1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
7 * Paul Mackerras <paulus@au1.ibm.com>
8 * Alexander Graf <agraf@suse.de>
9 * Kevin Wolf <mail@kevin-wolf.de>
11 * Description: KVM functions specific to running on Book 3S
12 * processors in hypervisor mode (specifically POWER7 and later).
14 * This file is derived from arch/powerpc/kvm/book3s.c,
15 * by Alexander Graf <agraf@suse.de>.
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
46 #include <asm/ftrace.h>
48 #include <asm/ppc-opcode.h>
49 #include <asm/asm-prototypes.h>
50 #include <asm/archrandom.h>
51 #include <asm/debug.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <linux/uaccess.h>
57 #include <asm/kvm_ppc.h>
58 #include <asm/kvm_book3s.h>
59 #include <asm/mmu_context.h>
60 #include <asm/lppaca.h>
61 #include <asm/processor.h>
62 #include <asm/cputhreads.h>
64 #include <asm/hvcall.h>
65 #include <asm/switch_to.h>
67 #include <asm/dbell.h>
69 #include <asm/pnv-pci.h>
74 #include <asm/hw_breakpoint.h>
78 #define CREATE_TRACE_POINTS
81 /* #define EXIT_DEBUG */
82 /* #define EXIT_DEBUG_SIMPLE */
83 /* #define EXIT_DEBUG_INT */
85 /* Used to indicate that a guest page fault needs to be handled */
86 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
87 /* Used to indicate that a guest passthrough interrupt needs to be handled */
88 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
90 /* Used as a "null" value for timebase values */
91 #define TB_NIL (~(u64)0)
93 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
95 static int dynamic_mt_modes = 6;
96 module_param(dynamic_mt_modes, int, 0644);
97 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
98 static int target_smt_mode;
99 module_param(target_smt_mode, int, 0644);
100 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
102 static bool indep_threads_mode = true;
103 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
106 static bool one_vm_per_core;
107 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
108 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
110 #ifdef CONFIG_KVM_XICS
111 static struct kernel_param_ops module_param_ops = {
112 .set = param_set_int,
113 .get = param_get_int,
116 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
117 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
119 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
120 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
123 /* If set, guests are allowed to create and control nested guests */
124 static bool nested = true;
125 module_param(nested, bool, S_IRUGO | S_IWUSR);
126 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
128 static inline bool nesting_enabled(struct kvm *kvm)
130 return kvm->arch.nested_enable && kvm_is_radix(kvm);
133 /* If set, the threads on each CPU core have to be in the same MMU mode */
134 static bool no_mixing_hpt_and_radix;
136 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
137 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
140 * RWMR values for POWER8. These control the rate at which PURR
141 * and SPURR count and should be set according to the number of
142 * online threads in the vcore being run.
144 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
145 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
146 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
147 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
148 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
149 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
150 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
151 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
153 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
165 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
169 struct kvm_vcpu *vcpu;
171 while (++i < MAX_SMT_THREADS) {
172 vcpu = READ_ONCE(vc->runnable_threads[i]);
181 /* Used to traverse the list of runnable threads for a given vcore */
182 #define for_each_runnable_thread(i, vcpu, vc) \
183 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
185 static bool kvmppc_ipi_thread(int cpu)
187 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
189 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
190 if (kvmhv_on_pseries())
193 /* On POWER9 we can use msgsnd to IPI any cpu */
194 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
195 msg |= get_hard_smp_processor_id(cpu);
197 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
201 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
202 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
204 if (cpu_first_thread_sibling(cpu) ==
205 cpu_first_thread_sibling(smp_processor_id())) {
206 msg |= cpu_thread_in_core(cpu);
208 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
215 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
216 if (cpu >= 0 && cpu < nr_cpu_ids) {
217 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
221 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
229 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
232 struct swait_queue_head *wqp;
234 wqp = kvm_arch_vcpu_wq(vcpu);
235 if (swq_has_sleeper(wqp)) {
237 ++vcpu->stat.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)
287 spin_lock_irqsave(&vc->stoltb_lock, flags);
288 vc->preempt_tb = mftb();
289 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
292 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
296 spin_lock_irqsave(&vc->stoltb_lock, flags);
297 if (vc->preempt_tb != TB_NIL) {
298 vc->stolen_tb += mftb() - vc->preempt_tb;
299 vc->preempt_tb = TB_NIL;
301 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
304 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
306 struct kvmppc_vcore *vc = vcpu->arch.vcore;
310 * We can test vc->runner without taking the vcore lock,
311 * because only this task ever sets vc->runner to this
312 * vcpu, and once it is set to this vcpu, only this task
313 * ever sets it to NULL.
315 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
316 kvmppc_core_end_stolen(vc);
318 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
319 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
320 vcpu->arch.busy_preempt != TB_NIL) {
321 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
322 vcpu->arch.busy_preempt = TB_NIL;
324 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
327 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
329 struct kvmppc_vcore *vc = vcpu->arch.vcore;
332 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
333 kvmppc_core_start_stolen(vc);
335 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
336 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
337 vcpu->arch.busy_preempt = mftb();
338 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
341 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
344 * Check for illegal transactional state bit combination
345 * and if we find it, force the TS field to a safe state.
347 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
349 vcpu->arch.shregs.msr = msr;
350 kvmppc_end_cede(vcpu);
353 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
355 vcpu->arch.pvr = pvr;
358 /* Dummy value used in computing PCR value below */
359 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
361 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
363 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
364 struct kvmppc_vcore *vc = vcpu->arch.vcore;
366 /* We can (emulate) our own architecture version and anything older */
367 if (cpu_has_feature(CPU_FTR_ARCH_300))
368 host_pcr_bit = PCR_ARCH_300;
369 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
370 host_pcr_bit = PCR_ARCH_207;
371 else if (cpu_has_feature(CPU_FTR_ARCH_206))
372 host_pcr_bit = PCR_ARCH_206;
374 host_pcr_bit = PCR_ARCH_205;
376 /* Determine lowest PCR bit needed to run guest in given PVR level */
377 guest_pcr_bit = host_pcr_bit;
379 switch (arch_compat) {
381 guest_pcr_bit = PCR_ARCH_205;
385 guest_pcr_bit = PCR_ARCH_206;
388 guest_pcr_bit = PCR_ARCH_207;
391 guest_pcr_bit = PCR_ARCH_300;
398 /* Check requested PCR bits don't exceed our capabilities */
399 if (guest_pcr_bit > host_pcr_bit)
402 spin_lock(&vc->lock);
403 vc->arch_compat = arch_compat;
404 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
405 vc->pcr = host_pcr_bit - guest_pcr_bit;
406 spin_unlock(&vc->lock);
411 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
415 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
416 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
417 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
418 for (r = 0; r < 16; ++r)
419 pr_err("r%2d = %.16lx r%d = %.16lx\n",
420 r, kvmppc_get_gpr(vcpu, r),
421 r+16, kvmppc_get_gpr(vcpu, r+16));
422 pr_err("ctr = %.16lx lr = %.16lx\n",
423 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
424 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
425 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
426 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
427 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
428 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
429 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
430 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
431 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
432 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
433 pr_err("fault dar = %.16lx dsisr = %.8x\n",
434 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
435 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
436 for (r = 0; r < vcpu->arch.slb_max; ++r)
437 pr_err(" ESID = %.16llx VSID = %.16llx\n",
438 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
439 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
440 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
441 vcpu->arch.last_inst);
444 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
446 return kvm_get_vcpu_by_id(kvm, id);
449 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
451 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
452 vpa->yield_count = cpu_to_be32(1);
455 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
456 unsigned long addr, unsigned long len)
458 /* check address is cacheline aligned */
459 if (addr & (L1_CACHE_BYTES - 1))
461 spin_lock(&vcpu->arch.vpa_update_lock);
462 if (v->next_gpa != addr || v->len != len) {
464 v->len = addr ? len : 0;
465 v->update_pending = 1;
467 spin_unlock(&vcpu->arch.vpa_update_lock);
471 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
480 static int vpa_is_registered(struct kvmppc_vpa *vpap)
482 if (vpap->update_pending)
483 return vpap->next_gpa != 0;
484 return vpap->pinned_addr != NULL;
487 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
489 unsigned long vcpuid, unsigned long vpa)
491 struct kvm *kvm = vcpu->kvm;
492 unsigned long len, nb;
494 struct kvm_vcpu *tvcpu;
497 struct kvmppc_vpa *vpap;
499 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
503 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
504 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
505 subfunc == H_VPA_REG_SLB) {
506 /* Registering new area - address must be cache-line aligned */
507 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
510 /* convert logical addr to kernel addr and read length */
511 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
514 if (subfunc == H_VPA_REG_VPA)
515 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
517 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
518 kvmppc_unpin_guest_page(kvm, va, vpa, false);
521 if (len > nb || len < sizeof(struct reg_vpa))
530 spin_lock(&tvcpu->arch.vpa_update_lock);
533 case H_VPA_REG_VPA: /* register VPA */
535 * The size of our lppaca is 1kB because of the way we align
536 * it for the guest to avoid crossing a 4kB boundary. We only
537 * use 640 bytes of the structure though, so we should accept
538 * clients that set a size of 640.
540 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
541 if (len < sizeof(struct lppaca))
543 vpap = &tvcpu->arch.vpa;
547 case H_VPA_REG_DTL: /* register DTL */
548 if (len < sizeof(struct dtl_entry))
550 len -= len % sizeof(struct dtl_entry);
552 /* Check that they have previously registered a VPA */
554 if (!vpa_is_registered(&tvcpu->arch.vpa))
557 vpap = &tvcpu->arch.dtl;
561 case H_VPA_REG_SLB: /* register SLB shadow buffer */
562 /* Check that they have previously registered a VPA */
564 if (!vpa_is_registered(&tvcpu->arch.vpa))
567 vpap = &tvcpu->arch.slb_shadow;
571 case H_VPA_DEREG_VPA: /* deregister VPA */
572 /* Check they don't still have a DTL or SLB buf registered */
574 if (vpa_is_registered(&tvcpu->arch.dtl) ||
575 vpa_is_registered(&tvcpu->arch.slb_shadow))
578 vpap = &tvcpu->arch.vpa;
582 case H_VPA_DEREG_DTL: /* deregister DTL */
583 vpap = &tvcpu->arch.dtl;
587 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
588 vpap = &tvcpu->arch.slb_shadow;
594 vpap->next_gpa = vpa;
596 vpap->update_pending = 1;
599 spin_unlock(&tvcpu->arch.vpa_update_lock);
604 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
606 struct kvm *kvm = vcpu->kvm;
612 * We need to pin the page pointed to by vpap->next_gpa,
613 * but we can't call kvmppc_pin_guest_page under the lock
614 * as it does get_user_pages() and down_read(). So we
615 * have to drop the lock, pin the page, then get the lock
616 * again and check that a new area didn't get registered
620 gpa = vpap->next_gpa;
621 spin_unlock(&vcpu->arch.vpa_update_lock);
625 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
626 spin_lock(&vcpu->arch.vpa_update_lock);
627 if (gpa == vpap->next_gpa)
629 /* sigh... unpin that one and try again */
631 kvmppc_unpin_guest_page(kvm, va, gpa, false);
634 vpap->update_pending = 0;
635 if (va && nb < vpap->len) {
637 * If it's now too short, it must be that userspace
638 * has changed the mappings underlying guest memory,
639 * so unregister the region.
641 kvmppc_unpin_guest_page(kvm, va, gpa, false);
644 if (vpap->pinned_addr)
645 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
648 vpap->pinned_addr = va;
651 vpap->pinned_end = va + vpap->len;
654 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
656 if (!(vcpu->arch.vpa.update_pending ||
657 vcpu->arch.slb_shadow.update_pending ||
658 vcpu->arch.dtl.update_pending))
661 spin_lock(&vcpu->arch.vpa_update_lock);
662 if (vcpu->arch.vpa.update_pending) {
663 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
664 if (vcpu->arch.vpa.pinned_addr)
665 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
667 if (vcpu->arch.dtl.update_pending) {
668 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
669 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
670 vcpu->arch.dtl_index = 0;
672 if (vcpu->arch.slb_shadow.update_pending)
673 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
674 spin_unlock(&vcpu->arch.vpa_update_lock);
678 * Return the accumulated stolen time for the vcore up until `now'.
679 * The caller should hold the vcore lock.
681 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
686 spin_lock_irqsave(&vc->stoltb_lock, flags);
688 if (vc->vcore_state != VCORE_INACTIVE &&
689 vc->preempt_tb != TB_NIL)
690 p += now - vc->preempt_tb;
691 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
695 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
696 struct kvmppc_vcore *vc)
698 struct dtl_entry *dt;
700 unsigned long stolen;
701 unsigned long core_stolen;
705 dt = vcpu->arch.dtl_ptr;
706 vpa = vcpu->arch.vpa.pinned_addr;
708 core_stolen = vcore_stolen_time(vc, now);
709 stolen = core_stolen - vcpu->arch.stolen_logged;
710 vcpu->arch.stolen_logged = core_stolen;
711 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
712 stolen += vcpu->arch.busy_stolen;
713 vcpu->arch.busy_stolen = 0;
714 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
717 memset(dt, 0, sizeof(struct dtl_entry));
718 dt->dispatch_reason = 7;
719 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
720 dt->timebase = cpu_to_be64(now + vc->tb_offset);
721 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
722 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
723 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
725 if (dt == vcpu->arch.dtl.pinned_end)
726 dt = vcpu->arch.dtl.pinned_addr;
727 vcpu->arch.dtl_ptr = dt;
728 /* order writing *dt vs. writing vpa->dtl_idx */
730 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
731 vcpu->arch.dtl.dirty = true;
734 /* See if there is a doorbell interrupt pending for a vcpu */
735 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
738 struct kvmppc_vcore *vc;
740 if (vcpu->arch.doorbell_request)
743 * Ensure that the read of vcore->dpdes comes after the read
744 * of vcpu->doorbell_request. This barrier matches the
745 * smp_wmb() in kvmppc_guest_entry_inject().
