2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
21 #include <linux/kvm_host.h>
22 #include <linux/kernel.h>
23 #include <linux/err.h>
24 #include <linux/slab.h>
25 #include <linux/preempt.h>
26 #include <linux/sched/signal.h>
27 #include <linux/sched/stat.h>
28 #include <linux/delay.h>
29 #include <linux/export.h>
31 #include <linux/anon_inodes.h>
32 #include <linux/cpu.h>
33 #include <linux/cpumask.h>
34 #include <linux/spinlock.h>
35 #include <linux/page-flags.h>
36 #include <linux/srcu.h>
37 #include <linux/miscdevice.h>
38 #include <linux/debugfs.h>
39 #include <linux/gfp.h>
40 #include <linux/vmalloc.h>
41 #include <linux/highmem.h>
42 #include <linux/hugetlb.h>
43 #include <linux/kvm_irqfd.h>
44 #include <linux/irqbypass.h>
45 #include <linux/module.h>
46 #include <linux/compiler.h>
49 #include <asm/ftrace.h>
51 #include <asm/ppc-opcode.h>
52 #include <asm/asm-prototypes.h>
53 #include <asm/archrandom.h>
54 #include <asm/debug.h>
55 #include <asm/disassemble.h>
56 #include <asm/cputable.h>
57 #include <asm/cacheflush.h>
58 #include <linux/uaccess.h>
60 #include <asm/kvm_ppc.h>
61 #include <asm/kvm_book3s.h>
62 #include <asm/mmu_context.h>
63 #include <asm/lppaca.h>
64 #include <asm/processor.h>
65 #include <asm/cputhreads.h>
67 #include <asm/hvcall.h>
68 #include <asm/switch_to.h>
70 #include <asm/dbell.h>
72 #include <asm/pnv-pci.h>
77 #include <asm/hw_breakpoint.h>
81 #define CREATE_TRACE_POINTS
84 /* #define EXIT_DEBUG */
85 /* #define EXIT_DEBUG_SIMPLE */
86 /* #define EXIT_DEBUG_INT */
88 /* Used to indicate that a guest page fault needs to be handled */
89 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
90 /* Used to indicate that a guest passthrough interrupt needs to be handled */
91 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
93 /* Used as a "null" value for timebase values */
94 #define TB_NIL (~(u64)0)
96 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
98 static int dynamic_mt_modes = 6;
99 module_param(dynamic_mt_modes, int, 0644);
100 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
101 static int target_smt_mode;
102 module_param(target_smt_mode, int, 0644);
103 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
105 static bool indep_threads_mode = true;
106 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
107 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
109 static bool one_vm_per_core;
110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
113 #ifdef CONFIG_KVM_XICS
114 static struct kernel_param_ops module_param_ops = {
115 .set = param_set_int,
116 .get = param_get_int,
119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
126 /* If set, guests are allowed to create and control nested guests */
127 static bool nested = true;
128 module_param(nested, bool, S_IRUGO | S_IWUSR);
129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
131 static inline bool nesting_enabled(struct kvm *kvm)
133 return kvm->arch.nested_enable && kvm_is_radix(kvm);
136 /* If set, the threads on each CPU core have to be in the same MMU mode */
137 static bool no_mixing_hpt_and_radix;
139 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
140 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
143 * RWMR values for POWER8. These control the rate at which PURR
144 * and SPURR count and should be set according to the number of
145 * online threads in the vcore being run.
