
4376 run->exit_reason = 0;
4377 vcpu->arch.ret = RESUME_GUEST;
4378 vcpu->arch.trap = 0;
4379 kvmppc_update_vpas(vcpu);
4382 * Synchronize with other threads in this virtual core
4384 vc = vcpu->arch.vcore;
4385 spin_lock(&vc->lock);
4386 vcpu->arch.ceded = 0;
4387 vcpu->arch.run_task = current;
4388 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4389 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4390 vcpu->arch.busy_preempt = TB_NIL;
4391 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4395 * This happens the first time this is called for a vcpu.
4396 * If the vcore is already running, we may be able to start
4397 * this thread straight away and have it join in.
4399 if (!signal_pending(current)) {
4400 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4401 vc->vcore_state == VCORE_RUNNING) &&
4402 !VCORE_IS_EXITING(vc)) {
4403 kvmppc_create_dtl_entry(vcpu, vc);
4404 kvmppc_start_thread(vcpu, vc);
4405 trace_kvm_guest_enter(vcpu);
4406 } else if (vc->vcore_state == VCORE_SLEEPING) {
4407 rcuwait_wake_up(&vc->wait);
4412 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4413 !signal_pending(current)) {
4414 /* See if the MMU is ready to go */
4415 if (!vcpu->kvm->arch.mmu_ready) {
4416 spin_unlock(&vc->lock);
4417 r = kvmhv_setup_mmu(vcpu);
4418 spin_lock(&vc->lock);
4420 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4422 hardware_entry_failure_reason = 0;
4428 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4429 kvmppc_vcore_end_preempt(vc);
4431 if (vc->vcore_state != VCORE_INACTIVE) {
4432 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4435 for_each_runnable_thread(i, v, vc) {
4436 kvmppc_core_prepare_to_enter(v);
4437 if (signal_pending(v->arch.run_task)) {
4438 kvmppc_remove_runnable(vc, v, mftb());
4439 v->stat.signal_exits++;
4440 v->run->exit_reason = KVM_EXIT_INTR;
4441 v->arch.ret = -EINTR;
4442 wake_up(&v->arch.cpu_run);
4445 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4448 for_each_runnable_thread(i, v, vc) {
4449 if (!kvmppc_vcpu_woken(v))
4450 n_ceded += v->arch.ceded;
4455 if (n_ceded == vc->n_runnable) {
4456 kvmppc_vcore_blocked(vc);
4457 } else if (need_resched()) {
4458 kvmppc_vcore_preempt(vc);
4459 /* Let something else run */
4460 cond_resched_lock(&vc->lock);
4461 if (vc->vcore_state == VCORE_PREEMPT)
4462 kvmppc_vcore_end_preempt(vc);
4464 kvmppc_run_core(vc);
4469 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4470 (vc->vcore_state == VCORE_RUNNING ||
4471 vc->vcore_state == VCORE_EXITING ||
4472 vc->vcore_state == VCORE_PIGGYBACK))
4473 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4475 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4476 kvmppc_vcore_end_preempt(vc);
4478 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4479 kvmppc_remove_runnable(vc, vcpu, mftb());
4480 vcpu->stat.signal_exits++;
4481 run->exit_reason = KVM_EXIT_INTR;
4482 vcpu->arch.ret = -EINTR;
4485 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4486 /* Wake up some vcpu to run the core */
4488 v = next_runnable_thread(vc, &i);
4489 wake_up(&v->arch.cpu_run);
4492 trace_kvmppc_run_vcpu_exit(vcpu);
4493 spin_unlock(&vc->lock);
4494 return vcpu->arch.ret;
4497 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4500 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4501 struct kvm_run *run = vcpu->run;
4504 struct kvmppc_vcore *vc;
4505 struct kvm *kvm = vcpu->kvm;
4506 struct kvm_nested_guest *nested = vcpu->arch.nested;
4507 unsigned long flags;
4510 trace_kvmppc_run_vcpu_enter(vcpu);
4512 run->exit_reason = 0;
4513 vcpu->arch.ret = RESUME_GUEST;
4514 vcpu->arch.trap = 0;
4516 vc = vcpu->arch.vcore;
4517 vcpu->arch.ceded = 0;
4518 vcpu->arch.run_task = current;
4519 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4520 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4522 /* See if the MMU is ready to go */
4523 if (unlikely(!kvm->arch.mmu_ready)) {
4524 r = kvmhv_setup_mmu(vcpu);
4526 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4527 run->fail_entry.hardware_entry_failure_reason = 0;
4536 kvmppc_update_vpas(vcpu);
4539 pcpu = smp_processor_id();
4540 if (kvm_is_radix(kvm))
4541 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4543 /* flags save not required, but irq_pmu has no disable/enable API */
4544 powerpc_local_irq_pmu_save(flags);
4546 if (signal_pending(current))
4548 if (need_resched() || !kvm->arch.mmu_ready)
4552 vcpu->arch.thread_cpu = pcpu;
4554 local_paca->kvm_hstate.kvm_vcpu = vcpu;
4555 local_paca->kvm_hstate.ptid = 0;
4556 local_paca->kvm_hstate.fake_suspend = 0;
4559 * Orders set cpu/thread_cpu vs testing for pending interrupts and
4560 * doorbells below. The other side is when these fields are set vs
4561 * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to
4562 * kick a vCPU to notice the pending interrupt.