748 vc = vcpu->arch.vcore;
749 thr = vcpu->vcpu_id - vc->first_vcpuid;
750 return !!(vc->dpdes & (1 << thr));
753 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
755 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
757 if ((!vcpu->arch.vcore->arch_compat) &&
758 cpu_has_feature(CPU_FTR_ARCH_207S))
763 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
764 unsigned long resource, unsigned long value1,
765 unsigned long value2)
768 case H_SET_MODE_RESOURCE_SET_CIABR:
769 if (!kvmppc_power8_compatible(vcpu))
774 return H_UNSUPPORTED_FLAG_START;
775 /* Guests can't breakpoint the hypervisor */
776 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
778 vcpu->arch.ciabr = value1;
780 case H_SET_MODE_RESOURCE_SET_DAWR:
781 if (!kvmppc_power8_compatible(vcpu))
783 if (!ppc_breakpoint_available())
786 return H_UNSUPPORTED_FLAG_START;
787 if (value2 & DABRX_HYP)
789 vcpu->arch.dawr = value1;
790 vcpu->arch.dawrx = value2;
797 /* Copy guest memory in place - must reside within a single memslot */
798 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
801 struct kvm_memory_slot *to_memslot = NULL;
802 struct kvm_memory_slot *from_memslot = NULL;
803 unsigned long to_addr, from_addr;
806 /* Get HPA for from address */
807 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
810 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
813 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
814 if (kvm_is_error_hva(from_addr))
816 from_addr |= (from & (PAGE_SIZE - 1));
818 /* Get HPA for to address */
819 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
822 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
825 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
826 if (kvm_is_error_hva(to_addr))
828 to_addr |= (to & (PAGE_SIZE - 1));
831 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
835 mark_page_dirty(kvm, to >> PAGE_SHIFT);
839 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
840 unsigned long dest, unsigned long src)
842 u64 pg_sz = SZ_4K; /* 4K page size */
843 u64 pg_mask = SZ_4K - 1;
846 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
847 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
848 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
851 /* dest (and src if copy_page flag set) must be page aligned */
852 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
855 /* zero and/or copy the page as determined by the flags */
856 if (flags & H_COPY_PAGE) {
857 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
860 } else if (flags & H_ZERO_PAGE) {
861 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
866 /* We can ignore the remaining flags */
871 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
873 struct kvmppc_vcore *vcore = target->arch.vcore;
876 * We expect to have been called by the real mode handler
877 * (kvmppc_rm_h_confer()) which would have directly returned
878 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
879 * have useful work to do and should not confer) so we don't
883 spin_lock(&vcore->lock);
884 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
885 vcore->vcore_state != VCORE_INACTIVE &&
887 target = vcore->runner;
888 spin_unlock(&vcore->lock);
890 return kvm_vcpu_yield_to(target);
893 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
896 struct lppaca *lppaca;
898 spin_lock(&vcpu->arch.vpa_update_lock);
899 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
901 yield_count = be32_to_cpu(lppaca->yield_count);
902 spin_unlock(&vcpu->arch.vpa_update_lock);
906 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
908 unsigned long req = kvmppc_get_gpr(vcpu, 3);
909 unsigned long target, ret = H_SUCCESS;
911 struct kvm_vcpu *tvcpu;
914 if (req <= MAX_HCALL_OPCODE &&
915 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
922 target = kvmppc_get_gpr(vcpu, 4);
923 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
928 tvcpu->arch.prodded = 1;
930 if (tvcpu->arch.ceded)
931 kvmppc_fast_vcpu_kick_hv(tvcpu);
934 target = kvmppc_get_gpr(vcpu, 4);
937 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
942 yield_count = kvmppc_get_gpr(vcpu, 5);
943 if (kvmppc_get_yield_count(tvcpu) != yield_count)
945 kvm_arch_vcpu_yield_to(tvcpu);
948 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
949 kvmppc_get_gpr(vcpu, 5),
950 kvmppc_get_gpr(vcpu, 6));
953 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
956 idx = srcu_read_lock(&vcpu->kvm->srcu);
957 rc = kvmppc_rtas_hcall(vcpu);
958 srcu_read_unlock(&vcpu->kvm->srcu, idx);
965 /* Send the error out to userspace via KVM_RUN */
967 case H_LOGICAL_CI_LOAD:
968 ret = kvmppc_h_logical_ci_load(vcpu);
969 if (ret == H_TOO_HARD)
972 case H_LOGICAL_CI_STORE:
973 ret = kvmppc_h_logical_ci_store(vcpu);
974 if (ret == H_TOO_HARD)
978 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
979 kvmppc_get_gpr(vcpu, 5),
980 kvmppc_get_gpr(vcpu, 6),
981 kvmppc_get_gpr(vcpu, 7));
982 if (ret == H_TOO_HARD)
991 if (kvmppc_xics_enabled(vcpu)) {
992 if (xics_on_xive()) {
993 ret = H_NOT_AVAILABLE;
996 ret = kvmppc_xics_hcall(vcpu, req);
1001 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1004 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1005 kvmppc_get_gpr(vcpu, 5));
1007 #ifdef CONFIG_SPAPR_TCE_IOMMU
1009 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1010 kvmppc_get_gpr(vcpu, 5));
1011 if (ret == H_TOO_HARD)
1015 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1016 kvmppc_get_gpr(vcpu, 5),
1017 kvmppc_get_gpr(vcpu, 6));
1018 if (ret == H_TOO_HARD)
1021 case H_PUT_TCE_INDIRECT:
1022 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1023 kvmppc_get_gpr(vcpu, 5),
1024 kvmppc_get_gpr(vcpu, 6),
1025 kvmppc_get_gpr(vcpu, 7));
1026 if (ret == H_TOO_HARD)
1030 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1031 kvmppc_get_gpr(vcpu, 5),
1032 kvmppc_get_gpr(vcpu, 6),
1033 kvmppc_get_gpr(vcpu, 7));
1034 if (ret == H_TOO_HARD)
1039 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1043 case H_SET_PARTITION_TABLE:
1045 if (nesting_enabled(vcpu->kvm))
1046 ret = kvmhv_set_partition_table(vcpu);
1048 case H_ENTER_NESTED:
1050 if (!nesting_enabled(vcpu->kvm))
1052 ret = kvmhv_enter_nested_guest(vcpu);
1053 if (ret == H_INTERRUPT) {
1054 kvmppc_set_gpr(vcpu, 3, 0);
1055 vcpu->arch.hcall_needed = 0;
1057 } else if (ret == H_TOO_HARD) {
1058 kvmppc_set_gpr(vcpu, 3, 0);
1059 vcpu->arch.hcall_needed = 0;
1063 case H_TLB_INVALIDATE:
1065 if (nesting_enabled(vcpu->kvm))
1066 ret = kvmhv_do_nested_tlbie(vcpu);
1068 case H_COPY_TOFROM_GUEST:
1070 if (nesting_enabled(vcpu->kvm))
1071 ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1074 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1075 kvmppc_get_gpr(vcpu, 5),
1076 kvmppc_get_gpr(vcpu, 6));
1081 kvmppc_set_gpr(vcpu, 3, ret);
1082 vcpu->arch.hcall_needed = 0;
1083 return RESUME_GUEST;
1087 * Handle H_CEDE in the nested virtualization case where we haven't
1088 * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1089 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1090 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1092 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1094 vcpu->arch.shregs.msr |= MSR_EE;
1095 vcpu->arch.ceded = 1;
1097 if (vcpu->arch.prodded) {
1098 vcpu->arch.prodded = 0;
1100 vcpu->arch.ceded = 0;
1104 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1110 case H_REGISTER_VPA:
1112 case H_LOGICAL_CI_LOAD:
1113 case H_LOGICAL_CI_STORE:
1114 #ifdef CONFIG_KVM_XICS
1126 /* See if it's in the real-mode table */
1127 return kvmppc_hcall_impl_hv_realmode(cmd);
1130 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1131 struct kvm_vcpu *vcpu)
1135 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1138 * Fetch failed, so return to guest and
1139 * try executing it again.
1141 return RESUME_GUEST;
1144 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1145 run->exit_reason = KVM_EXIT_DEBUG;
1146 run->debug.arch.address = kvmppc_get_pc(vcpu);
1149 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1150 return RESUME_GUEST;
1154 static void do_nothing(void *x)
1158 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1160 int thr, cpu, pcpu, nthreads;
1162 unsigned long dpdes;
1164 nthreads = vcpu->kvm->arch.emul_smt_mode;
1166 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1167 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1168 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1172 * If the vcpu is currently running on a physical cpu thread,
1173 * interrupt it in order to pull it out of the guest briefly,
1174 * which will update its vcore->dpdes value.
1176 pcpu = READ_ONCE(v->cpu);
1178 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1179 if (kvmppc_doorbell_pending(v))
1186 * On POWER9, emulate doorbell-related instructions in order to
1187 * give the guest the illusion of running on a multi-threaded core.
1188 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1191 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1195 struct kvm *kvm = vcpu->kvm;
1196 struct kvm_vcpu *tvcpu;
1198 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1199 return RESUME_GUEST;
1200 if (get_op(inst) != 31)
1201 return EMULATE_FAIL;
1203 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1204 switch (get_xop(inst)) {
1205 case OP_31_XOP_MSGSNDP:
1206 arg = kvmppc_get_gpr(vcpu, rb);
1207 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1210 if (arg >= kvm->arch.emul_smt_mode)
1212 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1215 if (!tvcpu->arch.doorbell_request) {
1216 tvcpu->arch.doorbell_request = 1;
1217 kvmppc_fast_vcpu_kick_hv(tvcpu);
1220 case OP_31_XOP_MSGCLRP:
1221 arg = kvmppc_get_gpr(vcpu, rb);
1222 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1224 vcpu->arch.vcore->dpdes = 0;
1225 vcpu->arch.doorbell_request = 0;
1227 case OP_31_XOP_MFSPR:
1228 switch (get_sprn(inst)) {
1233 arg = kvmppc_read_dpdes(vcpu);
1236 return EMULATE_FAIL;
1238 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1241 return EMULATE_FAIL;
1243 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1244 return RESUME_GUEST;
1247 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1248 struct task_struct *tsk)
1250 int r = RESUME_HOST;
1252 vcpu->stat.sum_exits++;
1255 * This can happen if an interrupt occurs in the last stages
1256 * of guest entry or the first stages of guest exit (i.e. after
1257 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1258 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1259 * That can happen due to a bug, or due to a machine check
1260 * occurring at just the wrong time.
1262 if (vcpu->arch.shregs.msr & MSR_HV) {
1263 printk(KERN_EMERG "KVM trap in HV mode!\n");
1264 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1265 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1266 vcpu->arch.shregs.msr);
1267 kvmppc_dump_regs(vcpu);
1268 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1269 run->hw.hardware_exit_reason = vcpu->arch.trap;
1272 run->exit_reason = KVM_EXIT_UNKNOWN;
1273 run->ready_for_interrupt_injection = 1;
1274 switch (vcpu->arch.trap) {
1275 /* We're good on these - the host merely wanted to get our attention */
1276 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1277 vcpu->stat.dec_exits++;
1280 case BOOK3S_INTERRUPT_EXTERNAL:
1281 case BOOK3S_INTERRUPT_H_DOORBELL:
1282 case BOOK3S_INTERRUPT_H_VIRT:
1283 vcpu->stat.ext_intr_exits++;
1286 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1287 case BOOK3S_INTERRUPT_HMI:
1288 case BOOK3S_INTERRUPT_PERFMON:
1289 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1292 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1293 /* Print the MCE event to host console. */
1294 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1297 * If the guest can do FWNMI, exit to userspace so it can
1298 * deliver a FWNMI to the guest.
1299 * Otherwise we synthesize a machine check for the guest
1300 * so that it knows that the machine check occurred.
1302 if (!vcpu->kvm->arch.fwnmi_enabled) {
1303 ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1304 kvmppc_core_queue_machine_check(vcpu, flags);
1309 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1310 run->exit_reason = KVM_EXIT_NMI;
1311 run->hw.hardware_exit_reason = vcpu->arch.trap;
1312 /* Clear out the old NMI status from run->flags */
1313 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1314 /* Now set the NMI status */
1315 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1316 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1318 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1322 case BOOK3S_INTERRUPT_PROGRAM:
1326 * Normally program interrupts are delivered directly
1327 * to the guest by the hardware, but we can get here
1328 * as a result of a hypervisor emulation interrupt
1329 * (e40) getting turned into a 700 by BML RTAS.
1331 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1332 kvmppc_core_queue_program(vcpu, flags);
1336 case BOOK3S_INTERRUPT_SYSCALL:
1338 /* hcall - punt to userspace */
1341 /* hypercall with MSR_PR has already been handled in rmode,
1342 * and never reaches here.
1345 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1346 for (i = 0; i < 9; ++i)
1347 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1348 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1349 vcpu->arch.hcall_needed = 1;
1354 * We get these next two if the guest accesses a page which it thinks
1355 * it has mapped but which is not actually present, either because
1356 * it is for an emulated I/O device or because the corresonding
1357 * host page has been paged out. Any other HDSI/HISI interrupts
1358 * have been handled already.
1360 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1361 r = RESUME_PAGE_FAULT;
1363 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1364 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1365 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1366 DSISR_SRR1_MATCH_64S;
1367 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1368 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1369 r = RESUME_PAGE_FAULT;
1372 * This occurs if the guest executes an illegal instruction.
1373 * If the guest debug is disabled, generate a program interrupt
1374 * to the guest. If guest debug is enabled, we need to check
1375 * whether the instruction is a software breakpoint instruction.
1376 * Accordingly return to Guest or Host.
1378 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1379 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1380 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1381 swab32(vcpu->arch.emul_inst) :
1382 vcpu->arch.emul_inst;
1383 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1384 r = kvmppc_emulate_debug_inst(run, vcpu);
1386 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1391 * This occurs if the guest (kernel or userspace), does something that
1392 * is prohibited by HFSCR.
1393 * On POWER9, this could be a doorbell instruction that we need
1395 * Otherwise, we just generate a program interrupt to the guest.
1397 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1399 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1400 cpu_has_feature(CPU_FTR_ARCH_300))
1401 r = kvmppc_emulate_doorbell_instr(vcpu);
1402 if (r == EMULATE_FAIL) {
1403 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1408 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1409 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1411 * This occurs for various TM-related instructions that
1412 * we need to emulate on POWER9 DD2.2. We have already
1413 * handled the cases where the guest was in real-suspend
1414 * mode and was transitioning to transactional state.
1416 r = kvmhv_p9_tm_emulation(vcpu);
1420 case BOOK3S_INTERRUPT_HV_RM_HARD:
1421 r = RESUME_PASSTHROUGH;
1424 kvmppc_dump_regs(vcpu);
1425 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1426 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1427 vcpu->arch.shregs.msr);
1428 run->hw.hardware_exit_reason = vcpu->arch.trap;
1436 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1441 vcpu->stat.sum_exits++;
1444 * This can happen if an interrupt occurs in the last stages
1445 * of guest entry or the first stages of guest exit (i.e. after
1446 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1447 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1448 * That can happen due to a bug, or due to a machine check
1449 * occurring at just the wrong time.
1451 if (vcpu->arch.shregs.msr & MSR_HV) {
1452 pr_emerg("KVM trap in HV mode while nested!\n");
1453 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1454 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1455 vcpu->arch.shregs.msr);
1456 kvmppc_dump_regs(vcpu);
1459 switch (vcpu->arch.trap) {
1460 /* We're good on these - the host merely wanted to get our attention */
1461 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1462 vcpu->stat.dec_exits++;
1465 case BOOK3S_INTERRUPT_EXTERNAL:
1466 vcpu->stat.ext_intr_exits++;
1469 case BOOK3S_INTERRUPT_H_DOORBELL:
1470 case BOOK3S_INTERRUPT_H_VIRT:
1471 vcpu->stat.ext_intr_exits++;
1474 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1475 case BOOK3S_INTERRUPT_HMI:
1476 case BOOK3S_INTERRUPT_PERFMON:
1477 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1480 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1481 /* Pass the machine check to the L1 guest */
1483 /* Print the MCE event to host console. */
1484 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1487 * We get these next two if the guest accesses a page which it thinks
1488 * it has mapped but which is not actually present, either because
1489 * it is for an emulated I/O device or because the corresonding
1490 * host page has been paged out.
1492 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1493 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1494 r = kvmhv_nested_page_fault(run, vcpu);
1495 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1497 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1498 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1499 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1500 DSISR_SRR1_MATCH_64S;
1501 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1502 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1503 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1504 r = kvmhv_nested_page_fault(run, vcpu);
1505 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1508 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1509 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1511 * This occurs for various TM-related instructions that
1512 * we need to emulate on POWER9 DD2.2. We have already
1513 * handled the cases where the guest was in real-suspend
1514 * mode and was transitioning to transactional state.
1516 r = kvmhv_p9_tm_emulation(vcpu);
1520 case BOOK3S_INTERRUPT_HV_RM_HARD:
1521 vcpu->arch.trap = 0;
1523 if (!xics_on_xive())
1524 kvmppc_xics_rm_complete(vcpu, 0);
1534 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1535 struct kvm_sregs *sregs)
1539 memset(sregs, 0, sizeof(struct kvm_sregs));
1540 sregs->pvr = vcpu->arch.pvr;
1541 for (i = 0; i < vcpu->arch.slb_max; i++) {
1542 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1543 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1549 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1550 struct kvm_sregs *sregs)
1554 /* Only accept the same PVR as the host's, since we can't spoof it */
1555 if (sregs->pvr != vcpu->arch.pvr)
1559 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1560 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1561 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1562 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1566 vcpu->arch.slb_max = j;
1571 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1572 bool preserve_top32)
1574 struct kvm *kvm = vcpu->kvm;
1575 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1578 spin_lock(&vc->lock);
1580 * If ILE (interrupt little-endian) has changed, update the
1581 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1583 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1584 struct kvm_vcpu *vcpu;
1587 kvm_for_each_vcpu(i, vcpu, kvm) {
1588 if (vcpu->arch.vcore != vc)
1590 if (new_lpcr & LPCR_ILE)
1591 vcpu->arch.intr_msr |= MSR_LE;
1593 vcpu->arch.intr_msr &= ~MSR_LE;
1598 * Userspace can only modify DPFD (default prefetch depth),
1599 * ILE (interrupt little-endian) and TC (translation control).
1600 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1602 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1603 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1606 * On POWER9, allow userspace to enable large decrementer for the
1607 * guest, whether or not the host has it enabled.