147 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
148 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
149 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
150 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
151 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
152 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
153 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
154 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
156 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
168 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
172 struct kvm_vcpu *vcpu;
174 while (++i < MAX_SMT_THREADS) {
175 vcpu = READ_ONCE(vc->runnable_threads[i]);
184 /* Used to traverse the list of runnable threads for a given vcore */
185 #define for_each_runnable_thread(i, vcpu, vc) \
186 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
188 static bool kvmppc_ipi_thread(int cpu)
190 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
192 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
193 if (kvmhv_on_pseries())
196 /* On POWER9 we can use msgsnd to IPI any cpu */
197 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
198 msg |= get_hard_smp_processor_id(cpu);
200 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
204 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
205 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
207 if (cpu_first_thread_sibling(cpu) ==
208 cpu_first_thread_sibling(smp_processor_id())) {
209 msg |= cpu_thread_in_core(cpu);
211 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
218 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
219 if (cpu >= 0 && cpu < nr_cpu_ids) {
220 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
224 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
232 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
235 struct swait_queue_head *wqp;
237 wqp = kvm_arch_vcpu_wq(vcpu);
238 if (swq_has_sleeper(wqp)) {
240 ++vcpu->stat.halt_wakeup;
243 cpu = READ_ONCE(vcpu->arch.thread_cpu);
244 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
247 /* CPU points to the first thread of the core */
249 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
250 smp_send_reschedule(cpu);
254 * We use the vcpu_load/put functions to measure stolen time.
255 * Stolen time is counted as time when either the vcpu is able to
256 * run as part of a virtual core, but the task running the vcore
257 * is preempted or sleeping, or when the vcpu needs something done
258 * in the kernel by the task running the vcpu, but that task is
259 * preempted or sleeping. Those two things have to be counted
260 * separately, since one of the vcpu tasks will take on the job
261 * of running the core, and the other vcpu tasks in the vcore will
262 * sleep waiting for it to do that, but that sleep shouldn't count
265 * Hence we accumulate stolen time when the vcpu can run as part of
266 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
267 * needs its task to do other things in the kernel (for example,
268 * service a page fault) in busy_stolen. We don't accumulate
269 * stolen time for a vcore when it is inactive, or for a vcpu
270 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
271 * a misnomer; it means that the vcpu task is not executing in
272 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
273 * the kernel. We don't have any way of dividing up that time
274 * between time that the vcpu is genuinely stopped, time that
275 * the task is actively working on behalf of the vcpu, and time
276 * that the task is preempted, so we don't count any of it as
279 * Updates to busy_stolen are protected by arch.tbacct_lock;
280 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
281 * lock. The stolen times are measured in units of timebase ticks.
282 * (Note that the != TB_NIL checks below are purely defensive;
283 * they should never fail.)
286 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
290 spin_lock_irqsave(&vc->stoltb_lock, flags);
291 vc->preempt_tb = mftb();
292 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
295 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
299 spin_lock_irqsave(&vc->stoltb_lock, flags);
300 if (vc->preempt_tb != TB_NIL) {
301 vc->stolen_tb += mftb() - vc->preempt_tb;
302 vc->preempt_tb = TB_NIL;
304 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
307 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
309 struct kvmppc_vcore *vc = vcpu->arch.vcore;
313 * We can test vc->runner without taking the vcore lock,
314 * because only this task ever sets vc->runner to this
315 * vcpu, and once it is set to this vcpu, only this task
316 * ever sets it to NULL.
318 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
319 kvmppc_core_end_stolen(vc);
321 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
322 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
323 vcpu->arch.busy_preempt != TB_NIL) {
324 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
325 vcpu->arch.busy_preempt = TB_NIL;
327 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
330 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
332 struct kvmppc_vcore *vc = vcpu->arch.vcore;
335 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
336 kvmppc_core_start_stolen(vc);
338 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
339 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
340 vcpu->arch.busy_preempt = mftb();
341 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
344 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
347 * Check for illegal transactional state bit combination
348 * and if we find it, force the TS field to a safe state.