4567 kvmppc_core_prepare_to_enter(vcpu);
4568 if (vcpu->arch.shregs.msr & MSR_EE) {
4569 if (xive_interrupt_pending(vcpu))
4570 kvmppc_inject_interrupt_hv(vcpu,
4571 BOOK3S_INTERRUPT_EXTERNAL, 0);
4572 } else if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4573 &vcpu->arch.pending_exceptions)) {
4576 } else if (vcpu->arch.pending_exceptions ||
4577 vcpu->arch.doorbell_request ||
4578 xive_interrupt_pending(vcpu)) {
4579 vcpu->arch.ret = RESUME_HOST;
4583 if (vcpu->arch.timer_running) {
4584 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4585 vcpu->arch.timer_running = 0;
4590 __kvmppc_create_dtl_entry(vcpu, pcpu, tb + vc->tb_offset, 0);
4592 trace_kvm_guest_enter(vcpu);
4594 guest_enter_irqoff();
4596 srcu_idx = srcu_read_lock(&kvm->srcu);
4598 this_cpu_disable_ftrace();
4600 /* Tell lockdep that we're about to enable interrupts */
4601 trace_hardirqs_on();
4603 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4604 vcpu->arch.trap = trap;
4606 trace_hardirqs_off();
4608 this_cpu_enable_ftrace();
4610 srcu_read_unlock(&kvm->srcu, srcu_idx);
4612 set_irq_happened(trap);
4614 context_tracking_guest_exit();
4615 if (!vtime_accounting_enabled_this_cpu()) {
4618 * Service IRQs here before vtime_account_guest_exit() so any
4619 * ticks that occurred while running the guest are accounted to
4620 * the guest. If vtime accounting is enabled, accounting uses
4621 * TB rather than ticks, so it can be done without enabling
4622 * interrupts here, which has the problem that it accounts
4623 * interrupt processing overhead to the host.
4625 local_irq_disable();
4627 vtime_account_guest_exit();
4630 vcpu->arch.thread_cpu = -1;
4632 powerpc_local_irq_pmu_restore(flags);
4637 * cancel pending decrementer exception if DEC is now positive, or if
4638 * entering a nested guest in which case the decrementer is now owned
4639 * by L2 and the L1 decrementer is provided in hdec_expires
4641 if (kvmppc_core_pending_dec(vcpu) &&
4642 ((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
4643 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4644 kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4645 kvmppc_core_dequeue_dec(vcpu);
4647 trace_kvm_guest_exit(vcpu);
4651 r = kvmppc_handle_exit_hv(vcpu, current);
4653 r = kvmppc_handle_nested_exit(vcpu);
4657 if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
4658 kvmppc_set_timer(vcpu);
4660 prepare_to_rcuwait(wait);
4662 set_current_state(TASK_INTERRUPTIBLE);
4663 if (signal_pending(current)) {
4664 vcpu->stat.signal_exits++;
4665 run->exit_reason = KVM_EXIT_INTR;
4666 vcpu->arch.ret = -EINTR;
4670 if (kvmppc_vcpu_check_block(vcpu))
4673 trace_kvmppc_vcore_blocked(vcpu, 0);
4675 trace_kvmppc_vcore_blocked(vcpu, 1);
4677 finish_rcuwait(wait);
4679 vcpu->arch.ceded = 0;
4682 trace_kvmppc_run_vcpu_exit(vcpu);
4684 return vcpu->arch.ret;
4687 vcpu->stat.signal_exits++;
4688 run->exit_reason = KVM_EXIT_INTR;
4689 vcpu->arch.ret = -EINTR;
4692 vcpu->arch.thread_cpu = -1;
4693 powerpc_local_irq_pmu_restore(flags);
4698 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4700 struct kvm_run *run = vcpu->run;
4706 if (!vcpu->arch.sane) {
4707 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4711 /* No need to go into the guest when all we'll do is come back out */
4712 if (signal_pending(current)) {
4713 run->exit_reason = KVM_EXIT_INTR;
4717 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4719 * Don't allow entry with a suspended transaction, because
4720 * the guest entry/exit code will lose it.
4722 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4723 (current->thread.regs->msr & MSR_TM)) {
4724 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4725 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4726 run->fail_entry.hardware_entry_failure_reason = 0;
4733 * Force online to 1 for the sake of old userspace which doesn't
4736 if (!vcpu->arch.online) {
4737 atomic_inc(&vcpu->arch.vcore->online_count);
4738 vcpu->arch.online = 1;
4741 kvmppc_core_prepare_to_enter(vcpu);
4744 atomic_inc(&kvm->arch.vcpus_running);
4745 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4749 if (IS_ENABLED(CONFIG_PPC_FPU))
4751 if (cpu_has_feature(CPU_FTR_ALTIVEC))
4753 if (cpu_has_feature(CPU_FTR_VSX))
4755 if ((cpu_has_feature(CPU_FTR_TM) ||
4756 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
4757 (vcpu->arch.hfscr & HFSCR_TM))
4759 msr = msr_check_and_set(msr);
4761 kvmppc_save_user_regs();
4763 kvmppc_save_current_sprs();
4765 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4766 vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4767 vcpu->arch.pgdir = kvm->mm->pgd;
4768 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4771 if (cpu_has_feature(CPU_FTR_ARCH_300))
4772 r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4773 vcpu->arch.vcore->lpcr);
4775 r = kvmppc_run_vcpu(vcpu);
4777 if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4778 if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4780 * These should have been caught reflected
4781 * into the guest by now. Final sanity check:
4782 * don't allow userspace to execute hcalls in
4788 trace_kvm_hcall_enter(vcpu);
4789 r = kvmppc_pseries_do_hcall(vcpu);
4790 trace_kvm_hcall_exit(vcpu, r);
4791 kvmppc_core_prepare_to_enter(vcpu);
4792 } else if (r == RESUME_PAGE_FAULT) {
4793 srcu_idx = srcu_read_lock(&kvm->srcu);
4794 r = kvmppc_book3s_hv_page_fault(vcpu,
4795 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4796 srcu_read_unlock(&kvm->srcu, srcu_idx);
4797 } else if (r == RESUME_PASSTHROUGH) {
4798 if (WARN_ON(xics_on_xive()))
4801 r = kvmppc_xics_rm_complete(vcpu, 0);
4803 } while (is_kvmppc_resume_guest(r));
4805 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4806 atomic_dec(&kvm->arch.vcpus_running);
4808 srr_regs_clobbered();
4813 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4814 int shift, int sllp)
4816 (*sps)->page_shift = shift;
4817 (*sps)->slb_enc = sllp;
4818 (*sps)->enc[0].page_shift = shift;
4819 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4821 * Add 16MB MPSS support (may get filtered out by userspace)
4824 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4826 (*sps)->enc[1].page_shift = 24;
4827 (*sps)->enc[1].pte_enc = penc;
4833 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4834 struct kvm_ppc_smmu_info *info)
4836 struct kvm_ppc_one_seg_page_size *sps;
4839 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4840 * POWER7 doesn't support keys for instruction accesses,
4841 * POWER8 and POWER9 do.