1609 if (cpu_has_feature(CPU_FTR_ARCH_300))
1612 /* Broken 32-bit version of LPCR must not clear top bits */
1615 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1616 spin_unlock(&vc->lock);
1619 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1620 union kvmppc_one_reg *val)
1626 case KVM_REG_PPC_DEBUG_INST:
1627 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1629 case KVM_REG_PPC_HIOR:
1630 *val = get_reg_val(id, 0);
1632 case KVM_REG_PPC_DABR:
1633 *val = get_reg_val(id, vcpu->arch.dabr);
1635 case KVM_REG_PPC_DABRX:
1636 *val = get_reg_val(id, vcpu->arch.dabrx);
1638 case KVM_REG_PPC_DSCR:
1639 *val = get_reg_val(id, vcpu->arch.dscr);
1641 case KVM_REG_PPC_PURR:
1642 *val = get_reg_val(id, vcpu->arch.purr);
1644 case KVM_REG_PPC_SPURR:
1645 *val = get_reg_val(id, vcpu->arch.spurr);
1647 case KVM_REG_PPC_AMR:
1648 *val = get_reg_val(id, vcpu->arch.amr);
1650 case KVM_REG_PPC_UAMOR:
1651 *val = get_reg_val(id, vcpu->arch.uamor);
1653 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1654 i = id - KVM_REG_PPC_MMCR0;
1655 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1657 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1658 i = id - KVM_REG_PPC_PMC1;
1659 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1661 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1662 i = id - KVM_REG_PPC_SPMC1;
1663 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1665 case KVM_REG_PPC_SIAR:
1666 *val = get_reg_val(id, vcpu->arch.siar);
1668 case KVM_REG_PPC_SDAR:
1669 *val = get_reg_val(id, vcpu->arch.sdar);
1671 case KVM_REG_PPC_SIER:
1672 *val = get_reg_val(id, vcpu->arch.sier);
1674 case KVM_REG_PPC_IAMR:
1675 *val = get_reg_val(id, vcpu->arch.iamr);
1677 case KVM_REG_PPC_PSPB:
1678 *val = get_reg_val(id, vcpu->arch.pspb);
1680 case KVM_REG_PPC_DPDES:
1681 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1683 case KVM_REG_PPC_VTB:
1684 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1686 case KVM_REG_PPC_DAWR:
1687 *val = get_reg_val(id, vcpu->arch.dawr);
1689 case KVM_REG_PPC_DAWRX:
1690 *val = get_reg_val(id, vcpu->arch.dawrx);
1692 case KVM_REG_PPC_CIABR:
1693 *val = get_reg_val(id, vcpu->arch.ciabr);
1695 case KVM_REG_PPC_CSIGR:
1696 *val = get_reg_val(id, vcpu->arch.csigr);
1698 case KVM_REG_PPC_TACR:
1699 *val = get_reg_val(id, vcpu->arch.tacr);
1701 case KVM_REG_PPC_TCSCR:
1702 *val = get_reg_val(id, vcpu->arch.tcscr);
1704 case KVM_REG_PPC_PID:
1705 *val = get_reg_val(id, vcpu->arch.pid);
1707 case KVM_REG_PPC_ACOP:
1708 *val = get_reg_val(id, vcpu->arch.acop);
1710 case KVM_REG_PPC_WORT:
1711 *val = get_reg_val(id, vcpu->arch.wort);
1713 case KVM_REG_PPC_TIDR:
1714 *val = get_reg_val(id, vcpu->arch.tid);
1716 case KVM_REG_PPC_PSSCR:
1717 *val = get_reg_val(id, vcpu->arch.psscr);
1719 case KVM_REG_PPC_VPA_ADDR:
1720 spin_lock(&vcpu->arch.vpa_update_lock);
1721 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1722 spin_unlock(&vcpu->arch.vpa_update_lock);
1724 case KVM_REG_PPC_VPA_SLB:
1725 spin_lock(&vcpu->arch.vpa_update_lock);
1726 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1727 val->vpaval.length = vcpu->arch.slb_shadow.len;
1728 spin_unlock(&vcpu->arch.vpa_update_lock);
1730 case KVM_REG_PPC_VPA_DTL:
1731 spin_lock(&vcpu->arch.vpa_update_lock);
1732 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1733 val->vpaval.length = vcpu->arch.dtl.len;
1734 spin_unlock(&vcpu->arch.vpa_update_lock);
1736 case KVM_REG_PPC_TB_OFFSET:
1737 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1739 case KVM_REG_PPC_LPCR:
1740 case KVM_REG_PPC_LPCR_64:
1741 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1743 case KVM_REG_PPC_PPR:
1744 *val = get_reg_val(id, vcpu->arch.ppr);
1746 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1747 case KVM_REG_PPC_TFHAR:
1748 *val = get_reg_val(id, vcpu->arch.tfhar);
1750 case KVM_REG_PPC_TFIAR:
1751 *val = get_reg_val(id, vcpu->arch.tfiar);
1753 case KVM_REG_PPC_TEXASR:
1754 *val = get_reg_val(id, vcpu->arch.texasr);
1756 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1757 i = id - KVM_REG_PPC_TM_GPR0;
1758 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1760 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1763 i = id - KVM_REG_PPC_TM_VSR0;
1765 for (j = 0; j < TS_FPRWIDTH; j++)
1766 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1768 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1769 val->vval = vcpu->arch.vr_tm.vr[i-32];
1775 case KVM_REG_PPC_TM_CR:
1776 *val = get_reg_val(id, vcpu->arch.cr_tm);
1778 case KVM_REG_PPC_TM_XER:
1779 *val = get_reg_val(id, vcpu->arch.xer_tm);
1781 case KVM_REG_PPC_TM_LR:
1782 *val = get_reg_val(id, vcpu->arch.lr_tm);
1784 case KVM_REG_PPC_TM_CTR:
1785 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1787 case KVM_REG_PPC_TM_FPSCR:
1788 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1790 case KVM_REG_PPC_TM_AMR:
1791 *val = get_reg_val(id, vcpu->arch.amr_tm);
1793 case KVM_REG_PPC_TM_PPR:
1794 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1796 case KVM_REG_PPC_TM_VRSAVE:
1797 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1799 case KVM_REG_PPC_TM_VSCR:
1800 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1801 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1805 case KVM_REG_PPC_TM_DSCR:
1806 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1808 case KVM_REG_PPC_TM_TAR:
1809 *val = get_reg_val(id, vcpu->arch.tar_tm);
1812 case KVM_REG_PPC_ARCH_COMPAT:
1813 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1815 case KVM_REG_PPC_DEC_EXPIRY:
1816 *val = get_reg_val(id, vcpu->arch.dec_expires +
1817 vcpu->arch.vcore->tb_offset);
1819 case KVM_REG_PPC_ONLINE:
1820 *val = get_reg_val(id, vcpu->arch.online);
1822 case KVM_REG_PPC_PTCR:
1823 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1833 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1834 union kvmppc_one_reg *val)
1838 unsigned long addr, len;
1841 case KVM_REG_PPC_HIOR:
1842 /* Only allow this to be set to zero */
1843 if (set_reg_val(id, *val))
1846 case KVM_REG_PPC_DABR:
1847 vcpu->arch.dabr = set_reg_val(id, *val);
1849 case KVM_REG_PPC_DABRX:
1850 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1852 case KVM_REG_PPC_DSCR:
1853 vcpu->arch.dscr = set_reg_val(id, *val);
1855 case KVM_REG_PPC_PURR:
1856 vcpu->arch.purr = set_reg_val(id, *val);
1858 case KVM_REG_PPC_SPURR:
1859 vcpu->arch.spurr = set_reg_val(id, *val);
1861 case KVM_REG_PPC_AMR:
1862 vcpu->arch.amr = set_reg_val(id, *val);
1864 case KVM_REG_PPC_UAMOR:
1865 vcpu->arch.uamor = set_reg_val(id, *val);
1867 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1868 i = id - KVM_REG_PPC_MMCR0;
1869 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1871 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1872 i = id - KVM_REG_PPC_PMC1;
1873 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1875 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1876 i = id - KVM_REG_PPC_SPMC1;
1877 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1879 case KVM_REG_PPC_SIAR:
1880 vcpu->arch.siar = set_reg_val(id, *val);
1882 case KVM_REG_PPC_SDAR:
1883 vcpu->arch.sdar = set_reg_val(id, *val);
1885 case KVM_REG_PPC_SIER:
1886 vcpu->arch.sier = set_reg_val(id, *val);
1888 case KVM_REG_PPC_IAMR:
1889 vcpu->arch.iamr = set_reg_val(id, *val);
1891 case KVM_REG_PPC_PSPB:
1892 vcpu->arch.pspb = set_reg_val(id, *val);
1894 case KVM_REG_PPC_DPDES:
1895 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1897 case KVM_REG_PPC_VTB:
1898 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1900 case KVM_REG_PPC_DAWR:
1901 vcpu->arch.dawr = set_reg_val(id, *val);
1903 case KVM_REG_PPC_DAWRX:
1904 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1906 case KVM_REG_PPC_CIABR:
1907 vcpu->arch.ciabr = set_reg_val(id, *val);
1908 /* Don't allow setting breakpoints in hypervisor code */
1909 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1910 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1912 case KVM_REG_PPC_CSIGR:
1913 vcpu->arch.csigr = set_reg_val(id, *val);
1915 case KVM_REG_PPC_TACR:
1916 vcpu->arch.tacr = set_reg_val(id, *val);
1918 case KVM_REG_PPC_TCSCR:
1919 vcpu->arch.tcscr = set_reg_val(id, *val);
1921 case KVM_REG_PPC_PID:
1922 vcpu->arch.pid = set_reg_val(id, *val);
1924 case KVM_REG_PPC_ACOP:
1925 vcpu->arch.acop = set_reg_val(id, *val);
1927 case KVM_REG_PPC_WORT:
1928 vcpu->arch.wort = set_reg_val(id, *val);
1930 case KVM_REG_PPC_TIDR:
1931 vcpu->arch.tid = set_reg_val(id, *val);
1933 case KVM_REG_PPC_PSSCR:
1934 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1936 case KVM_REG_PPC_VPA_ADDR:
1937 addr = set_reg_val(id, *val);
1939 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1940 vcpu->arch.dtl.next_gpa))
1942 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1944 case KVM_REG_PPC_VPA_SLB:
1945 addr = val->vpaval.addr;
1946 len = val->vpaval.length;
1948 if (addr && !vcpu->arch.vpa.next_gpa)
1950 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1952 case KVM_REG_PPC_VPA_DTL:
1953 addr = val->vpaval.addr;
1954 len = val->vpaval.length;
1956 if (addr && (len < sizeof(struct dtl_entry) ||
1957 !vcpu->arch.vpa.next_gpa))
1959 len -= len % sizeof(struct dtl_entry);
1960 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1962 case KVM_REG_PPC_TB_OFFSET:
1963 /* round up to multiple of 2^24 */
1964 vcpu->arch.vcore->tb_offset =
1965 ALIGN(set_reg_val(id, *val), 1UL << 24);
1967 case KVM_REG_PPC_LPCR:
1968 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1970 case KVM_REG_PPC_LPCR_64:
1971 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1973 case KVM_REG_PPC_PPR:
1974 vcpu->arch.ppr = set_reg_val(id, *val);
1976 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1977 case KVM_REG_PPC_TFHAR:
1978 vcpu->arch.tfhar = set_reg_val(id, *val);
1980 case KVM_REG_PPC_TFIAR:
1981 vcpu->arch.tfiar = set_reg_val(id, *val);
1983 case KVM_REG_PPC_TEXASR:
1984 vcpu->arch.texasr = set_reg_val(id, *val);
1986 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1987 i = id - KVM_REG_PPC_TM_GPR0;
1988 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1990 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1993 i = id - KVM_REG_PPC_TM_VSR0;
1995 for (j = 0; j < TS_FPRWIDTH; j++)
1996 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1998 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1999 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2004 case KVM_REG_PPC_TM_CR:
2005 vcpu->arch.cr_tm = set_reg_val(id, *val);
2007 case KVM_REG_PPC_TM_XER:
2008 vcpu->arch.xer_tm = set_reg_val(id, *val);
2010 case KVM_REG_PPC_TM_LR:
2011 vcpu->arch.lr_tm = set_reg_val(id, *val);
2013 case KVM_REG_PPC_TM_CTR:
2014 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2016 case KVM_REG_PPC_TM_FPSCR:
2017 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2019 case KVM_REG_PPC_TM_AMR:
2020 vcpu->arch.amr_tm = set_reg_val(id, *val);
2022 case KVM_REG_PPC_TM_PPR:
2023 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2025 case KVM_REG_PPC_TM_VRSAVE:
2026 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2028 case KVM_REG_PPC_TM_VSCR:
2029 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2030 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2034 case KVM_REG_PPC_TM_DSCR:
2035 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2037 case KVM_REG_PPC_TM_TAR:
2038 vcpu->arch.tar_tm = set_reg_val(id, *val);
2041 case KVM_REG_PPC_ARCH_COMPAT:
2042 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2044 case KVM_REG_PPC_DEC_EXPIRY:
2045 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2046 vcpu->arch.vcore->tb_offset;
2048 case KVM_REG_PPC_ONLINE:
2049 i = set_reg_val(id, *val);
2050 if (i && !vcpu->arch.online)
2051 atomic_inc(&vcpu->arch.vcore->online_count);
2052 else if (!i && vcpu->arch.online)
2053 atomic_dec(&vcpu->arch.vcore->online_count);
2054 vcpu->arch.online = i;
2056 case KVM_REG_PPC_PTCR:
2057 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2068 * On POWER9, threads are independent and can be in different partitions.
2069 * Therefore we consider each thread to be a subcore.
2070 * There is a restriction that all threads have to be in the same
2071 * MMU mode (radix or HPT), unfortunately, but since we only support
2072 * HPT guests on a HPT host so far, that isn't an impediment yet.
2074 static int threads_per_vcore(struct kvm *kvm)
2076 if (kvm->arch.threads_indep)
2078 return threads_per_subcore;
2081 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2083 struct kvmppc_vcore *vcore;
2085 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2090 spin_lock_init(&vcore->lock);
2091 spin_lock_init(&vcore->stoltb_lock);
2092 init_swait_queue_head(&vcore->wq);
2093 vcore->preempt_tb = TB_NIL;
2094 vcore->lpcr = kvm->arch.lpcr;
2095 vcore->first_vcpuid = id;
2097 INIT_LIST_HEAD(&vcore->preempt_list);
2102 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2103 static struct debugfs_timings_element {
2107 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2108 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2109 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2110 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2111 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2114 #define N_TIMINGS (ARRAY_SIZE(timings))
2116 struct debugfs_timings_state {
2117 struct kvm_vcpu *vcpu;
2118 unsigned int buflen;
2119 char buf[N_TIMINGS * 100];
2122 static int debugfs_timings_open(struct inode *inode, struct file *file)
2124 struct kvm_vcpu *vcpu = inode->i_private;
2125 struct debugfs_timings_state *p;
2127 p = kzalloc(sizeof(*p), GFP_KERNEL);
2131 kvm_get_kvm(vcpu->kvm);
2133 file->private_data = p;
2135 return nonseekable_open(inode, file);
2138 static int debugfs_timings_release(struct inode *inode, struct file *file)
2140 struct debugfs_timings_state *p = file->private_data;
2142 kvm_put_kvm(p->vcpu->kvm);
2147 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2148 size_t len, loff_t *ppos)
2150 struct debugfs_timings_state *p = file->private_data;
2151 struct kvm_vcpu *vcpu = p->vcpu;
2153 struct kvmhv_tb_accumulator tb;
2162 buf_end = s + sizeof(p->buf);
2163 for (i = 0; i < N_TIMINGS; ++i) {
2164 struct kvmhv_tb_accumulator *acc;
2166 acc = (struct kvmhv_tb_accumulator *)
2167 ((unsigned long)vcpu + timings[i].offset);
2169 for (loops = 0; loops < 1000; ++loops) {
2170 count = acc->seqcount;
2175 if (count == acc->seqcount) {
2183 snprintf(s, buf_end - s, "%s: stuck\n",
2186 snprintf(s, buf_end - s,
2187 "%s: %llu %llu %llu %llu\n",
2188 timings[i].name, count / 2,
2189 tb_to_ns(tb.tb_total),
2190 tb_to_ns(tb.tb_min),
2191 tb_to_ns(tb.tb_max));
2194 p->buflen = s - p->buf;
2198 if (pos >= p->buflen)
2200 if (len > p->buflen - pos)
2201 len = p->buflen - pos;
2202 n = copy_to_user(buf, p->buf + pos, len);
2212 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2213 size_t len, loff_t *ppos)
2218 static const struct file_operations debugfs_timings_ops = {
2219 .owner = THIS_MODULE,
2220 .open = debugfs_timings_open,
2221 .release = debugfs_timings_release,
2222 .read = debugfs_timings_read,
2223 .write = debugfs_timings_write,
2224 .llseek = generic_file_llseek,
2227 /* Create a debugfs directory for the vcpu */
2228 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2231 struct kvm *kvm = vcpu->kvm;
2233 snprintf(buf, sizeof(buf), "vcpu%u", id);
2234 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2236 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2237 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2239 vcpu->arch.debugfs_timings =
2240 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2241 vcpu, &debugfs_timings_ops);
2244 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2245 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2248 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2250 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2253 struct kvm_vcpu *vcpu;
2256 struct kvmppc_vcore *vcore;
2259 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2263 err = kvm_vcpu_init(vcpu, kvm, id);
2267 vcpu->arch.shared = &vcpu->arch.shregs;
2268 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2270 * The shared struct is never shared on HV,
2271 * so we can always use host endianness
2273 #ifdef __BIG_ENDIAN__
2274 vcpu->arch.shared_big_endian = true;
2276 vcpu->arch.shared_big_endian = false;
2279 vcpu->arch.mmcr[0] = MMCR0_FC;
2280 vcpu->arch.ctrl = CTRL_RUNLATCH;
2281 /* default to host PVR, since we can't spoof it */
2282 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2283 spin_lock_init(&vcpu->arch.vpa_update_lock);
2284 spin_lock_init(&vcpu->arch.tbacct_lock);
2285 vcpu->arch.busy_preempt = TB_NIL;
2286 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2289 * Set the default HFSCR for the guest from the host value.