350 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
352 vcpu->arch.shregs.msr = msr;
353 kvmppc_end_cede(vcpu);
356 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
358 vcpu->arch.pvr = pvr;
361 /* Dummy value used in computing PCR value below */
362 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
364 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
366 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
367 struct kvmppc_vcore *vc = vcpu->arch.vcore;
369 /* We can (emulate) our own architecture version and anything older */
370 if (cpu_has_feature(CPU_FTR_ARCH_300))
371 host_pcr_bit = PCR_ARCH_300;
372 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
373 host_pcr_bit = PCR_ARCH_207;
374 else if (cpu_has_feature(CPU_FTR_ARCH_206))
375 host_pcr_bit = PCR_ARCH_206;
377 host_pcr_bit = PCR_ARCH_205;
379 /* Determine lowest PCR bit needed to run guest in given PVR level */
380 guest_pcr_bit = host_pcr_bit;
382 switch (arch_compat) {
384 guest_pcr_bit = PCR_ARCH_205;
388 guest_pcr_bit = PCR_ARCH_206;
391 guest_pcr_bit = PCR_ARCH_207;
394 guest_pcr_bit = PCR_ARCH_300;
401 /* Check requested PCR bits don't exceed our capabilities */
402 if (guest_pcr_bit > host_pcr_bit)
405 spin_lock(&vc->lock);
406 vc->arch_compat = arch_compat;
407 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
408 vc->pcr = host_pcr_bit - guest_pcr_bit;
409 spin_unlock(&vc->lock);
414 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
418 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
419 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
420 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
421 for (r = 0; r < 16; ++r)
422 pr_err("r%2d = %.16lx r%d = %.16lx\n",
423 r, kvmppc_get_gpr(vcpu, r),
424 r+16, kvmppc_get_gpr(vcpu, r+16));
425 pr_err("ctr = %.16lx lr = %.16lx\n",
426 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
427 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
428 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
429 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
430 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
431 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
432 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
433 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
434 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
435 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
436 pr_err("fault dar = %.16lx dsisr = %.8x\n",
437 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
438 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
439 for (r = 0; r < vcpu->arch.slb_max; ++r)
440 pr_err(" ESID = %.16llx VSID = %.16llx\n",
441 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
442 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
443 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
444 vcpu->arch.last_inst);
447 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
449 return kvm_get_vcpu_by_id(kvm, id);
452 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
454 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
455 vpa->yield_count = cpu_to_be32(1);
458 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
459 unsigned long addr, unsigned long len)
461 /* check address is cacheline aligned */
462 if (addr & (L1_CACHE_BYTES - 1))
464 spin_lock(&vcpu->arch.vpa_update_lock);
465 if (v->next_gpa != addr || v->len != len) {
467 v->len = addr ? len : 0;
468 v->update_pending = 1;
470 spin_unlock(&vcpu->arch.vpa_update_lock);
474 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
483 static int vpa_is_registered(struct kvmppc_vpa *vpap)
485 if (vpap->update_pending)
486 return vpap->next_gpa != 0;
487 return vpap->pinned_addr != NULL;
490 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
492 unsigned long vcpuid, unsigned long vpa)
494 struct kvm *kvm = vcpu->kvm;
495 unsigned long len, nb;
497 struct kvm_vcpu *tvcpu;
500 struct kvmppc_vpa *vpap;
502 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
506 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
507 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
508 subfunc == H_VPA_REG_SLB) {
509 /* Registering new area - address must be cache-line aligned */
510 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
513 /* convert logical addr to kernel addr and read length */
514 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
517 if (subfunc == H_VPA_REG_VPA)
518 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
520 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
521 kvmppc_unpin_guest_page(kvm, va, vpa, false);
524 if (len > nb || len < sizeof(struct reg_vpa))
533 spin_lock(&tvcpu->arch.vpa_update_lock);
536 case H_VPA_REG_VPA: /* register VPA */
538 * The size of our lppaca is 1kB because of the way we align
539 * it for the guest to avoid crossing a 4kB boundary. We only
540 * use 640 bytes of the structure though, so we should accept
541 * clients that set a size of 640.