4843 info->data_keys = 32;
4844 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4846 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4847 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4848 info->slb_size = 32;
4850 /* We only support these sizes for now, and no muti-size segments */
4851 sps = &info->sps[0];
4852 kvmppc_add_seg_page_size(&sps, 12, 0);
4853 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4854 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4856 /* If running as a nested hypervisor, we don't support HPT guests */
4857 if (kvmhv_on_pseries())
4858 info->flags |= KVM_PPC_NO_HASH;
4864 * Get (and clear) the dirty memory log for a memory slot.
4866 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4867 struct kvm_dirty_log *log)
4869 struct kvm_memslots *slots;
4870 struct kvm_memory_slot *memslot;
4873 unsigned long *buf, *p;
4874 struct kvm_vcpu *vcpu;
4876 mutex_lock(&kvm->slots_lock);
4879 if (log->slot >= KVM_USER_MEM_SLOTS)
4882 slots = kvm_memslots(kvm);
4883 memslot = id_to_memslot(slots, log->slot);
4885 if (!memslot || !memslot->dirty_bitmap)
4889 * Use second half of bitmap area because both HPT and radix
4890 * accumulate bits in the first half.
4892 n = kvm_dirty_bitmap_bytes(memslot);
4893 buf = memslot->dirty_bitmap + n / sizeof(long);
4896 if (kvm_is_radix(kvm))
4897 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4899 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4904 * We accumulate dirty bits in the first half of the
4905 * memslot's dirty_bitmap area, for when pages are paged
4906 * out or modified by the host directly. Pick up these
4907 * bits and add them to the map.
4909 p = memslot->dirty_bitmap;
4910 for (i = 0; i < n / sizeof(long); ++i)
4911 buf[i] |= xchg(&p[i], 0);
4913 /* Harvest dirty bits from VPA and DTL updates */
4914 /* Note: we never modify the SLB shadow buffer areas */
4915 kvm_for_each_vcpu(i, vcpu, kvm) {
4916 spin_lock(&vcpu->arch.vpa_update_lock);
4917 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4918 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4919 spin_unlock(&vcpu->arch.vpa_update_lock);
4923 if (copy_to_user(log->dirty_bitmap, buf, n))
4928 mutex_unlock(&kvm->slots_lock);
4932 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4934 vfree(slot->arch.rmap);
4935 slot->arch.rmap = NULL;
4938 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4939 const struct kvm_memory_slot *old,
4940 struct kvm_memory_slot *new,
4941 enum kvm_mr_change change)
4943 if (change == KVM_MR_CREATE) {
4944 unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
4946 if ((size >> PAGE_SHIFT) > totalram_pages())
4949 new->arch.rmap = vzalloc(size);
4950 if (!new->arch.rmap)
4952 } else if (change != KVM_MR_DELETE) {
4953 new->arch.rmap = old->arch.rmap;
4959 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4960 struct kvm_memory_slot *old,
4961 const struct kvm_memory_slot *new,
4962 enum kvm_mr_change change)
4965 * If we are creating or modifying a memslot, it might make
4966 * some address that was previously cached as emulated
4967 * MMIO be no longer emulated MMIO, so invalidate
4968 * all the caches of emulated MMIO translations.
4970 if (change != KVM_MR_DELETE)
4971 atomic64_inc(&kvm->arch.mmio_update);
4974 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4975 * have already called kvm_arch_flush_shadow_memslot() to
4976 * flush shadow mappings. For KVM_MR_CREATE we have no
4977 * previous mappings. So the only case to handle is
4978 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4980 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4981 * to get rid of any THP PTEs in the partition-scoped page tables
4982 * so we can track dirtiness at the page level; we flush when
4983 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4986 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4987 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4988 kvmppc_radix_flush_memslot(kvm, old);
4990 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4992 if (!kvm->arch.secure_guest)
4998 * @TODO kvmppc_uvmem_memslot_create() can fail and
4999 * return error. Fix this.
5001 kvmppc_uvmem_memslot_create(kvm, new);
5004 kvmppc_uvmem_memslot_delete(kvm, old);
5007 /* TODO: Handle KVM_MR_MOVE */
5013 * Update LPCR values in kvm->arch and in vcores.
5014 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
5015 * of kvm->arch.lpcr update).