2290 * This value is only used on POWER9.
2291 * On POWER9, we want to virtualize the doorbell facility, so we
2292 * don't set the HFSCR_MSGP bit, and that causes those instructions
2293 * to trap and then we emulate them.
2295 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2296 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2297 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2298 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2299 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2300 vcpu->arch.hfscr |= HFSCR_TM;
2302 if (cpu_has_feature(CPU_FTR_TM_COMP))
2303 vcpu->arch.hfscr |= HFSCR_TM;
2305 kvmppc_mmu_book3s_hv_init(vcpu);
2307 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2309 init_waitqueue_head(&vcpu->arch.cpu_run);
2311 mutex_lock(&kvm->lock);
2314 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2315 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2316 pr_devel("KVM: VCPU ID too high\n");
2317 core = KVM_MAX_VCORES;
2319 BUG_ON(kvm->arch.smt_mode != 1);
2320 core = kvmppc_pack_vcpu_id(kvm, id);
2323 core = id / kvm->arch.smt_mode;
2325 if (core < KVM_MAX_VCORES) {
2326 vcore = kvm->arch.vcores[core];
2327 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2328 pr_devel("KVM: collision on id %u", id);
2330 } else if (!vcore) {
2332 * Take mmu_setup_lock for mutual exclusion
2333 * with kvmppc_update_lpcr().
2336 vcore = kvmppc_vcore_create(kvm,
2337 id & ~(kvm->arch.smt_mode - 1));
2338 mutex_lock(&kvm->arch.mmu_setup_lock);
2339 kvm->arch.vcores[core] = vcore;
2340 kvm->arch.online_vcores++;
2341 mutex_unlock(&kvm->arch.mmu_setup_lock);
2344 mutex_unlock(&kvm->lock);
2349 spin_lock(&vcore->lock);
2350 ++vcore->num_threads;
2351 spin_unlock(&vcore->lock);
2352 vcpu->arch.vcore = vcore;
2353 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2354 vcpu->arch.thread_cpu = -1;
2355 vcpu->arch.prev_cpu = -1;
2357 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2358 kvmppc_sanity_check(vcpu);
2360 debugfs_vcpu_init(vcpu, id);
2365 kmem_cache_free(kvm_vcpu_cache, vcpu);
2367 return ERR_PTR(err);
2370 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2371 unsigned long flags)
2378 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2380 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2382 * On POWER8 (or POWER7), the threading mode is "strict",
2383 * so we pack smt_mode vcpus per vcore.
2385 if (smt_mode > threads_per_subcore)
2389 * On POWER9, the threading mode is "loose",
2390 * so each vcpu gets its own vcore.
2395 mutex_lock(&kvm->lock);
2397 if (!kvm->arch.online_vcores) {
2398 kvm->arch.smt_mode = smt_mode;
2399 kvm->arch.emul_smt_mode = esmt;
2402 mutex_unlock(&kvm->lock);
2407 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2409 if (vpa->pinned_addr)
2410 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2414 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2416 spin_lock(&vcpu->arch.vpa_update_lock);
2417 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2418 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2419 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2420 spin_unlock(&vcpu->arch.vpa_update_lock);
2421 kvm_vcpu_uninit(vcpu);
2422 kmem_cache_free(kvm_vcpu_cache, vcpu);
2425 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2427 /* Indicate we want to get back into the guest */
2431 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2433 unsigned long dec_nsec, now;
2436 if (now > vcpu->arch.dec_expires) {
2437 /* decrementer has already gone negative */
2438 kvmppc_core_queue_dec(vcpu);
2439 kvmppc_core_prepare_to_enter(vcpu);
2442 dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2443 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2444 vcpu->arch.timer_running = 1;
2447 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2449 vcpu->arch.ceded = 0;
2450 if (vcpu->arch.timer_running) {
2451 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2452 vcpu->arch.timer_running = 0;
2456 extern int __kvmppc_vcore_entry(void);
2458 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2459 struct kvm_vcpu *vcpu)
2463 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2465 spin_lock_irq(&vcpu->arch.tbacct_lock);
2467 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2468 vcpu->arch.stolen_logged;
2469 vcpu->arch.busy_preempt = now;
2470 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2471 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2473 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2476 static int kvmppc_grab_hwthread(int cpu)
2478 struct paca_struct *tpaca;
2479 long timeout = 10000;
2481 tpaca = paca_ptrs[cpu];
2483 /* Ensure the thread won't go into the kernel if it wakes */
2484 tpaca->kvm_hstate.kvm_vcpu = NULL;
2485 tpaca->kvm_hstate.kvm_vcore = NULL;
2486 tpaca->kvm_hstate.napping = 0;
2488 tpaca->kvm_hstate.hwthread_req = 1;
2491 * If the thread is already executing in the kernel (e.g. handling
2492 * a stray interrupt), wait for it to get back to nap mode.
2493 * The smp_mb() is to ensure that our setting of hwthread_req
2494 * is visible before we look at hwthread_state, so if this
2495 * races with the code at system_reset_pSeries and the thread
2496 * misses our setting of hwthread_req, we are sure to see its
2497 * setting of hwthread_state, and vice versa.
2500 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2501 if (--timeout <= 0) {
2502 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2510 static void kvmppc_release_hwthread(int cpu)
2512 struct paca_struct *tpaca;
2514 tpaca = paca_ptrs[cpu];
2515 tpaca->kvm_hstate.hwthread_req = 0;
2516 tpaca->kvm_hstate.kvm_vcpu = NULL;
2517 tpaca->kvm_hstate.kvm_vcore = NULL;
2518 tpaca->kvm_hstate.kvm_split_mode = NULL;
2521 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2523 struct kvm_nested_guest *nested = vcpu->arch.nested;
2524 cpumask_t *cpu_in_guest;
2527 cpu = cpu_first_thread_sibling(cpu);
2529 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2530 cpu_in_guest = &nested->cpu_in_guest;
2532 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2533 cpu_in_guest = &kvm->arch.cpu_in_guest;
2536 * Make sure setting of bit in need_tlb_flush precedes
2537 * testing of cpu_in_guest bits. The matching barrier on
2538 * the other side is the first smp_mb() in kvmppc_run_core().
2541 for (i = 0; i < threads_per_core; ++i)
2542 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2543 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2546 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2548 struct kvm_nested_guest *nested = vcpu->arch.nested;
2549 struct kvm *kvm = vcpu->kvm;
2552 if (!cpu_has_feature(CPU_FTR_HVMODE))
2556 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2558 prev_cpu = vcpu->arch.prev_cpu;
2561 * With radix, the guest can do TLB invalidations itself,
2562 * and it could choose to use the local form (tlbiel) if
2563 * it is invalidating a translation that has only ever been
2564 * used on one vcpu. However, that doesn't mean it has
2565 * only ever been used on one physical cpu, since vcpus
2566 * can move around between pcpus. To cope with this, when
2567 * a vcpu moves from one pcpu to another, we need to tell
2568 * any vcpus running on the same core as this vcpu previously
2569 * ran to flush the TLB. The TLB is shared between threads,
2570 * so we use a single bit in .need_tlb_flush for all 4 threads.
2572 if (prev_cpu != pcpu) {
2573 if (prev_cpu >= 0 &&
2574 cpu_first_thread_sibling(prev_cpu) !=
2575 cpu_first_thread_sibling(pcpu))
2576 radix_flush_cpu(kvm, prev_cpu, vcpu);
2578 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2580 vcpu->arch.prev_cpu = pcpu;
2584 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2587 struct paca_struct *tpaca;
2588 struct kvm *kvm = vc->kvm;
2592 if (vcpu->arch.timer_running) {
2593 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2594 vcpu->arch.timer_running = 0;
2596 cpu += vcpu->arch.ptid;
2597 vcpu->cpu = vc->pcpu;
2598 vcpu->arch.thread_cpu = cpu;
2599 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2601 tpaca = paca_ptrs[cpu];
2602 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2603 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2604 tpaca->kvm_hstate.fake_suspend = 0;
2605 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2607 tpaca->kvm_hstate.kvm_vcore = vc;
2608 if (cpu != smp_processor_id())
2609 kvmppc_ipi_thread(cpu);
2612 static void kvmppc_wait_for_nap(int n_threads)
2614 int cpu = smp_processor_id();
2619 for (loops = 0; loops < 1000000; ++loops) {
2621 * Check if all threads are finished.
2622 * We set the vcore pointer when starting a thread
2623 * and the thread clears it when finished, so we look
2624 * for any threads that still have a non-NULL vcore ptr.
2626 for (i = 1; i < n_threads; ++i)
2627 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2629 if (i == n_threads) {
2636 for (i = 1; i < n_threads; ++i)
2637 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2638 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2642 * Check that we are on thread 0 and that any other threads in
2643 * this core are off-line. Then grab the threads so they can't
2646 static int on_primary_thread(void)
2648 int cpu = smp_processor_id();
2651 /* Are we on a primary subcore? */
2652 if (cpu_thread_in_subcore(cpu))
2656 while (++thr < threads_per_subcore)
2657 if (cpu_online(cpu + thr))
2660 /* Grab all hw threads so they can't go into the kernel */
2661 for (thr = 1; thr < threads_per_subcore; ++thr) {
2662 if (kvmppc_grab_hwthread(cpu + thr)) {
2663 /* Couldn't grab one; let the others go */
2665 kvmppc_release_hwthread(cpu + thr);
2666 } while (--thr > 0);
2674 * A list of virtual cores for each physical CPU.
2675 * These are vcores that could run but their runner VCPU tasks are
2676 * (or may be) preempted.
2678 struct preempted_vcore_list {
2679 struct list_head list;
2683 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2685 static void init_vcore_lists(void)
2689 for_each_possible_cpu(cpu) {
2690 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2691 spin_lock_init(&lp->lock);
2692 INIT_LIST_HEAD(&lp->list);
2696 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2698 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2700 vc->vcore_state = VCORE_PREEMPT;
2701 vc->pcpu = smp_processor_id();
2702 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2703 spin_lock(&lp->lock);
2704 list_add_tail(&vc->preempt_list, &lp->list);
2705 spin_unlock(&lp->lock);
2708 /* Start accumulating stolen time */
2709 kvmppc_core_start_stolen(vc);
2712 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2714 struct preempted_vcore_list *lp;
2716 kvmppc_core_end_stolen(vc);
2717 if (!list_empty(&vc->preempt_list)) {
2718 lp = &per_cpu(preempted_vcores, vc->pcpu);
2719 spin_lock(&lp->lock);
2720 list_del_init(&vc->preempt_list);
2721 spin_unlock(&lp->lock);
2723 vc->vcore_state = VCORE_INACTIVE;
2727 * This stores information about the virtual cores currently
2728 * assigned to a physical core.
2732 int max_subcore_threads;
2734 int subcore_threads[MAX_SUBCORES];
2735 struct kvmppc_vcore *vc[MAX_SUBCORES];
2739 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2740 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2742 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2744 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2746 memset(cip, 0, sizeof(*cip));
2747 cip->n_subcores = 1;
2748 cip->max_subcore_threads = vc->num_threads;
2749 cip->total_threads = vc->num_threads;
2750 cip->subcore_threads[0] = vc->num_threads;
2754 static bool subcore_config_ok(int n_subcores, int n_threads)
2757 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2758 * split-core mode, with one thread per subcore.
2760 if (cpu_has_feature(CPU_FTR_ARCH_300))
2761 return n_subcores <= 4 && n_threads == 1;
2763 /* On POWER8, can only dynamically split if unsplit to begin with */
2764 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2766 if (n_subcores > MAX_SUBCORES)
2768 if (n_subcores > 1) {
2769 if (!(dynamic_mt_modes & 2))
2771 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2775 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2778 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2780 vc->entry_exit_map = 0;
2782 vc->napping_threads = 0;
2783 vc->conferring_threads = 0;
2784 vc->tb_offset_applied = 0;
2787 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2789 int n_threads = vc->num_threads;
2792 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2795 /* In one_vm_per_core mode, require all vcores to be from the same vm */
2796 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2799 /* Some POWER9 chips require all threads to be in the same MMU mode */
2800 if (no_mixing_hpt_and_radix &&
2801 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2804 if (n_threads < cip->max_subcore_threads)
2805 n_threads = cip->max_subcore_threads;
2806 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2808 cip->max_subcore_threads = n_threads;
2810 sub = cip->n_subcores;
2812 cip->total_threads += vc->num_threads;
2813 cip->subcore_threads[sub] = vc->num_threads;
2815 init_vcore_to_run(vc);
2816 list_del_init(&vc->preempt_list);
2822 * Work out whether it is possible to piggyback the execution of
2823 * vcore *pvc onto the execution of the other vcores described in *cip.
2825 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2828 if (cip->total_threads + pvc->num_threads > target_threads)
2831 return can_dynamic_split(pvc, cip);
2834 static void prepare_threads(struct kvmppc_vcore *vc)
2837 struct kvm_vcpu *vcpu;
2839 for_each_runnable_thread(i, vcpu, vc) {
2840 if (signal_pending(vcpu->arch.run_task))
2841 vcpu->arch.ret = -EINTR;
2842 else if (vcpu->arch.vpa.update_pending ||
2843 vcpu->arch.slb_shadow.update_pending ||
2844 vcpu->arch.dtl.update_pending)
2845 vcpu->arch.ret = RESUME_GUEST;
2848 kvmppc_remove_runnable(vc, vcpu);
2849 wake_up(&vcpu->arch.cpu_run);
2853 static void collect_piggybacks(struct core_info *cip, int target_threads)
2855 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2856 struct kvmppc_vcore *pvc, *vcnext;
2858 spin_lock(&lp->lock);
2859 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2860 if (!spin_trylock(&pvc->lock))
2862 prepare_threads(pvc);
2863 if (!pvc->n_runnable) {
2864 list_del_init(&pvc->preempt_list);
2865 if (pvc->runner == NULL) {
2866 pvc->vcore_state = VCORE_INACTIVE;
2867 kvmppc_core_end_stolen(pvc);
2869 spin_unlock(&pvc->lock);
2872 if (!can_piggyback(pvc, cip, target_threads)) {
2873 spin_unlock(&pvc->lock);
2876 kvmppc_core_end_stolen(pvc);
2877 pvc->vcore_state = VCORE_PIGGYBACK;
2878 if (cip->total_threads >= target_threads)
2881 spin_unlock(&lp->lock);
2884 static bool recheck_signals(struct core_info *cip)
2887 struct kvm_vcpu *vcpu;
2889 for (sub = 0; sub < cip->n_subcores; ++sub)
2890 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2891 if (signal_pending(vcpu->arch.run_task))
2896 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2898 int still_running = 0, i;
2901 struct kvm_vcpu *vcpu;
2903 spin_lock(&vc->lock);
2905 for_each_runnable_thread(i, vcpu, vc) {
2907 * It's safe to unlock the vcore in the loop here, because
2908 * for_each_runnable_thread() is safe against removal of
2909 * the vcpu, and the vcore state is VCORE_EXITING here,
2910 * so any vcpus becoming runnable will have their arch.trap
2911 * set to zero and can't actually run in the guest.
2913 spin_unlock(&vc->lock);
2914 /* cancel pending dec exception if dec is positive */
2915 if (now < vcpu->arch.dec_expires &&
2916 kvmppc_core_pending_dec(vcpu))
2917 kvmppc_core_dequeue_dec(vcpu);
2919 trace_kvm_guest_exit(vcpu);
2922 if (vcpu->arch.trap)
2923 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2924 vcpu->arch.run_task);
2926 vcpu->arch.ret = ret;
2927 vcpu->arch.trap = 0;
2929 spin_lock(&vc->lock);
2930 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2931 if (vcpu->arch.pending_exceptions)
2932 kvmppc_core_prepare_to_enter(vcpu);
2933 if (vcpu->arch.ceded)
2934 kvmppc_set_timer(vcpu);
2938 kvmppc_remove_runnable(vc, vcpu);
2939 wake_up(&vcpu->arch.cpu_run);
2943 if (still_running > 0) {
2944 kvmppc_vcore_preempt(vc);
2945 } else if (vc->runner) {
2946 vc->vcore_state = VCORE_PREEMPT;
2947 kvmppc_core_start_stolen(vc);
2949 vc->vcore_state = VCORE_INACTIVE;
2951 if (vc->n_runnable > 0 && vc->runner == NULL) {
2952 /* make sure there's a candidate runner awake */
2954 vcpu = next_runnable_thread(vc, &i);
2955 wake_up(&vcpu->arch.cpu_run);
2958 spin_unlock(&vc->lock);
2962 * Clear core from the list of active host cores as we are about to
2963 * enter the guest. Only do this if it is the primary thread of the
2964 * core (not if a subcore) that is entering the guest.
2966 static inline int kvmppc_clear_host_core(unsigned int cpu)
2970 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2973 * Memory barrier can be omitted here as we will do a smp_wmb()
2974 * later in kvmppc_start_thread and we need ensure that state is
2975 * visible to other CPUs only after we enter guest.
2977 core = cpu >> threads_shift;
2978 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2983 * Advertise this core as an active host core since we exited the guest
2984 * Only need to do this if it is the primary thread of the core that is
2987 static inline int kvmppc_set_host_core(unsigned int cpu)
2991 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2995 * Memory barrier can be omitted here because we do a spin_unlock
2996 * immediately after this which provides the memory barrier.
2998 core = cpu >> threads_shift;
2999 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3003 static void set_irq_happened(int trap)
3006 case BOOK3S_INTERRUPT_EXTERNAL:
3007 local_paca->irq_happened |= PACA_IRQ_EE;
3009 case BOOK3S_INTERRUPT_H_DOORBELL:
3010 local_paca->irq_happened |= PACA_IRQ_DBELL;
3012 case BOOK3S_INTERRUPT_HMI:
3013 local_paca->irq_happened |= PACA_IRQ_HMI;
3015 case BOOK3S_INTERRUPT_SYSTEM_RESET:
3016 replay_system_reset();
3022 * Run a set of guest threads on a physical core.
3023 * Called with vc->lock held.
3025 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3027 struct kvm_vcpu *vcpu;
3030 struct core_info core_info;
3031 struct kvmppc_vcore *pvc;
3032 struct kvm_split_mode split_info, *sip;
3033 int split, subcore_size, active;
3036 unsigned long cmd_bit, stat_bit;
3039 int controlled_threads;
3045 * Remove from the list any threads that have a signal pending
3046 * or need a VPA update done
3048 prepare_threads(vc);
3050 /* if the runner is no longer runnable, let the caller pick a new one */
3051 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3057 init_vcore_to_run(vc);
3058 vc->preempt_tb = TB_NIL;
3061 * Number of threads that we will be controlling: the same as
3062 * the number of threads per subcore, except on POWER9,
3063 * where it's 1 because the threads are (mostly) independent.
3065 controlled_threads = threads_per_vcore(vc->kvm);
3068 * Make sure we are running on primary threads, and that secondary
3069 * threads are offline. Also check if the number of threads in this
3070 * guest are greater than the current system threads per guest.
3071 * On POWER9, we need to be not in independent-threads mode if
3072 * this is a HPT guest on a radix host machine where the
3073 * CPU threads may not be in different MMU modes.
3075 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3076 !kvm_is_radix(vc->kvm);
3077 if (((controlled_threads > 1) &&
3078 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3079 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3080 for_each_runnable_thread(i, vcpu, vc) {
3081 vcpu->arch.ret = -EBUSY;
3082 kvmppc_remove_runnable(vc, vcpu);
3083 wake_up(&vcpu->arch.cpu_run);
3089 * See if we could run any other vcores on the physical core
3090 * along with this one.
3092 init_core_info(&core_info, vc);
3093 pcpu = smp_processor_id();
3094 target_threads = controlled_threads;
3095 if (target_smt_mode && target_smt_mode < target_threads)
3096 target_threads = target_smt_mode;
3097 if (vc->num_threads < target_threads)
3098 collect_piggybacks(&core_info, target_threads);
3101 * On radix, arrange for TLB flushing if necessary.
3102 * This has to be done before disabling interrupts since
3103 * it uses smp_call_function().
3105 pcpu = smp_processor_id();
3106 if (kvm_is_radix(vc->kvm)) {
3107 for (sub = 0; sub < core_info.n_subcores; ++sub)
3108 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3109 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3113 * Hard-disable interrupts, and check resched flag and signals.
3114 * If we need to reschedule or deliver a signal, clean up
3115 * and return without going into the guest(s).
3116 * If the mmu_ready flag has been cleared, don't go into the
3117 * guest because that means a HPT resize operation is in progress.
3119 local_irq_disable();
3121 if (lazy_irq_pending() || need_resched() ||
3122 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3124 vc->vcore_state = VCORE_INACTIVE;
3125 /* Unlock all except the primary vcore */
3126 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3127 pvc = core_info.vc[sub];
3128 /* Put back on to the preempted vcores list */
3129 kvmppc_vcore_preempt(pvc);
3130 spin_unlock(&pvc->lock);
3132 for (i = 0; i < controlled_threads; ++i)
3133 kvmppc_release_hwthread(pcpu + i);
3137 kvmppc_clear_host_core(pcpu);
3139 /* Decide on micro-threading (split-core) mode */
3140 subcore_size = threads_per_subcore;
3141 cmd_bit = stat_bit = 0;
3142 split = core_info.n_subcores;
3144 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3145 && !cpu_has_feature(CPU_FTR_ARCH_300);
3147 if (split > 1 || hpt_on_radix) {
3149 memset(&split_info, 0, sizeof(split_info));
3150 for (sub = 0; sub < core_info.n_subcores; ++sub)
3151 split_info.vc[sub] = core_info.vc[sub];
3154 if (split == 2 && (dynamic_mt_modes & 2)) {
3155 cmd_bit = HID0_POWER8_1TO2LPAR;
3156 stat_bit = HID0_POWER8_2LPARMODE;
3159 cmd_bit = HID0_POWER8_1TO4LPAR;
3160 stat_bit = HID0_POWER8_4LPARMODE;
3162 subcore_size = MAX_SMT_THREADS / split;
3163 split_info.rpr = mfspr(SPRN_RPR);
3164 split_info.pmmar = mfspr(SPRN_PMMAR);
3165 split_info.ldbar = mfspr(SPRN_LDBAR);
3166 split_info.subcore_size = subcore_size;
3168 split_info.subcore_size = 1;
3170 /* Use the split_info for LPCR/LPIDR changes */
3171 split_info.lpcr_req = vc->lpcr;
3172 split_info.lpidr_req = vc->kvm->arch.lpid;
3173 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3174 split_info.do_set = 1;
3178 /* order writes to split_info before kvm_split_mode pointer */
3182 for (thr = 0; thr < controlled_threads; ++thr) {
3183 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3185 paca->kvm_hstate.tid = thr;
3186 paca->kvm_hstate.napping = 0;
3187 paca->kvm_hstate.kvm_split_mode = sip;
3190 /* Initiate micro-threading (split-core) on POWER8 if required */
3192 unsigned long hid0 = mfspr(SPRN_HID0);
3194 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3196 mtspr(SPRN_HID0, hid0);
3199 hid0 = mfspr(SPRN_HID0);
3200 if (hid0 & stat_bit)
3207 * On POWER8, set RWMR register.
3208 * Since it only affects PURR and SPURR, it doesn't affect
3209 * the host, so we don't save/restore the host value.
3212 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3213 int n_online = atomic_read(&vc->online_count);
3216 * Use the 8-thread value if we're doing split-core
3217 * or if the vcore's online count looks bogus.
3219 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3220 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3221 rwmr_val = p8_rwmr_values[n_online];
3222 mtspr(SPRN_RWMR, rwmr_val);
3225 /* Start all the threads */
3227 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3228 thr = is_power8 ? subcore_thread_map[sub] : sub;
3231 pvc = core_info.vc[sub];
3232 pvc->pcpu = pcpu + thr;
3233 for_each_runnable_thread(i, vcpu, pvc) {
3234 kvmppc_start_thread(vcpu, pvc);
3235 kvmppc_create_dtl_entry(vcpu, pvc);
3236 trace_kvm_guest_enter(vcpu);
3237 if (!vcpu->arch.ptid)
3239 active |= 1 << (thr + vcpu->arch.ptid);
3242 * We need to start the first thread of each subcore
3243 * even if it doesn't have a vcpu.
3246 kvmppc_start_thread(NULL, pvc);
3250 * Ensure that split_info.do_nap is set after setting
3251 * the vcore pointer in the PACA of the secondaries.
3256 * When doing micro-threading, poke the inactive threads as well.
3257 * This gets them to the nap instruction after kvm_do_nap,
3258 * which reduces the time taken to unsplit later.
3259 * For POWER9 HPT guest on radix host, we need all the secondary
3260 * threads woken up so they can do the LPCR/LPIDR change.
3262 if (cmd_bit || hpt_on_radix) {
3263 split_info.do_nap = 1; /* ask secondaries to nap when done */
3264 for (thr = 1; thr < threads_per_subcore; ++thr)
3265 if (!(active & (1 << thr)))
3266 kvmppc_ipi_thread(pcpu + thr);
3269 vc->vcore_state = VCORE_RUNNING;
3272 trace_kvmppc_run_core(vc, 0);
3274 for (sub = 0; sub < core_info.n_subcores; ++sub)
3275 spin_unlock(&core_info.vc[sub]->lock);
3277 guest_enter_irqoff();
3279 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3281 this_cpu_disable_ftrace();
3284 * Interrupts will be enabled once we get into the guest,
3285 * so tell lockdep that we're about to enable interrupts.
3287 trace_hardirqs_on();
3289 trap = __kvmppc_vcore_entry();
3291 trace_hardirqs_off();
3293 this_cpu_enable_ftrace();
3295 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3297 set_irq_happened(trap);
3299 spin_lock(&vc->lock);
3300 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3301 vc->vcore_state = VCORE_EXITING;
3303 /* wait for secondary threads to finish writing their state to memory */
3304 kvmppc_wait_for_nap(controlled_threads);
3306 /* Return to whole-core mode if we split the core earlier */
3308 unsigned long hid0 = mfspr(SPRN_HID0);
3309 unsigned long loops = 0;
3311 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3312 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3314 mtspr(SPRN_HID0, hid0);
3317 hid0 = mfspr(SPRN_HID0);
3318 if (!(hid0 & stat_bit))
3323 } else if (hpt_on_radix) {
3324 /* Wait for all threads to have seen final sync */
3325 for (thr = 1; thr < controlled_threads; ++thr) {
3326 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3328 while (paca->kvm_hstate.kvm_split_mode) {
3335 split_info.do_nap = 0;
3337 kvmppc_set_host_core(pcpu);
3342 /* Let secondaries go back to the offline loop */
3343 for (i = 0; i < controlled_threads; ++i) {
3344 kvmppc_release_hwthread(pcpu + i);
3345 if (sip && sip->napped[i])
3346 kvmppc_ipi_thread(pcpu + i);
3347 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3350 spin_unlock(&vc->lock);
3352 /* make sure updates to secondary vcpu structs are visible now */
3357 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3358 pvc = core_info.vc[sub];
3359 post_guest_process(pvc, pvc == vc);
3362 spin_lock(&vc->lock);
3365 vc->vcore_state = VCORE_INACTIVE;
3366 trace_kvmppc_run_core(vc, 1);
3370 * Load up hypervisor-mode registers on P9.
3372 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3375 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3377 u64 tb, purr, spurr;
3379 unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3380 unsigned long host_ciabr = mfspr(SPRN_CIABR);
3381 unsigned long host_dawr = mfspr(SPRN_DAWR);
3382 unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3383 unsigned long host_psscr = mfspr(SPRN_PSSCR);
3384 unsigned long host_pidr = mfspr(SPRN_PID);
3386 hdec = time_limit - mftb();
3388 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3389 mtspr(SPRN_HDEC, hdec);
3391 if (vc->tb_offset) {
3392 u64 new_tb = mftb() + vc->tb_offset;
3393 mtspr(SPRN_TBU40, new_tb);
3395 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3396 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3397 vc->tb_offset_applied = vc->tb_offset;
3401 mtspr(SPRN_PCR, vc->pcr);
3402 mtspr(SPRN_DPDES, vc->dpdes);
3403 mtspr(SPRN_VTB, vc->vtb);
3405 local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3406 local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3407 mtspr(SPRN_PURR, vcpu->arch.purr);
3408 mtspr(SPRN_SPURR, vcpu->arch.spurr);
3410 if (dawr_enabled()) {
3411 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3412 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3414 mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3415 mtspr(SPRN_IC, vcpu->arch.ic);
3416 mtspr(SPRN_PID, vcpu->arch.pid);
3418 mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3419 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3421 mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3423 mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3424 mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3425 mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3426 mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3428 mtspr(SPRN_AMOR, ~0UL);
3430 mtspr(SPRN_LPCR, lpcr);
3433 kvmppc_xive_push_vcpu(vcpu);
3435 mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3436 mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3438 trap = __kvmhv_vcpu_entry_p9(vcpu);
3440 /* Advance host PURR/SPURR by the amount used by guest */
3441 purr = mfspr(SPRN_PURR);
3442 spurr = mfspr(SPRN_SPURR);
3443 mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3444 purr - vcpu->arch.purr);
3445 mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3446 spurr - vcpu->arch.spurr);
3447 vcpu->arch.purr = purr;
3448 vcpu->arch.spurr = spurr;
3450 vcpu->arch.ic = mfspr(SPRN_IC);
3451 vcpu->arch.pid = mfspr(SPRN_PID);
3452 vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3454 vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3455 vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3456 vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3457 vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3459 /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3460 mtspr(SPRN_PSSCR, host_psscr |
3461 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3462 mtspr(SPRN_HFSCR, host_hfscr);
3463 mtspr(SPRN_CIABR, host_ciabr);
3464 mtspr(SPRN_DAWR, host_dawr);
3465 mtspr(SPRN_DAWRX, host_dawrx);
3466 mtspr(SPRN_PID, host_pidr);
3469 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3470 * case we interrupted the guest between a tlbie and a ptesync.
3472 asm volatile("eieio; tlbsync; ptesync");
3474 mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid); /* restore host LPID */
3477 vc->dpdes = mfspr(SPRN_DPDES);
3478 vc->vtb = mfspr(SPRN_VTB);
3479 mtspr(SPRN_DPDES, 0);
3483 if (vc->tb_offset_applied) {
3484 u64 new_tb = mftb() - vc->tb_offset_applied;
3485 mtspr(SPRN_TBU40, new_tb);
3487 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3488 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3489 vc->tb_offset_applied = 0;
3492 mtspr(SPRN_HDEC, 0x7fffffff);
3493 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3499 * Virtual-mode guest entry for POWER9 and later when the host and
3500 * guest are both using the radix MMU. The LPIDR has already been set.