5017 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
5022 if ((kvm->arch.lpcr & mask) == lpcr)
5025 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
5027 for (i = 0; i < KVM_MAX_VCORES; ++i) {
5028 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
5032 spin_lock(&vc->lock);
5033 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
5034 verify_lpcr(kvm, vc->lpcr);
5035 spin_unlock(&vc->lock);
5036 if (++cores_done >= kvm->arch.online_vcores)
5041 void kvmppc_setup_partition_table(struct kvm *kvm)
5043 unsigned long dw0, dw1;
5045 if (!kvm_is_radix(kvm)) {
5046 /* PS field - page size for VRMA */
5047 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
5048 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
5049 /* HTABSIZE and HTABORG fields */
5050 dw0 |= kvm->arch.sdr1;
5052 /* Second dword as set by userspace */
5053 dw1 = kvm->arch.process_table;
5055 dw0 = PATB_HR | radix__get_tree_size() |
5056 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
5057 dw1 = PATB_GR | kvm->arch.process_table;
5059 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
5063 * Set up HPT (hashed page table) and RMA (real-mode area).
5064 * Must be called with kvm->arch.mmu_setup_lock held.
5066 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
5069 struct kvm *kvm = vcpu->kvm;
5071 struct kvm_memory_slot *memslot;
5072 struct vm_area_struct *vma;
5073 unsigned long lpcr = 0, senc;
5074 unsigned long psize, porder;
5077 /* Allocate hashed page table (if not done already) and reset it */
5078 if (!kvm->arch.hpt.virt) {
5079 int order = KVM_DEFAULT_HPT_ORDER;
5080 struct kvm_hpt_info info;
5082 err = kvmppc_allocate_hpt(&info, order);
5083 /* If we get here, it means userspace didn't specify a
5084 * size explicitly. So, try successively smaller
5085 * sizes if the default failed. */
5086 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5087 err = kvmppc_allocate_hpt(&info, order);
5090 pr_err("KVM: Couldn't alloc HPT\n");
5094 kvmppc_set_hpt(kvm, &info);
5097 /* Look up the memslot for guest physical address 0 */
5098 srcu_idx = srcu_read_lock(&kvm->srcu);
5099 memslot = gfn_to_memslot(kvm, 0);
5101 /* We must have some memory at 0 by now */
5103 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5106 /* Look up the VMA for the start of this memory slot */
5107 hva = memslot->userspace_addr;
5108 mmap_read_lock(kvm->mm);
5109 vma = vma_lookup(kvm->mm, hva);
5110 if (!vma || (vma->vm_flags & VM_IO))
5113 psize = vma_kernel_pagesize(vma);
5115 mmap_read_unlock(kvm->mm);
5117 /* We can handle 4k, 64k or 16M pages in the VRMA */
5118 if (psize >= 0x1000000)
5120 else if (psize >= 0x10000)
5124 porder = __ilog2(psize);
5126 senc = slb_pgsize_encoding(psize);
5127 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5128 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5129 /* Create HPTEs in the hash page table for the VRMA */
5130 kvmppc_map_vrma(vcpu, memslot, porder);
5132 /* Update VRMASD field in the LPCR */
5133 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5134 /* the -4 is to account for senc values starting at 0x10 */
5135 lpcr = senc << (LPCR_VRMASD_SH - 4);
5136 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5139 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5143 srcu_read_unlock(&kvm->srcu, srcu_idx);
5148 mmap_read_unlock(kvm->mm);
5153 * Must be called with kvm->arch.mmu_setup_lock held and
5154 * mmu_ready = 0 and no vcpus running.
5156 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5158 unsigned long lpcr, lpcr_mask;
5160 if (nesting_enabled(kvm))
5161 kvmhv_release_all_nested(kvm);
5162 kvmppc_rmap_reset(kvm);
5163 kvm->arch.process_table = 0;
5164 /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5165 spin_lock(&kvm->mmu_lock);
5166 kvm->arch.radix = 0;
5167 spin_unlock(&kvm->mmu_lock);
5168 kvmppc_free_radix(kvm);
5171 lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5172 if (cpu_has_feature(CPU_FTR_ARCH_31))
5173 lpcr_mask |= LPCR_HAIL;
5174 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5180 * Must be called with kvm->arch.mmu_setup_lock held and
5181 * mmu_ready = 0 and no vcpus running.
5183 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5185 unsigned long lpcr, lpcr_mask;
5188 err = kvmppc_init_vm_radix(kvm);
5191 kvmppc_rmap_reset(kvm);
5192 /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5193 spin_lock(&kvm->mmu_lock);
5194 kvm->arch.radix = 1;
5195 spin_unlock(&kvm->mmu_lock);
5196 kvmppc_free_hpt(&kvm->arch.hpt);
5198 lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5199 lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5200 if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5201 lpcr_mask |= LPCR_HAIL;
5202 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5203 (kvm->arch.host_lpcr & LPCR_HAIL))
5206 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5211 #ifdef CONFIG_KVM_XICS
5213 * Allocate a per-core structure for managing state about which cores are
5214 * running in the host versus the guest and for exchanging data between
5215 * real mode KVM and CPU running in the host.
5216 * This is only done for the first VM.
5217 * The allocated structure stays even if all VMs have stopped.
5218 * It is only freed when the kvm-hv module is unloaded.
5219 * It's OK for this routine to fail, we just don't support host
5220 * core operations like redirecting H_IPI wakeups.
5222 void kvmppc_alloc_host_rm_ops(void)
5224 struct kvmppc_host_rm_ops *ops;
5225 unsigned long l_ops;
5229 if (cpu_has_feature(CPU_FTR_ARCH_300))
5232 /* Not the first time here ? */
5233 if (kvmppc_host_rm_ops_hv != NULL)
5236 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5240 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5241 ops->rm_core = kzalloc(size, GFP_KERNEL);
5243 if (!ops->rm_core) {
5250 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5251 if (!cpu_online(cpu))
5254 core = cpu >> threads_shift;
5255 ops->rm_core[core].rm_state.in_host = 1;
5258 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5261 * Make the contents of the kvmppc_host_rm_ops structure visible
5262 * to other CPUs before we assign it to the global variable.