3502 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3505 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3506 unsigned long host_dscr = mfspr(SPRN_DSCR);
3507 unsigned long host_tidr = mfspr(SPRN_TIDR);
3508 unsigned long host_iamr = mfspr(SPRN_IAMR);
3509 unsigned long host_amr = mfspr(SPRN_AMR);
3514 dec = mfspr(SPRN_DEC);
3517 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3518 local_paca->kvm_hstate.dec_expires = dec + tb;
3519 if (local_paca->kvm_hstate.dec_expires < time_limit)
3520 time_limit = local_paca->kvm_hstate.dec_expires;
3522 vcpu->arch.ceded = 0;
3524 kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */
3526 kvmppc_subcore_enter_guest();
3528 vc->entry_exit_map = 1;
3531 if (vcpu->arch.vpa.pinned_addr) {
3532 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3533 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3534 lp->yield_count = cpu_to_be32(yield_count);
3535 vcpu->arch.vpa.dirty = 1;
3538 if (cpu_has_feature(CPU_FTR_TM) ||
3539 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3540 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3542 kvmhv_load_guest_pmu(vcpu);
3544 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3545 load_fp_state(&vcpu->arch.fp);
3546 #ifdef CONFIG_ALTIVEC
3547 load_vr_state(&vcpu->arch.vr);
3549 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3551 mtspr(SPRN_DSCR, vcpu->arch.dscr);
3552 mtspr(SPRN_IAMR, vcpu->arch.iamr);
3553 mtspr(SPRN_PSPB, vcpu->arch.pspb);
3554 mtspr(SPRN_FSCR, vcpu->arch.fscr);
3555 mtspr(SPRN_TAR, vcpu->arch.tar);
3556 mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3557 mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3558 mtspr(SPRN_BESCR, vcpu->arch.bescr);
3559 mtspr(SPRN_WORT, vcpu->arch.wort);
3560 mtspr(SPRN_TIDR, vcpu->arch.tid);
3561 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3562 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3563 mtspr(SPRN_AMR, vcpu->arch.amr);
3564 mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3566 if (!(vcpu->arch.ctrl & 1))
3567 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3569 mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3571 if (kvmhv_on_pseries()) {
3573 * We need to save and restore the guest visible part of the
3574 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3575 * doesn't do this for us. Note only required if pseries since
3576 * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3578 unsigned long host_psscr;
3579 /* call our hypervisor to load up HV regs and go */
3580 struct hv_guest_state hvregs;
3582 host_psscr = mfspr(SPRN_PSSCR_PR);
3583 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3584 kvmhv_save_hv_regs(vcpu, &hvregs);
3586 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3587 hvregs.version = HV_GUEST_STATE_VERSION;
3588 if (vcpu->arch.nested) {
3589 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3590 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3592 hvregs.lpid = vcpu->kvm->arch.lpid;
3593 hvregs.vcpu_token = vcpu->vcpu_id;
3595 hvregs.hdec_expiry = time_limit;
3596 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3597 __pa(&vcpu->arch.regs));
3598 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3599 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3600 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3601 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3602 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3603 mtspr(SPRN_PSSCR_PR, host_psscr);
3605 /* H_CEDE has to be handled now, not later */
3606 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3607 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3608 kvmppc_nested_cede(vcpu);
3612 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3615 vcpu->arch.slb_max = 0;
3616 dec = mfspr(SPRN_DEC);
3617 if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3620 vcpu->arch.dec_expires = dec + tb;
3622 vcpu->arch.thread_cpu = -1;
3623 vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3625 vcpu->arch.iamr = mfspr(SPRN_IAMR);
3626 vcpu->arch.pspb = mfspr(SPRN_PSPB);
3627 vcpu->arch.fscr = mfspr(SPRN_FSCR);
3628 vcpu->arch.tar = mfspr(SPRN_TAR);
3629 vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3630 vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3631 vcpu->arch.bescr = mfspr(SPRN_BESCR);
3632 vcpu->arch.wort = mfspr(SPRN_WORT);
3633 vcpu->arch.tid = mfspr(SPRN_TIDR);
3634 vcpu->arch.amr = mfspr(SPRN_AMR);
3635 vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3636 vcpu->arch.dscr = mfspr(SPRN_DSCR);
3638 mtspr(SPRN_PSPB, 0);
3639 mtspr(SPRN_WORT, 0);
3640 mtspr(SPRN_UAMOR, 0);
3641 mtspr(SPRN_DSCR, host_dscr);
3642 mtspr(SPRN_TIDR, host_tidr);
3643 mtspr(SPRN_IAMR, host_iamr);
3644 mtspr(SPRN_PSPB, 0);
3646 if (host_amr != vcpu->arch.amr)
3647 mtspr(SPRN_AMR, host_amr);
3649 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3650 store_fp_state(&vcpu->arch.fp);
3651 #ifdef CONFIG_ALTIVEC
3652 store_vr_state(&vcpu->arch.vr);
3654 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3656 if (cpu_has_feature(CPU_FTR_TM) ||
3657 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3658 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3661 if (vcpu->arch.vpa.pinned_addr) {
3662 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3663 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3664 lp->yield_count = cpu_to_be32(yield_count);
3665 vcpu->arch.vpa.dirty = 1;
3666 save_pmu = lp->pmcregs_in_use;
3668 /* Must save pmu if this guest is capable of running nested guests */
3669 save_pmu |= nesting_enabled(vcpu->kvm);
3671 kvmhv_save_guest_pmu(vcpu, save_pmu);
3673 vc->entry_exit_map = 0x101;
3676 mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3677 mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3679 kvmhv_load_host_pmu();
3681 kvmppc_subcore_exit_guest();
3687 * Wait for some other vcpu thread to execute us, and
3688 * wake us up when we need to handle something in the host.
3690 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3691 struct kvm_vcpu *vcpu, int wait_state)
3695 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3696 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3697 spin_unlock(&vc->lock);
3699 spin_lock(&vc->lock);
3701 finish_wait(&vcpu->arch.cpu_run, &wait);
3704 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3706 if (!halt_poll_ns_grow)
3709 vc->halt_poll_ns *= halt_poll_ns_grow;
3710 if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3711 vc->halt_poll_ns = halt_poll_ns_grow_start;
3714 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3716 if (halt_poll_ns_shrink == 0)
3717 vc->halt_poll_ns = 0;
3719 vc->halt_poll_ns /= halt_poll_ns_shrink;
3722 #ifdef CONFIG_KVM_XICS
3723 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3725 if (!xics_on_xive())
3727 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3728 vcpu->arch.xive_saved_state.cppr;
3731 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3735 #endif /* CONFIG_KVM_XICS */
3737 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3739 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3740 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3747 * Check to see if any of the runnable vcpus on the vcore have pending
3748 * exceptions or are no longer ceded
3750 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3752 struct kvm_vcpu *vcpu;
3755 for_each_runnable_thread(i, vcpu, vc) {
3756 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3764 * All the vcpus in this vcore are idle, so wait for a decrementer
3765 * or external interrupt to one of the vcpus. vc->lock is held.
3767 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3769 ktime_t cur, start_poll, start_wait;
3772 DECLARE_SWAITQUEUE(wait);
3774 /* Poll for pending exceptions and ceded state */
3775 cur = start_poll = ktime_get();
3776 if (vc->halt_poll_ns) {
3777 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3778 ++vc->runner->stat.halt_attempted_poll;
3780 vc->vcore_state = VCORE_POLLING;
3781 spin_unlock(&vc->lock);
3784 if (kvmppc_vcore_check_block(vc)) {
3789 } while (single_task_running() && ktime_before(cur, stop));
3791 spin_lock(&vc->lock);
3792 vc->vcore_state = VCORE_INACTIVE;
3795 ++vc->runner->stat.halt_successful_poll;
3800 prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3802 if (kvmppc_vcore_check_block(vc)) {
3803 finish_swait(&vc->wq, &wait);
3805 /* If we polled, count this as a successful poll */
3806 if (vc->halt_poll_ns)
3807 ++vc->runner->stat.halt_successful_poll;
3811 start_wait = ktime_get();
3813 vc->vcore_state = VCORE_SLEEPING;
3814 trace_kvmppc_vcore_blocked(vc, 0);
3815 spin_unlock(&vc->lock);
3817 finish_swait(&vc->wq, &wait);
3818 spin_lock(&vc->lock);
3819 vc->vcore_state = VCORE_INACTIVE;
3820 trace_kvmppc_vcore_blocked(vc, 1);
3821 ++vc->runner->stat.halt_successful_wait;
3826 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3828 /* Attribute wait time */
3830 vc->runner->stat.halt_wait_ns +=
3831 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3832 /* Attribute failed poll time */
3833 if (vc->halt_poll_ns)
3834 vc->runner->stat.halt_poll_fail_ns +=
3835 ktime_to_ns(start_wait) -
3836 ktime_to_ns(start_poll);
3838 /* Attribute successful poll time */
3839 if (vc->halt_poll_ns)
3840 vc->runner->stat.halt_poll_success_ns +=
3842 ktime_to_ns(start_poll);
3845 /* Adjust poll time */
3847 if (block_ns <= vc->halt_poll_ns)
3849 /* We slept and blocked for longer than the max halt time */
3850 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3851 shrink_halt_poll_ns(vc);
3852 /* We slept and our poll time is too small */
3853 else if (vc->halt_poll_ns < halt_poll_ns &&
3854 block_ns < halt_poll_ns)
3855 grow_halt_poll_ns(vc);
3856 if (vc->halt_poll_ns > halt_poll_ns)
3857 vc->halt_poll_ns = halt_poll_ns;
3859 vc->halt_poll_ns = 0;
3861 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3865 * This never fails for a radix guest, as none of the operations it does
3866 * for a radix guest can fail or have a way to report failure.
3867 * kvmhv_run_single_vcpu() relies on this fact.
3869 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3872 struct kvm *kvm = vcpu->kvm;
3874 mutex_lock(&kvm->arch.mmu_setup_lock);
3875 if (!kvm->arch.mmu_ready) {
3876 if (!kvm_is_radix(kvm))
3877 r = kvmppc_hv_setup_htab_rma(vcpu);
3879 if (cpu_has_feature(CPU_FTR_ARCH_300))
3880 kvmppc_setup_partition_table(kvm);
3881 kvm->arch.mmu_ready = 1;
3884 mutex_unlock(&kvm->arch.mmu_setup_lock);
3888 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3891 struct kvmppc_vcore *vc;
3894 trace_kvmppc_run_vcpu_enter(vcpu);
3896 kvm_run->exit_reason = 0;
3897 vcpu->arch.ret = RESUME_GUEST;
3898 vcpu->arch.trap = 0;
3899 kvmppc_update_vpas(vcpu);
3902 * Synchronize with other threads in this virtual core
3904 vc = vcpu->arch.vcore;
3905 spin_lock(&vc->lock);
3906 vcpu->arch.ceded = 0;
3907 vcpu->arch.run_task = current;
3908 vcpu->arch.kvm_run = kvm_run;
3909 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3910 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3911 vcpu->arch.busy_preempt = TB_NIL;
3912 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3916 * This happens the first time this is called for a vcpu.
3917 * If the vcore is already running, we may be able to start
3918 * this thread straight away and have it join in.
3920 if (!signal_pending(current)) {
3921 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3922 vc->vcore_state == VCORE_RUNNING) &&
3923 !VCORE_IS_EXITING(vc)) {
3924 kvmppc_create_dtl_entry(vcpu, vc);
3925 kvmppc_start_thread(vcpu, vc);
3926 trace_kvm_guest_enter(vcpu);
3927 } else if (vc->vcore_state == VCORE_SLEEPING) {
3928 swake_up_one(&vc->wq);
3933 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3934 !signal_pending(current)) {
3935 /* See if the MMU is ready to go */
3936 if (!vcpu->kvm->arch.mmu_ready) {
3937 spin_unlock(&vc->lock);
3938 r = kvmhv_setup_mmu(vcpu);
3939 spin_lock(&vc->lock);
3941 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3942 kvm_run->fail_entry.
3943 hardware_entry_failure_reason = 0;
3949 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3950 kvmppc_vcore_end_preempt(vc);
3952 if (vc->vcore_state != VCORE_INACTIVE) {
3953 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3956 for_each_runnable_thread(i, v, vc) {
3957 kvmppc_core_prepare_to_enter(v);
3958 if (signal_pending(v->arch.run_task)) {
3959 kvmppc_remove_runnable(vc, v);
3960 v->stat.signal_exits++;
3961 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3962 v->arch.ret = -EINTR;
3963 wake_up(&v->arch.cpu_run);
3966 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3969 for_each_runnable_thread(i, v, vc) {
3970 if (!kvmppc_vcpu_woken(v))
3971 n_ceded += v->arch.ceded;
3976 if (n_ceded == vc->n_runnable) {
3977 kvmppc_vcore_blocked(vc);
3978 } else if (need_resched()) {
3979 kvmppc_vcore_preempt(vc);
3980 /* Let something else run */
3981 cond_resched_lock(&vc->lock);
3982 if (vc->vcore_state == VCORE_PREEMPT)
3983 kvmppc_vcore_end_preempt(vc);
3985 kvmppc_run_core(vc);
3990 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3991 (vc->vcore_state == VCORE_RUNNING ||
3992 vc->vcore_state == VCORE_EXITING ||
3993 vc->vcore_state == VCORE_PIGGYBACK))
3994 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3996 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3997 kvmppc_vcore_end_preempt(vc);
3999 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4000 kvmppc_remove_runnable(vc, vcpu);
4001 vcpu->stat.signal_exits++;
4002 kvm_run->exit_reason = KVM_EXIT_INTR;
4003 vcpu->arch.ret = -EINTR;
4006 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4007 /* Wake up some vcpu to run the core */
4009 v = next_runnable_thread(vc, &i);
4010 wake_up(&v->arch.cpu_run);
4013 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4014 spin_unlock(&vc->lock);
4015 return vcpu->arch.ret;
4018 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4019 struct kvm_vcpu *vcpu, u64 time_limit,
4024 struct kvmppc_vcore *vc;
4025 struct kvm *kvm = vcpu->kvm;
4026 struct kvm_nested_guest *nested = vcpu->arch.nested;
4028 trace_kvmppc_run_vcpu_enter(vcpu);
4030 kvm_run->exit_reason = 0;
4031 vcpu->arch.ret = RESUME_GUEST;
4032 vcpu->arch.trap = 0;
4034 vc = vcpu->arch.vcore;
4035 vcpu->arch.ceded = 0;
4036 vcpu->arch.run_task = current;
4037 vcpu->arch.kvm_run = kvm_run;
4038 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4039 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4040 vcpu->arch.busy_preempt = TB_NIL;
4041 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4042 vc->runnable_threads[0] = vcpu;
4046 /* See if the MMU is ready to go */
4047 if (!kvm->arch.mmu_ready)
4048 kvmhv_setup_mmu(vcpu);
4053 kvmppc_update_vpas(vcpu);
4055 init_vcore_to_run(vc);
4056 vc->preempt_tb = TB_NIL;
4059 pcpu = smp_processor_id();
4061 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4063 local_irq_disable();
4065 if (signal_pending(current))
4067 if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4071 kvmppc_core_prepare_to_enter(vcpu);
4072 if (vcpu->arch.doorbell_request) {
4075 vcpu->arch.doorbell_request = 0;
4077 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4078 &vcpu->arch.pending_exceptions))
4080 } else if (vcpu->arch.pending_exceptions ||
4081 vcpu->arch.doorbell_request ||
4082 xive_interrupt_pending(vcpu)) {
4083 vcpu->arch.ret = RESUME_HOST;
4087 kvmppc_clear_host_core(pcpu);
4089 local_paca->kvm_hstate.tid = 0;
4090 local_paca->kvm_hstate.napping = 0;
4091 local_paca->kvm_hstate.kvm_split_mode = NULL;
4092 kvmppc_start_thread(vcpu, vc);
4093 kvmppc_create_dtl_entry(vcpu, vc);
4094 trace_kvm_guest_enter(vcpu);
4096 vc->vcore_state = VCORE_RUNNING;
4097 trace_kvmppc_run_core(vc, 0);
4099 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4100 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4101 mtspr(SPRN_LPID, lpid);
4103 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4106 guest_enter_irqoff();
4108 srcu_idx = srcu_read_lock(&kvm->srcu);
4110 this_cpu_disable_ftrace();
4112 /* Tell lockdep that we're about to enable interrupts */
4113 trace_hardirqs_on();
4115 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4116 vcpu->arch.trap = trap;
4118 trace_hardirqs_off();
4120 this_cpu_enable_ftrace();
4122 srcu_read_unlock(&kvm->srcu, srcu_idx);
4124 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4125 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4129 set_irq_happened(trap);
4131 kvmppc_set_host_core(pcpu);
4136 cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4141 * cancel pending decrementer exception if DEC is now positive, or if
4142 * entering a nested guest in which case the decrementer is now owned
4143 * by L2 and the L1 decrementer is provided in hdec_expires
4145 if (kvmppc_core_pending_dec(vcpu) &&
4146 ((get_tb() < vcpu->arch.dec_expires) ||
4147 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4148 kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4149 kvmppc_core_dequeue_dec(vcpu);
4151 trace_kvm_guest_exit(vcpu);
4155 r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4157 r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4161 if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4162 !kvmppc_vcpu_woken(vcpu)) {
4163 kvmppc_set_timer(vcpu);
4164 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4165 if (signal_pending(current)) {
4166 vcpu->stat.signal_exits++;
4167 kvm_run->exit_reason = KVM_EXIT_INTR;
4168 vcpu->arch.ret = -EINTR;
4171 spin_lock(&vc->lock);
4172 kvmppc_vcore_blocked(vc);
4173 spin_unlock(&vc->lock);
4176 vcpu->arch.ceded = 0;
4178 vc->vcore_state = VCORE_INACTIVE;
4179 trace_kvmppc_run_core(vc, 1);
4182 kvmppc_remove_runnable(vc, vcpu);
4183 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4185 return vcpu->arch.ret;
4188 vcpu->stat.signal_exits++;
4189 kvm_run->exit_reason = KVM_EXIT_INTR;
4190 vcpu->arch.ret = -EINTR;
4197 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4201 unsigned long ebb_regs[3] = {}; /* shut up GCC */
4202 unsigned long user_tar = 0;
4203 unsigned int user_vrsave;
4206 if (!vcpu->arch.sane) {
4207 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4212 * Don't allow entry with a suspended transaction, because
4213 * the guest entry/exit code will lose it.
4214 * If the guest has TM enabled, save away their TM-related SPRs
4215 * (they will get restored by the TM unavailable interrupt).
4217 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4218 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4219 (current->thread.regs->msr & MSR_TM)) {
4220 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4221 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4222 run->fail_entry.hardware_entry_failure_reason = 0;
4225 /* Enable TM so we can read the TM SPRs */
4226 mtmsr(mfmsr() | MSR_TM);
4227 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4228 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4229 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4230 current->thread.regs->msr &= ~MSR_TM;
4235 * Force online to 1 for the sake of old userspace which doesn't
4238 if (!vcpu->arch.online) {
4239 atomic_inc(&vcpu->arch.vcore->online_count);
4240 vcpu->arch.online = 1;
4243 kvmppc_core_prepare_to_enter(vcpu);
4245 /* No need to go into the guest when all we'll do is come back out */
4246 if (signal_pending(current)) {
4247 run->exit_reason = KVM_EXIT_INTR;
4252 atomic_inc(&kvm->arch.vcpus_running);
4253 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4256 flush_all_to_thread(current);
4258 /* Save userspace EBB and other register values */
4259 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4260 ebb_regs[0] = mfspr(SPRN_EBBHR);
4261 ebb_regs[1] = mfspr(SPRN_EBBRR);
4262 ebb_regs[2] = mfspr(SPRN_BESCR);
4263 user_tar = mfspr(SPRN_TAR);
4265 user_vrsave = mfspr(SPRN_VRSAVE);
4267 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4268 vcpu->arch.pgdir = current->mm->pgd;
4269 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4273 * The early POWER9 chips that can't mix radix and HPT threads
4274 * on the same core also need the workaround for the problem
4275 * where the TLB would prefetch entries in the guest exit path
4276 * for radix guests using the guest PIDR value and LPID 0.
4277 * The workaround is in the old path (kvmppc_run_vcpu())
4278 * but not the new path (kvmhv_run_single_vcpu()).
4280 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4281 !no_mixing_hpt_and_radix)
4282 r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4283 vcpu->arch.vcore->lpcr);
4285 r = kvmppc_run_vcpu(run, vcpu);
4287 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4288 !(vcpu->arch.shregs.msr & MSR_PR)) {
4289 trace_kvm_hcall_enter(vcpu);
4290 r = kvmppc_pseries_do_hcall(vcpu);
4291 trace_kvm_hcall_exit(vcpu, r);
4292 kvmppc_core_prepare_to_enter(vcpu);
4293 } else if (r == RESUME_PAGE_FAULT) {
4294 srcu_idx = srcu_read_lock(&kvm->srcu);
4295 r = kvmppc_book3s_hv_page_fault(run, vcpu,
4296 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4297 srcu_read_unlock(&kvm->srcu, srcu_idx);
4298 } else if (r == RESUME_PASSTHROUGH) {
4299 if (WARN_ON(xics_on_xive()))
4302 r = kvmppc_xics_rm_complete(vcpu, 0);
4304 } while (is_kvmppc_resume_guest(r));
4306 /* Restore userspace EBB and other register values */
4307 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4308 mtspr(SPRN_EBBHR, ebb_regs[0]);
4309 mtspr(SPRN_EBBRR, ebb_regs[1]);
4310 mtspr(SPRN_BESCR, ebb_regs[2]);
4311 mtspr(SPRN_TAR, user_tar);
4312 mtspr(SPRN_FSCR, current->thread.fscr);
4314 mtspr(SPRN_VRSAVE, user_vrsave);
4316 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4317 atomic_dec(&kvm->arch.vcpus_running);
4321 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4322 int shift, int sllp)
4324 (*sps)->page_shift = shift;
4325 (*sps)->slb_enc = sllp;
4326 (*sps)->enc[0].page_shift = shift;
4327 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4329 * Add 16MB MPSS support (may get filtered out by userspace)
4332 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4334 (*sps)->enc[1].page_shift = 24;
4335 (*sps)->enc[1].pte_enc = penc;
4341 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4342 struct kvm_ppc_smmu_info *info)
4344 struct kvm_ppc_one_seg_page_size *sps;
4347 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4348 * POWER7 doesn't support keys for instruction accesses,
4349 * POWER8 and POWER9 do.
4351 info->data_keys = 32;
4352 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4354 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4355 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4356 info->slb_size = 32;
4358 /* We only support these sizes for now, and no muti-size segments */
4359 sps = &info->sps[0];
4360 kvmppc_add_seg_page_size(&sps, 12, 0);
4361 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4362 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4364 /* If running as a nested hypervisor, we don't support HPT guests */
4365 if (kvmhv_on_pseries())
4366 info->flags |= KVM_PPC_NO_HASH;
4372 * Get (and clear) the dirty memory log for a memory slot.
4374 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4375 struct kvm_dirty_log *log)
4377 struct kvm_memslots *slots;
4378 struct kvm_memory_slot *memslot;
4381 unsigned long *buf, *p;
4382 struct kvm_vcpu *vcpu;
4384 mutex_lock(&kvm->slots_lock);
4387 if (log->slot >= KVM_USER_MEM_SLOTS)
4390 slots = kvm_memslots(kvm);
4391 memslot = id_to_memslot(slots, log->slot);
4393 if (!memslot->dirty_bitmap)
4397 * Use second half of bitmap area because both HPT and radix
4398 * accumulate bits in the first half.
4400 n = kvm_dirty_bitmap_bytes(memslot);
4401 buf = memslot->dirty_bitmap + n / sizeof(long);
4404 if (kvm_is_radix(kvm))
4405 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4407 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4412 * We accumulate dirty bits in the first half of the
4413 * memslot's dirty_bitmap area, for when pages are paged
4414 * out or modified by the host directly. Pick up these
4415 * bits and add them to the map.
4417 p = memslot->dirty_bitmap;
4418 for (i = 0; i < n / sizeof(long); ++i)
4419 buf[i] |= xchg(&p[i], 0);
4421 /* Harvest dirty bits from VPA and DTL updates */
4422 /* Note: we never modify the SLB shadow buffer areas */
4423 kvm_for_each_vcpu(i, vcpu, kvm) {
4424 spin_lock(&vcpu->arch.vpa_update_lock);
4425 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4426 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4427 spin_unlock(&vcpu->arch.vpa_update_lock);
4431 if (copy_to_user(log->dirty_bitmap, buf, n))
4436 mutex_unlock(&kvm->slots_lock);
4440 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4441 struct kvm_memory_slot *dont)
4443 if (!dont || free->arch.rmap != dont->arch.rmap) {
4444 vfree(free->arch.rmap);
4445 free->arch.rmap = NULL;
4449 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4450 unsigned long npages)
4452 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4453 if (!slot->arch.rmap)
4459 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4460 struct kvm_memory_slot *memslot,
4461 const struct kvm_userspace_memory_region *mem)
4466 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4467 const struct kvm_userspace_memory_region *mem,
4468 const struct kvm_memory_slot *old,
4469 const struct kvm_memory_slot *new,
4470 enum kvm_mr_change change)
4472 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4475 * If we are making a new memslot, it might make
4476 * some address that was previously cached as emulated
4477 * MMIO be no longer emulated MMIO, so invalidate
4478 * all the caches of emulated MMIO translations.
4481 atomic64_inc(&kvm->arch.mmio_update);
4484 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4485 * have already called kvm_arch_flush_shadow_memslot() to
4486 * flush shadow mappings. For KVM_MR_CREATE we have no
4487 * previous mappings. So the only case to handle is
4488 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4490 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4491 * to get rid of any THP PTEs in the partition-scoped page tables
4492 * so we can track dirtiness at the page level; we flush when
4493 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4496 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4497 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4498 kvmppc_radix_flush_memslot(kvm, old);
4502 * Update LPCR values in kvm->arch and in vcores.
4503 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4504 * of kvm->arch.lpcr update).
4506 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4511 if ((kvm->arch.lpcr & mask) == lpcr)
4514 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4516 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4517 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4520 spin_lock(&vc->lock);
4521 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4522 spin_unlock(&vc->lock);
4523 if (++cores_done >= kvm->arch.online_vcores)
4528 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4533 void kvmppc_setup_partition_table(struct kvm *kvm)
4535 unsigned long dw0, dw1;
4537 if (!kvm_is_radix(kvm)) {
4538 /* PS field - page size for VRMA */
4539 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4540 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4541 /* HTABSIZE and HTABORG fields */
4542 dw0 |= kvm->arch.sdr1;
4544 /* Second dword as set by userspace */
4545 dw1 = kvm->arch.process_table;
4547 dw0 = PATB_HR | radix__get_tree_size() |
4548 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4549 dw1 = PATB_GR | kvm->arch.process_table;
4551 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4555 * Set up HPT (hashed page table) and RMA (real-mode area).
4556 * Must be called with kvm->arch.mmu_setup_lock held.
4558 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4561 struct kvm *kvm = vcpu->kvm;
4563 struct kvm_memory_slot *memslot;
4564 struct vm_area_struct *vma;
4565 unsigned long lpcr = 0, senc;
4566 unsigned long psize, porder;
4569 /* Allocate hashed page table (if not done already) and reset it */
4570 if (!kvm->arch.hpt.virt) {
4571 int order = KVM_DEFAULT_HPT_ORDER;
4572 struct kvm_hpt_info info;
4574 err = kvmppc_allocate_hpt(&info, order);
4575 /* If we get here, it means userspace didn't specify a
4576 * size explicitly. So, try successively smaller
4577 * sizes if the default failed. */
4578 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4579 err = kvmppc_allocate_hpt(&info, order);
4582 pr_err("KVM: Couldn't alloc HPT\n");
4586 kvmppc_set_hpt(kvm, &info);
4589 /* Look up the memslot for guest physical address 0 */
4590 srcu_idx = srcu_read_lock(&kvm->srcu);
4591 memslot = gfn_to_memslot(kvm, 0);
4593 /* We must have some memory at 0 by now */
4595 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4598 /* Look up the VMA for the start of this memory slot */
4599 hva = memslot->userspace_addr;
4600 down_read(¤t->mm->mmap_sem);
4601 vma = find_vma(current->mm, hva);
4602 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4605 psize = vma_kernel_pagesize(vma);
4607 up_read(¤t->mm->mmap_sem);
4609 /* We can handle 4k, 64k or 16M pages in the VRMA */
4610 if (psize >= 0x1000000)
4612 else if (psize >= 0x10000)
4616 porder = __ilog2(psize);
4618 senc = slb_pgsize_encoding(psize);
4619 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4620 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4621 /* Create HPTEs in the hash page table for the VRMA */
4622 kvmppc_map_vrma(vcpu, memslot, porder);
4624 /* Update VRMASD field in the LPCR */
4625 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4626 /* the -4 is to account for senc values starting at 0x10 */
4627 lpcr = senc << (LPCR_VRMASD_SH - 4);
4628 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4631 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4635 srcu_read_unlock(&kvm->srcu, srcu_idx);
4640 up_read(¤t->mm->mmap_sem);
4645 * Must be called with kvm->arch.mmu_setup_lock held and
4646 * mmu_ready = 0 and no vcpus running.
4648 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4650 if (nesting_enabled(kvm))
4651 kvmhv_release_all_nested(kvm);
4652 kvmppc_rmap_reset(kvm);
4653 kvm->arch.process_table = 0;
4654 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4655 spin_lock(&kvm->mmu_lock);
4656 kvm->arch.radix = 0;
4657 spin_unlock(&kvm->mmu_lock);
4658 kvmppc_free_radix(kvm);
4659 kvmppc_update_lpcr(kvm, LPCR_VPM1,
4660 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4665 * Must be called with kvm->arch.mmu_setup_lock held and
4666 * mmu_ready = 0 and no vcpus running.
4668 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4672 err = kvmppc_init_vm_radix(kvm);
4675 kvmppc_rmap_reset(kvm);
4676 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4677 spin_lock(&kvm->mmu_lock);
4678 kvm->arch.radix = 1;
4679 spin_unlock(&kvm->mmu_lock);
4680 kvmppc_free_hpt(&kvm->arch.hpt);
4681 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4682 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4686 #ifdef CONFIG_KVM_XICS
4688 * Allocate a per-core structure for managing state about which cores are
4689 * running in the host versus the guest and for exchanging data between
4690 * real mode KVM and CPU running in the host.
4691 * This is only done for the first VM.
4692 * The allocated structure stays even if all VMs have stopped.
4693 * It is only freed when the kvm-hv module is unloaded.
4694 * It's OK for this routine to fail, we just don't support host
4695 * core operations like redirecting H_IPI wakeups.
4697 void kvmppc_alloc_host_rm_ops(void)
4699 struct kvmppc_host_rm_ops *ops;
4700 unsigned long l_ops;
4704 /* Not the first time here ? */
4705 if (kvmppc_host_rm_ops_hv != NULL)
4708 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4712 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4713 ops->rm_core = kzalloc(size, GFP_KERNEL);
4715 if (!ops->rm_core) {
4722 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4723 if (!cpu_online(cpu))
4726 core = cpu >> threads_shift;
4727 ops->rm_core[core].rm_state.in_host = 1;
4730 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4733 * Make the contents of the kvmppc_host_rm_ops structure visible
4734 * to other CPUs before we assign it to the global variable.
4735 * Do an atomic assignment (no locks used here), but if someone
4736 * beats us to it, just free our copy and return.
4739 l_ops = (unsigned long) ops;
4741 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4743 kfree(ops->rm_core);
4748 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4749 "ppc/kvm_book3s:prepare",
4750 kvmppc_set_host_core,
4751 kvmppc_clear_host_core);
4755 void kvmppc_free_host_rm_ops(void)
4757 if (kvmppc_host_rm_ops_hv) {
4758 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4759 kfree(kvmppc_host_rm_ops_hv->rm_core);
4760 kfree(kvmppc_host_rm_ops_hv);
4761 kvmppc_host_rm_ops_hv = NULL;
4766 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4768 unsigned long lpcr, lpid;
4772 mutex_init(&kvm->arch.mmu_setup_lock);
4774 /* Allocate the guest's logical partition ID */
4776 lpid = kvmppc_alloc_lpid();
4779 kvm->arch.lpid = lpid;
4781 kvmppc_alloc_host_rm_ops();
4783 kvmhv_vm_nested_init(kvm);
4786 * Since we don't flush the TLB when tearing down a VM,
4787 * and this lpid might have previously been used,
4788 * make sure we flush on each core before running the new VM.