5263 * Do an atomic assignment (no locks used here), but if someone
5264 * beats us to it, just free our copy and return.
5267 l_ops = (unsigned long) ops;
5269 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5271 kfree(ops->rm_core);
5276 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5277 "ppc/kvm_book3s:prepare",
5278 kvmppc_set_host_core,
5279 kvmppc_clear_host_core);
5283 void kvmppc_free_host_rm_ops(void)
5285 if (kvmppc_host_rm_ops_hv) {
5286 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5287 kfree(kvmppc_host_rm_ops_hv->rm_core);
5288 kfree(kvmppc_host_rm_ops_hv);
5289 kvmppc_host_rm_ops_hv = NULL;
5294 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5296 unsigned long lpcr, lpid;
5299 mutex_init(&kvm->arch.uvmem_lock);
5300 INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5301 mutex_init(&kvm->arch.mmu_setup_lock);
5303 /* Allocate the guest's logical partition ID */
5305 lpid = kvmppc_alloc_lpid();
5308 kvm->arch.lpid = lpid;
5310 kvmppc_alloc_host_rm_ops();
5312 kvmhv_vm_nested_init(kvm);
5315 * Since we don't flush the TLB when tearing down a VM,
5316 * and this lpid might have previously been used,
5317 * make sure we flush on each core before running the new VM.
5318 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5319 * does this flush for us.
5321 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5322 cpumask_setall(&kvm->arch.need_tlb_flush);
5324 /* Start out with the default set of hcalls enabled */
5325 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5326 sizeof(kvm->arch.enabled_hcalls));
5328 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5329 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5331 /* Init LPCR for virtual RMA mode */
5332 if (cpu_has_feature(CPU_FTR_HVMODE)) {
5333 kvm->arch.host_lpid = mfspr(SPRN_LPID);
5334 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5335 lpcr &= LPCR_PECE | LPCR_LPES;
5338 * The L2 LPES mode will be set by the L0 according to whether
5339 * or not it needs to take external interrupts in HV mode.
5343 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5344 LPCR_VPM0 | LPCR_VPM1;
5345 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5346 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5347 /* On POWER8 turn on online bit to enable PURR/SPURR */
5348 if (cpu_has_feature(CPU_FTR_ARCH_207S))
5351 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5352 * Set HVICE bit to enable hypervisor virtualization interrupts.
5353 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5354 * be unnecessary but better safe than sorry in case we re-enable
5355 * EE in HV mode with this LPCR still set)
5357 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5359 lpcr |= LPCR_HVICE | LPCR_HEIC;
5362 * If xive is enabled, we route 0x500 interrupts directly
5370 * If the host uses radix, the guest starts out as radix.
5372 if (radix_enabled()) {
5373 kvm->arch.radix = 1;
5374 kvm->arch.mmu_ready = 1;
5376 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5377 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5378 cpu_has_feature(CPU_FTR_ARCH_31) &&
5379 (kvm->arch.host_lpcr & LPCR_HAIL))
5381 ret = kvmppc_init_vm_radix(kvm);
5383 kvmppc_free_lpid(kvm->arch.lpid);
5386 kvmppc_setup_partition_table(kvm);
5389 verify_lpcr(kvm, lpcr);
5390 kvm->arch.lpcr = lpcr;
5392 /* Initialization for future HPT resizes */
5393 kvm->arch.resize_hpt = NULL;
5396 * Work out how many sets the TLB has, for the use of
5397 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5399 if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5401 * P10 will flush all the congruence class with a single tlbiel
5403 kvm->arch.tlb_sets = 1;
5404 } else if (radix_enabled())
5405 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
5406 else if (cpu_has_feature(CPU_FTR_ARCH_300))
5407 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
5408 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5409 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
5411 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
5414 * Track that we now have a HV mode VM active. This blocks secondary
5415 * CPU threads from coming online.
5417 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5418 kvm_hv_vm_activated();
5421 * Initialize smt_mode depending on processor.