4789 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4790 * does this flush for us.
4792 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4793 cpumask_setall(&kvm->arch.need_tlb_flush);
4795 /* Start out with the default set of hcalls enabled */
4796 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4797 sizeof(kvm->arch.enabled_hcalls));
4799 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4800 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4802 /* Init LPCR for virtual RMA mode */
4803 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4804 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4805 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4806 lpcr &= LPCR_PECE | LPCR_LPES;
4810 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4811 LPCR_VPM0 | LPCR_VPM1;
4812 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4813 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4814 /* On POWER8 turn on online bit to enable PURR/SPURR */
4815 if (cpu_has_feature(CPU_FTR_ARCH_207S))
4818 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4819 * Set HVICE bit to enable hypervisor virtualization interrupts.
4820 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4821 * be unnecessary but better safe than sorry in case we re-enable
4822 * EE in HV mode with this LPCR still set)
4824 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4826 lpcr |= LPCR_HVICE | LPCR_HEIC;
4829 * If xive is enabled, we route 0x500 interrupts directly
4837 * If the host uses radix, the guest starts out as radix.
4839 if (radix_enabled()) {
4840 kvm->arch.radix = 1;
4841 kvm->arch.mmu_ready = 1;
4843 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4844 ret = kvmppc_init_vm_radix(kvm);
4846 kvmppc_free_lpid(kvm->arch.lpid);
4849 kvmppc_setup_partition_table(kvm);
4852 kvm->arch.lpcr = lpcr;
4854 /* Initialization for future HPT resizes */
4855 kvm->arch.resize_hpt = NULL;
4858 * Work out how many sets the TLB has, for the use of
4859 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4861 if (radix_enabled())
4862 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
4863 else if (cpu_has_feature(CPU_FTR_ARCH_300))
4864 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
4865 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4866 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
4868 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
4871 * Track that we now have a HV mode VM active. This blocks secondary
4872 * CPU threads from coming online.
4873 * On POWER9, we only need to do this if the "indep_threads_mode"
4874 * module parameter has been set to N.
4876 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4877 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4878 pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4879 kvm->arch.threads_indep = true;
4881 kvm->arch.threads_indep = indep_threads_mode;
4884 if (!kvm->arch.threads_indep)
4885 kvm_hv_vm_activated();
4888 * Initialize smt_mode depending on processor.
4889 * POWER8 and earlier have to use "strict" threading, where
4890 * all vCPUs in a vcore have to run on the same (sub)core,
4891 * whereas on POWER9 the threads can each run a different
4894 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4895 kvm->arch.smt_mode = threads_per_subcore;
4897 kvm->arch.smt_mode = 1;
4898 kvm->arch.emul_smt_mode = 1;
4901 * Create a debugfs directory for the VM
4903 snprintf(buf, sizeof(buf), "vm%d", current->pid);
4904 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4905 kvmppc_mmu_debugfs_init(kvm);
4906 if (radix_enabled())
4907 kvmhv_radix_debugfs_init(kvm);
4912 static void kvmppc_free_vcores(struct kvm *kvm)
4916 for (i = 0; i < KVM_MAX_VCORES; ++i)
4917 kfree(kvm->arch.vcores[i]);
4918 kvm->arch.online_vcores = 0;
4921 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4923 debugfs_remove_recursive(kvm->arch.debugfs_dir);
4925 if (!kvm->arch.threads_indep)
4926 kvm_hv_vm_deactivated();
4928 kvmppc_free_vcores(kvm);
4931 if (kvm_is_radix(kvm))
4932 kvmppc_free_radix(kvm);
4934 kvmppc_free_hpt(&kvm->arch.hpt);
4936 /* Perform global invalidation and return lpid to the pool */
4937 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4938 if (nesting_enabled(kvm))
4939 kvmhv_release_all_nested(kvm);
4940 kvm->arch.process_table = 0;
4941 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4943 kvmppc_free_lpid(kvm->arch.lpid);
4945 kvmppc_free_pimap(kvm);
4948 /* We don't need to emulate any privileged instructions or dcbz */
4949 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4950 unsigned int inst, int *advance)
4952 return EMULATE_FAIL;
4955 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4958 return EMULATE_FAIL;
4961 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4964 return EMULATE_FAIL;
4967 static int kvmppc_core_check_processor_compat_hv(void)
4969 if (cpu_has_feature(CPU_FTR_HVMODE) &&
4970 cpu_has_feature(CPU_FTR_ARCH_206))
4973 /* POWER9 in radix mode is capable of being a nested hypervisor. */
4974 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4980 #ifdef CONFIG_KVM_XICS
4982 void kvmppc_free_pimap(struct kvm *kvm)
4984 kfree(kvm->arch.pimap);
4987 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4989 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4992 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4994 struct irq_desc *desc;
4995 struct kvmppc_irq_map *irq_map;
4996 struct kvmppc_passthru_irqmap *pimap;
4997 struct irq_chip *chip;
5000 if (!kvm_irq_bypass)
5003 desc = irq_to_desc(host_irq);
5007 mutex_lock(&kvm->lock);
5009 pimap = kvm->arch.pimap;
5010 if (pimap == NULL) {
5011 /* First call, allocate structure to hold IRQ map */
5012 pimap = kvmppc_alloc_pimap();
5013 if (pimap == NULL) {
5014 mutex_unlock(&kvm->lock);
5017 kvm->arch.pimap = pimap;
5021 * For now, we only support interrupts for which the EOI operation
5022 * is an OPAL call followed by a write to XIRR, since that's
5023 * what our real-mode EOI code does, or a XIVE interrupt
5025 chip = irq_data_get_irq_chip(&desc->irq_data);
5026 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5027 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5028 host_irq, guest_gsi);
5029 mutex_unlock(&kvm->lock);
5034 * See if we already have an entry for this guest IRQ number.
5035 * If it's mapped to a hardware IRQ number, that's an error,
5036 * otherwise re-use this entry.
5038 for (i = 0; i < pimap->n_mapped; i++) {
5039 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5040 if (pimap->mapped[i].r_hwirq) {
5041 mutex_unlock(&kvm->lock);
5048 if (i == KVMPPC_PIRQ_MAPPED) {
5049 mutex_unlock(&kvm->lock);
5050 return -EAGAIN; /* table is full */
5053 irq_map = &pimap->mapped[i];
5055 irq_map->v_hwirq = guest_gsi;
5056 irq_map->desc = desc;
5059 * Order the above two stores before the next to serialize with
5060 * the KVM real mode handler.
5063 irq_map->r_hwirq = desc->irq_data.hwirq;
5065 if (i == pimap->n_mapped)
5069 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5071 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5073 irq_map->r_hwirq = 0;
5075 mutex_unlock(&kvm->lock);
5080 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5082 struct irq_desc *desc;
5083 struct kvmppc_passthru_irqmap *pimap;
5086 if (!kvm_irq_bypass)
5089 desc = irq_to_desc(host_irq);
5093 mutex_lock(&kvm->lock);
5094 if (!kvm->arch.pimap)
5097 pimap = kvm->arch.pimap;
5099 for (i = 0; i < pimap->n_mapped; i++) {
5100 if (guest_gsi == pimap->mapped[i].v_hwirq)
5104 if (i == pimap->n_mapped) {
5105 mutex_unlock(&kvm->lock);
5110 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5112 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5114 /* invalidate the entry (what do do on error from the above ?) */
5115 pimap->mapped[i].r_hwirq = 0;
5118 * We don't free this structure even when the count goes to
5119 * zero. The structure is freed when we destroy the VM.
5122 mutex_unlock(&kvm->lock);
5126 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5127 struct irq_bypass_producer *prod)
5130 struct kvm_kernel_irqfd *irqfd =
5131 container_of(cons, struct kvm_kernel_irqfd, consumer);
5133 irqfd->producer = prod;
5135 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5137 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5138 prod->irq, irqfd->gsi, ret);
5143 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5144 struct irq_bypass_producer *prod)
5147 struct kvm_kernel_irqfd *irqfd =
5148 container_of(cons, struct kvm_kernel_irqfd, consumer);
5150 irqfd->producer = NULL;
5153 * When producer of consumer is unregistered, we change back to
5154 * default external interrupt handling mode - KVM real mode
5155 * will switch back to host.
5157 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5159 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5160 prod->irq, irqfd->gsi, ret);
5164 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5165 unsigned int ioctl, unsigned long arg)
5167 struct kvm *kvm __maybe_unused = filp->private_data;
5168 void __user *argp = (void __user *)arg;
5173 case KVM_PPC_ALLOCATE_HTAB: {
5177 if (get_user(htab_order, (u32 __user *)argp))
5179 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5186 case KVM_PPC_GET_HTAB_FD: {
5187 struct kvm_get_htab_fd ghf;
5190 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5192 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5196 case KVM_PPC_RESIZE_HPT_PREPARE: {
5197 struct kvm_ppc_resize_hpt rhpt;
5200 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5203 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5207 case KVM_PPC_RESIZE_HPT_COMMIT: {
5208 struct kvm_ppc_resize_hpt rhpt;
5211 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5214 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5226 * List of hcall numbers to enable by default.
5227 * For compatibility with old userspace, we enable by default
5228 * all hcalls that were implemented before the hcall-enabling
5229 * facility was added. Note this list should not include H_RTAS.
5231 static unsigned int default_hcall_list[] = {
5245 #ifdef CONFIG_KVM_XICS
5256 static void init_default_hcalls(void)
5261 for (i = 0; default_hcall_list[i]; ++i) {
5262 hcall = default_hcall_list[i];
5263 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5264 __set_bit(hcall / 4, default_enabled_hcalls);
5268 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5274 /* If not on a POWER9, reject it */
5275 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5278 /* If any unknown flags set, reject it */
5279 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5282 /* GR (guest radix) bit in process_table field must match */
5283 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5284 if (!!(cfg->process_table & PATB_GR) != radix)
5287 /* Process table size field must be reasonable, i.e. <= 24 */
5288 if ((cfg->process_table & PRTS_MASK) > 24)
5291 /* We can change a guest to/from radix now, if the host is radix */
5292 if (radix && !radix_enabled())
5295 /* If we're a nested hypervisor, we currently only support radix */
5296 if (kvmhv_on_pseries() && !radix)
5299 mutex_lock(&kvm->arch.mmu_setup_lock);
5300 if (radix != kvm_is_radix(kvm)) {
5301 if (kvm->arch.mmu_ready) {
5302 kvm->arch.mmu_ready = 0;
5303 /* order mmu_ready vs. vcpus_running */
5305 if (atomic_read(&kvm->arch.vcpus_running)) {
5306 kvm->arch.mmu_ready = 1;
5312 err = kvmppc_switch_mmu_to_radix(kvm);
5314 err = kvmppc_switch_mmu_to_hpt(kvm);
5319 kvm->arch.process_table = cfg->process_table;
5320 kvmppc_setup_partition_table(kvm);
5322 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5323 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5327 mutex_unlock(&kvm->arch.mmu_setup_lock);
5331 static int kvmhv_enable_nested(struct kvm *kvm)
5335 if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5338 /* kvm == NULL means the caller is testing if the capability exists */
5340 kvm->arch.nested_enable = true;
5344 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5349 if (kvmhv_vcpu_is_radix(vcpu)) {
5350 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5356 /* For now quadrants are the only way to access nested guest memory */
5357 if (rc && vcpu->arch.nested)
5363 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5368 if (kvmhv_vcpu_is_radix(vcpu)) {
5369 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5375 /* For now quadrants are the only way to access nested guest memory */
5376 if (rc && vcpu->arch.nested)
5382 static struct kvmppc_ops kvm_ops_hv = {
5383 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5384 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5385 .get_one_reg = kvmppc_get_one_reg_hv,
5386 .set_one_reg = kvmppc_set_one_reg_hv,
5387 .vcpu_load = kvmppc_core_vcpu_load_hv,
5388 .vcpu_put = kvmppc_core_vcpu_put_hv,
5389 .set_msr = kvmppc_set_msr_hv,
5390 .vcpu_run = kvmppc_vcpu_run_hv,
5391 .vcpu_create = kvmppc_core_vcpu_create_hv,
5392 .vcpu_free = kvmppc_core_vcpu_free_hv,
5393 .check_requests = kvmppc_core_check_requests_hv,
5394 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
5395 .flush_memslot = kvmppc_core_flush_memslot_hv,
5396 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5397 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
5398 .unmap_hva_range = kvm_unmap_hva_range_hv,
5399 .age_hva = kvm_age_hva_hv,
5400 .test_age_hva = kvm_test_age_hva_hv,
5401 .set_spte_hva = kvm_set_spte_hva_hv,
5402 .mmu_destroy = kvmppc_mmu_destroy_hv,
5403 .free_memslot = kvmppc_core_free_memslot_hv,
5404 .create_memslot = kvmppc_core_create_memslot_hv,
5405 .init_vm = kvmppc_core_init_vm_hv,
5406 .destroy_vm = kvmppc_core_destroy_vm_hv,
5407 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5408 .emulate_op = kvmppc_core_emulate_op_hv,
5409 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5410 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5411 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5412 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
5413 .hcall_implemented = kvmppc_hcall_impl_hv,
5414 #ifdef CONFIG_KVM_XICS
5415 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5416 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5418 .configure_mmu = kvmhv_configure_mmu,
5419 .get_rmmu_info = kvmhv_get_rmmu_info,
5420 .set_smt_mode = kvmhv_set_smt_mode,
5421 .enable_nested = kvmhv_enable_nested,
5422 .load_from_eaddr = kvmhv_load_from_eaddr,
5423 .store_to_eaddr = kvmhv_store_to_eaddr,
5426 static int kvm_init_subcore_bitmap(void)
5429 int nr_cores = cpu_nr_cores();
5430 struct sibling_subcore_state *sibling_subcore_state;
5432 for (i = 0; i < nr_cores; i++) {
5433 int first_cpu = i * threads_per_core;
5434 int node = cpu_to_node(first_cpu);
5436 /* Ignore if it is already allocated. */
5437 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5440 sibling_subcore_state =
5441 kzalloc_node(sizeof(struct sibling_subcore_state),
5443 if (!sibling_subcore_state)
5447 for (j = 0; j < threads_per_core; j++) {
5448 int cpu = first_cpu + j;
5450 paca_ptrs[cpu]->sibling_subcore_state =
5451 sibling_subcore_state;
5457 static int kvmppc_radix_possible(void)
5459 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5462 static int kvmppc_book3s_init_hv(void)
5466 * FIXME!! Do we need to check on all cpus ?
5468 r = kvmppc_core_check_processor_compat_hv();
5472 r = kvmhv_nested_init();
5476 r = kvm_init_subcore_bitmap();
5481 * We need a way of accessing the XICS interrupt controller,
5482 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5483 * indirectly, via OPAL.
5486 if (!xics_on_xive() && !kvmhv_on_pseries() &&
5487 !local_paca->kvm_hstate.xics_phys) {
5488 struct device_node *np;
5490 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5492 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5495 /* presence of intc confirmed - node can be dropped again */
5500 kvm_ops_hv.owner = THIS_MODULE;
5501 kvmppc_hv_ops = &kvm_ops_hv;
5503 init_default_hcalls();
5507 r = kvmppc_mmu_hv_init();
5511 if (kvmppc_radix_possible())
5512 r = kvmppc_radix_init();
5515 * POWER9 chips before version 2.02 can't have some threads in
5516 * HPT mode and some in radix mode on the same core.
5518 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5519 unsigned int pvr = mfspr(SPRN_PVR);
5520 if ((pvr >> 16) == PVR_POWER9 &&
5521 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5522 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5523 no_mixing_hpt_and_radix = true;
5529 static void kvmppc_book3s_exit_hv(void)
5531 kvmppc_free_host_rm_ops();
5532 if (kvmppc_radix_possible())
5533 kvmppc_radix_exit();
5534 kvmppc_hv_ops = NULL;
5535 kvmhv_nested_exit();
5538 module_init(kvmppc_book3s_init_hv);
5539 module_exit(kvmppc_book3s_exit_hv);
5540 MODULE_LICENSE("GPL");
5541 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5542 MODULE_ALIAS("devname:kvm");