5422 * POWER8 and earlier have to use "strict" threading, where
5423 * all vCPUs in a vcore have to run on the same (sub)core,
5424 * whereas on POWER9 the threads can each run a different
5427 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5428 kvm->arch.smt_mode = threads_per_subcore;
5430 kvm->arch.smt_mode = 1;
5431 kvm->arch.emul_smt_mode = 1;
5436 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm)
5438 kvmppc_mmu_debugfs_init(kvm);
5439 if (radix_enabled())
5440 kvmhv_radix_debugfs_init(kvm);
5444 static void kvmppc_free_vcores(struct kvm *kvm)
5448 for (i = 0; i < KVM_MAX_VCORES; ++i)
5449 kfree(kvm->arch.vcores[i]);
5450 kvm->arch.online_vcores = 0;
5453 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5455 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5456 kvm_hv_vm_deactivated();
5458 kvmppc_free_vcores(kvm);
5461 if (kvm_is_radix(kvm))
5462 kvmppc_free_radix(kvm);
5464 kvmppc_free_hpt(&kvm->arch.hpt);
5466 /* Perform global invalidation and return lpid to the pool */
5467 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5468 if (nesting_enabled(kvm))
5469 kvmhv_release_all_nested(kvm);
5470 kvm->arch.process_table = 0;
5471 if (kvm->arch.secure_guest)
5472 uv_svm_terminate(kvm->arch.lpid);
5473 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5476 kvmppc_free_lpid(kvm->arch.lpid);
5478 kvmppc_free_pimap(kvm);
5481 /* We don't need to emulate any privileged instructions or dcbz */
5482 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5483 unsigned int inst, int *advance)
5485 return EMULATE_FAIL;
5488 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5491 return EMULATE_FAIL;
5494 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5497 return EMULATE_FAIL;
5500 static int kvmppc_core_check_processor_compat_hv(void)
5502 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5503 cpu_has_feature(CPU_FTR_ARCH_206))
5506 /* POWER9 in radix mode is capable of being a nested hypervisor. */
5507 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5513 #ifdef CONFIG_KVM_XICS
5515 void kvmppc_free_pimap(struct kvm *kvm)
5517 kfree(kvm->arch.pimap);
5520 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5522 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5525 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5527 struct irq_desc *desc;
5528 struct kvmppc_irq_map *irq_map;
5529 struct kvmppc_passthru_irqmap *pimap;
5530 struct irq_chip *chip;
5532 struct irq_data *host_data;
5534 if (!kvm_irq_bypass)
5537 desc = irq_to_desc(host_irq);
5541 mutex_lock(&kvm->lock);
5543 pimap = kvm->arch.pimap;
5544 if (pimap == NULL) {
5545 /* First call, allocate structure to hold IRQ map */
5546 pimap = kvmppc_alloc_pimap();
5547 if (pimap == NULL) {
5548 mutex_unlock(&kvm->lock);
5551 kvm->arch.pimap = pimap;
5555 * For now, we only support interrupts for which the EOI operation
5556 * is an OPAL call followed by a write to XIRR, since that's
5557 * what our real-mode EOI code does, or a XIVE interrupt
5559 chip = irq_data_get_irq_chip(&desc->irq_data);
5560 if (!chip || !is_pnv_opal_msi(chip)) {
5561 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5562 host_irq, guest_gsi);
5563 mutex_unlock(&kvm->lock);
5568 * See if we already have an entry for this guest IRQ number.
5569 * If it's mapped to a hardware IRQ number, that's an error,
5570 * otherwise re-use this entry.
5572 for (i = 0; i < pimap->n_mapped; i++) {
5573 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5574 if (pimap->mapped[i].r_hwirq) {
5575 mutex_unlock(&kvm->lock);
5582 if (i == KVMPPC_PIRQ_MAPPED) {
5583 mutex_unlock(&kvm->lock);
5584 return -EAGAIN; /* table is full */
5587 irq_map = &pimap->mapped[i];
5589 irq_map->v_hwirq = guest_gsi;
5590 irq_map->desc = desc;
5593 * Order the above two stores before the next to serialize with
5594 * the KVM real mode handler.
5599 * The 'host_irq' number is mapped in the PCI-MSI domain but
5600 * the underlying calls, which will EOI the interrupt in real
5601 * mode, need an HW IRQ number mapped in the XICS IRQ domain.
5603 host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
5604 irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
5606 if (i == pimap->n_mapped)
5610 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
5612 kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
5614 irq_map->r_hwirq = 0;
5616 mutex_unlock(&kvm->lock);
5621 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5623 struct irq_desc *desc;
5624 struct kvmppc_passthru_irqmap *pimap;
5627 if (!kvm_irq_bypass)
5630 desc = irq_to_desc(host_irq);
5634 mutex_lock(&kvm->lock);
5635 if (!kvm->arch.pimap)
5638 pimap = kvm->arch.pimap;
5640 for (i = 0; i < pimap->n_mapped; i++) {
5641 if (guest_gsi == pimap->mapped[i].v_hwirq)
5645 if (i == pimap->n_mapped) {
5646 mutex_unlock(&kvm->lock);
5651 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
5653 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5655 /* invalidate the entry (what do do on error from the above ?) */
5656 pimap->mapped[i].r_hwirq = 0;
5659 * We don't free this structure even when the count goes to
5660 * zero. The structure is freed when we destroy the VM.
5663 mutex_unlock(&kvm->lock);
5667 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5668 struct irq_bypass_producer *prod)
5671 struct kvm_kernel_irqfd *irqfd =
5672 container_of(cons, struct kvm_kernel_irqfd, consumer);
5674 irqfd->producer = prod;
5676 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5678 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5679 prod->irq, irqfd->gsi, ret);
5684 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5685 struct irq_bypass_producer *prod)
5688 struct kvm_kernel_irqfd *irqfd =
5689 container_of(cons, struct kvm_kernel_irqfd, consumer);
5691 irqfd->producer = NULL;
5694 * When producer of consumer is unregistered, we change back to
5695 * default external interrupt handling mode - KVM real mode
5696 * will switch back to host.
5698 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5700 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5701 prod->irq, irqfd->gsi, ret);
5705 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5706 unsigned int ioctl, unsigned long arg)
5708 struct kvm *kvm __maybe_unused = filp->private_data;
5709 void __user *argp = (void __user *)arg;
5714 case KVM_PPC_ALLOCATE_HTAB: {
5717 /* If we're a nested hypervisor, we currently only support radix */
5718 if (kvmhv_on_pseries()) {
5724 if (get_user(htab_order, (u32 __user *)argp))
5726 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5733 case KVM_PPC_GET_HTAB_FD: {
5734 struct kvm_get_htab_fd ghf;
5737 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5739 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5743 case KVM_PPC_RESIZE_HPT_PREPARE: {
5744 struct kvm_ppc_resize_hpt rhpt;
5747 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5750 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5754 case KVM_PPC_RESIZE_HPT_COMMIT: {
5755 struct kvm_ppc_resize_hpt rhpt;
5758 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5761 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5773 * List of hcall numbers to enable by default.
5774 * For compatibility with old userspace, we enable by default
5775 * all hcalls that were implemented before the hcall-enabling
5776 * facility was added. Note this list should not include H_RTAS.
5778 static unsigned int default_hcall_list[] = {
5784 #ifdef CONFIG_SPAPR_TCE_IOMMU
5794 #ifdef CONFIG_KVM_XICS
5805 static void init_default_hcalls(void)
5810 for (i = 0; default_hcall_list[i]; ++i) {
5811 hcall = default_hcall_list[i];
5812 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5813 __set_bit(hcall / 4, default_enabled_hcalls);
5817 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5823 /* If not on a POWER9, reject it */
5824 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5827 /* If any unknown flags set, reject it */
5828 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5831 /* GR (guest radix) bit in process_table field must match */
5832 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5833 if (!!(cfg->process_table & PATB_GR) != radix)
5836 /* Process table size field must be reasonable, i.e. <= 24 */
5837 if ((cfg->process_table & PRTS_MASK) > 24)
5840 /* We can change a guest to/from radix now, if the host is radix */
5841 if (radix && !radix_enabled())
5844 /* If we're a nested hypervisor, we currently only support radix */
5845 if (kvmhv_on_pseries() && !radix)
5848 mutex_lock(&kvm->arch.mmu_setup_lock);
5849 if (radix != kvm_is_radix(kvm)) {
5850 if (kvm->arch.mmu_ready) {
5851 kvm->arch.mmu_ready = 0;
5852 /* order mmu_ready vs. vcpus_running */
5854 if (atomic_read(&kvm->arch.vcpus_running)) {
5855 kvm->arch.mmu_ready = 1;
5861 err = kvmppc_switch_mmu_to_radix(kvm);
5863 err = kvmppc_switch_mmu_to_hpt(kvm);
5868 kvm->arch.process_table = cfg->process_table;
5869 kvmppc_setup_partition_table(kvm);
5871 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5872 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5876 mutex_unlock(&kvm->arch.mmu_setup_lock);
5880 static int kvmhv_enable_nested(struct kvm *kvm)
5884 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5886 if (!radix_enabled())
5889 /* kvm == NULL means the caller is testing if the capability exists */
5891 kvm->arch.nested_enable = true;
5895 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5900 if (kvmhv_vcpu_is_radix(vcpu)) {
5901 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5907 /* For now quadrants are the only way to access nested guest memory */
5908 if (rc && vcpu->arch.nested)
5914 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5919 if (kvmhv_vcpu_is_radix(vcpu)) {
5920 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5926 /* For now quadrants are the only way to access nested guest memory */
5927 if (rc && vcpu->arch.nested)
5933 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5935 unpin_vpa(kvm, vpa);
5937 vpa->pinned_addr = NULL;
5939 vpa->update_pending = 0;
5943 * Enable a guest to become a secure VM, or test whether
5944 * that could be enabled.
5945 * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5946 * tested (kvm == NULL) or enabled (kvm != NULL).
5948 static int kvmhv_enable_svm(struct kvm *kvm)
5950 if (!kvmppc_uvmem_available())
5953 kvm->arch.svm_enabled = 1;
5958 * IOCTL handler to turn off secure mode of guest
5960 * - Release all device pages
5961 * - Issue ucall to terminate the guest on the UV side
5962 * - Unpin the VPA pages.
5963 * - Reinit the partition scoped page tables
5965 static int kvmhv_svm_off(struct kvm *kvm)
5967 struct kvm_vcpu *vcpu;
5973 if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5976 mutex_lock(&kvm->arch.mmu_setup_lock);
5977 mmu_was_ready = kvm->arch.mmu_ready;
5978 if (kvm->arch.mmu_ready) {
5979 kvm->arch.mmu_ready = 0;
5980 /* order mmu_ready vs. vcpus_running */
5982 if (atomic_read(&kvm->arch.vcpus_running)) {
5983 kvm->arch.mmu_ready = 1;
5989 srcu_idx = srcu_read_lock(&kvm->srcu);
5990 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5991 struct kvm_memory_slot *memslot;
5992 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5998 kvm_for_each_memslot(memslot, bkt, slots) {
5999 kvmppc_uvmem_drop_pages(memslot, kvm, true);
6000 uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
6003 srcu_read_unlock(&kvm->srcu, srcu_idx);
6005 ret = uv_svm_terminate(kvm->arch.lpid);
6006 if (ret != U_SUCCESS) {
6012 * When secure guest is reset, all the guest pages are sent
6013 * to UV via UV_PAGE_IN before the non-boot vcpus get a
6014 * chance to run and unpin their VPA pages. Unpinning of all
6015 * VPA pages is done here explicitly so that VPA pages
6016 * can be migrated to the secure side.
6018 * This is required to for the secure SMP guest to reboot
6021 kvm_for_each_vcpu(i, vcpu, kvm) {
6022 spin_lock(&vcpu->arch.vpa_update_lock);
6023 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
6024 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
6025 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
6026 spin_unlock(&vcpu->arch.vpa_update_lock);
6029 kvmppc_setup_partition_table(kvm);
6030 kvm->arch.secure_guest = 0;
6031 kvm->arch.mmu_ready = mmu_was_ready;
6033 mutex_unlock(&kvm->arch.mmu_setup_lock);
6037 static int kvmhv_enable_dawr1(struct kvm *kvm)
6039 if (!cpu_has_feature(CPU_FTR_DAWR1))
6042 /* kvm == NULL means the caller is testing if the capability exists */
6044 kvm->arch.dawr1_enabled = true;
6048 static bool kvmppc_hash_v3_possible(void)
6050 if (!cpu_has_feature(CPU_FTR_ARCH_300))
6053 if (!cpu_has_feature(CPU_FTR_HVMODE))
6057 * POWER9 chips before version 2.02 can't have some threads in
6058 * HPT mode and some in radix mode on the same core.
6060 if (radix_enabled()) {
6061 unsigned int pvr = mfspr(SPRN_PVR);
6062 if ((pvr >> 16) == PVR_POWER9 &&
6063 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
6064 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
6071 static struct kvmppc_ops kvm_ops_hv = {
6072 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6073 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6074 .get_one_reg = kvmppc_get_one_reg_hv,
6075 .set_one_reg = kvmppc_set_one_reg_hv,
6076 .vcpu_load = kvmppc_core_vcpu_load_hv,
6077 .vcpu_put = kvmppc_core_vcpu_put_hv,
6078 .inject_interrupt = kvmppc_inject_interrupt_hv,
6079 .set_msr = kvmppc_set_msr_hv,
6080 .vcpu_run = kvmppc_vcpu_run_hv,
6081 .vcpu_create = kvmppc_core_vcpu_create_hv,
6082 .vcpu_free = kvmppc_core_vcpu_free_hv,
6083 .check_requests = kvmppc_core_check_requests_hv,
6084 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
6085 .flush_memslot = kvmppc_core_flush_memslot_hv,
6086 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6087 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
6088 .unmap_gfn_range = kvm_unmap_gfn_range_hv,
6089 .age_gfn = kvm_age_gfn_hv,
6090 .test_age_gfn = kvm_test_age_gfn_hv,
6091 .set_spte_gfn = kvm_set_spte_gfn_hv,
6092 .free_memslot = kvmppc_core_free_memslot_hv,
6093 .init_vm = kvmppc_core_init_vm_hv,
6094 .destroy_vm = kvmppc_core_destroy_vm_hv,
6095 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6096 .emulate_op = kvmppc_core_emulate_op_hv,
6097 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6098 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6099 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6100 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
6101 .hcall_implemented = kvmppc_hcall_impl_hv,
6102 #ifdef CONFIG_KVM_XICS
6103 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6104 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6106 .configure_mmu = kvmhv_configure_mmu,
6107 .get_rmmu_info = kvmhv_get_rmmu_info,
6108 .set_smt_mode = kvmhv_set_smt_mode,
6109 .enable_nested = kvmhv_enable_nested,
6110 .load_from_eaddr = kvmhv_load_from_eaddr,
6111 .store_to_eaddr = kvmhv_store_to_eaddr,
6112 .enable_svm = kvmhv_enable_svm,
6113 .svm_off = kvmhv_svm_off,
6114 .enable_dawr1 = kvmhv_enable_dawr1,
6115 .hash_v3_possible = kvmppc_hash_v3_possible,
6116 .create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv,
6117 .create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv,
6120 static int kvm_init_subcore_bitmap(void)
6123 int nr_cores = cpu_nr_cores();
6124 struct sibling_subcore_state *sibling_subcore_state;
6126 for (i = 0; i < nr_cores; i++) {
6127 int first_cpu = i * threads_per_core;
6128 int node = cpu_to_node(first_cpu);
6130 /* Ignore if it is already allocated. */
6131 if (paca_ptrs[first_cpu]->sibling_subcore_state)
6134 sibling_subcore_state =
6135 kzalloc_node(sizeof(struct sibling_subcore_state),
6137 if (!sibling_subcore_state)
6141 for (j = 0; j < threads_per_core; j++) {
6142 int cpu = first_cpu + j;
6144 paca_ptrs[cpu]->sibling_subcore_state =
6145 sibling_subcore_state;
6151 static int kvmppc_radix_possible(void)
6153 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6156 static int kvmppc_book3s_init_hv(void)
6160 if (!tlbie_capable) {
6161 pr_err("KVM-HV: Host does not support TLBIE\n");
6166 * FIXME!! Do we need to check on all cpus ?
6168 r = kvmppc_core_check_processor_compat_hv();
6172 r = kvmhv_nested_init();
6176 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6177 r = kvm_init_subcore_bitmap();
6183 * We need a way of accessing the XICS interrupt controller,
6184 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6185 * indirectly, via OPAL.
6188 if (!xics_on_xive() && !kvmhv_on_pseries() &&
6189 !local_paca->kvm_hstate.xics_phys) {
6190 struct device_node *np;
6192 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6194 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6198 /* presence of intc confirmed - node can be dropped again */
6203 init_default_hcalls();
6207 r = kvmppc_mmu_hv_init();
6211 if (kvmppc_radix_possible()) {
6212 r = kvmppc_radix_init();
6217 r = kvmppc_uvmem_init();
6219 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6223 kvm_ops_hv.owner = THIS_MODULE;
6224 kvmppc_hv_ops = &kvm_ops_hv;
6229 kvmhv_nested_exit();
6230 kvmppc_radix_exit();
6235 static void kvmppc_book3s_exit_hv(void)
6237 kvmppc_uvmem_free();
6238 kvmppc_free_host_rm_ops();
6239 if (kvmppc_radix_possible())
6240 kvmppc_radix_exit();
6241 kvmppc_hv_ops = NULL;
6242 kvmhv_nested_exit();
6245 module_init(kvmppc_book3s_init_hv);
6246 module_exit(kvmppc_book3s_exit_hv);
6247 MODULE_LICENSE("GPL");
6248 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6249 MODULE_ALIAS("devname:kvm");