543 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
544 if (len < sizeof(struct lppaca))
546 vpap = &tvcpu->arch.vpa;
550 case H_VPA_REG_DTL: /* register DTL */
551 if (len < sizeof(struct dtl_entry))
553 len -= len % sizeof(struct dtl_entry);
555 /* Check that they have previously registered a VPA */
557 if (!vpa_is_registered(&tvcpu->arch.vpa))
560 vpap = &tvcpu->arch.dtl;
564 case H_VPA_REG_SLB: /* register SLB shadow buffer */
565 /* Check that they have previously registered a VPA */
567 if (!vpa_is_registered(&tvcpu->arch.vpa))
570 vpap = &tvcpu->arch.slb_shadow;
574 case H_VPA_DEREG_VPA: /* deregister VPA */
575 /* Check they don't still have a DTL or SLB buf registered */
577 if (vpa_is_registered(&tvcpu->arch.dtl) ||
578 vpa_is_registered(&tvcpu->arch.slb_shadow))
581 vpap = &tvcpu->arch.vpa;
585 case H_VPA_DEREG_DTL: /* deregister DTL */
586 vpap = &tvcpu->arch.dtl;
590 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
591 vpap = &tvcpu->arch.slb_shadow;
597 vpap->next_gpa = vpa;
599 vpap->update_pending = 1;
602 spin_unlock(&tvcpu->arch.vpa_update_lock);
607 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
609 struct kvm *kvm = vcpu->kvm;
615 * We need to pin the page pointed to by vpap->next_gpa,
616 * but we can't call kvmppc_pin_guest_page under the lock
617 * as it does get_user_pages() and down_read(). So we
618 * have to drop the lock, pin the page, then get the lock
619 * again and check that a new area didn't get registered
623 gpa = vpap->next_gpa;
624 spin_unlock(&vcpu->arch.vpa_update_lock);
628 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
629 spin_lock(&vcpu->arch.vpa_update_lock);
630 if (gpa == vpap->next_gpa)
632 /* sigh... unpin that one and try again */
634 kvmppc_unpin_guest_page(kvm, va, gpa, false);
637 vpap->update_pending = 0;
638 if (va && nb < vpap->len) {
640 * If it's now too short, it must be that userspace
641 * has changed the mappings underlying guest memory,
642 * so unregister the region.
644 kvmppc_unpin_guest_page(kvm, va, gpa, false);
647 if (vpap->pinned_addr)
648 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
651 vpap->pinned_addr = va;
654 vpap->pinned_end = va + vpap->len;
657 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
659 if (!(vcpu->arch.vpa.update_pending ||
660 vcpu->arch.slb_shadow.update_pending ||
661 vcpu->arch.dtl.update_pending))
664 spin_lock(&vcpu->arch.vpa_update_lock);
665 if (vcpu->arch.vpa.update_pending) {
666 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
667 if (vcpu->arch.vpa.pinned_addr)
668 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
670 if (vcpu->arch.dtl.update_pending) {
671 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
672 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
673 vcpu->arch.dtl_index = 0;
675 if (vcpu->arch.slb_shadow.update_pending)
676 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
677 spin_unlock(&vcpu->arch.vpa_update_lock);
681 * Return the accumulated stolen time for the vcore up until `now'.
682 * The caller should hold the vcore lock.
684 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
689 spin_lock_irqsave(&vc->stoltb_lock, flags);
691 if (vc->vcore_state != VCORE_INACTIVE &&
692 vc->preempt_tb != TB_NIL)
693 p += now - vc->preempt_tb;
694 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
698 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
699 struct kvmppc_vcore *vc)
701 struct dtl_entry *dt;
703 unsigned long stolen;
704 unsigned long core_stolen;
708 dt = vcpu->arch.dtl_ptr;
709 vpa = vcpu->arch.vpa.pinned_addr;
711 core_stolen = vcore_stolen_time(vc, now);
712 stolen = core_stolen - vcpu->arch.stolen_logged;
713 vcpu->arch.stolen_logged = core_stolen;
714 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
715 stolen += vcpu->arch.busy_stolen;
716 vcpu->arch.busy_stolen = 0;
717 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
720 memset(dt, 0, sizeof(struct dtl_entry));
721 dt->dispatch_reason = 7;
722 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
723 dt->timebase = cpu_to_be64(now + vc->tb_offset);
724 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
725 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
726 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
728 if (dt == vcpu->arch.dtl.pinned_end)
729 dt = vcpu->arch.dtl.pinned_addr;
730 vcpu->arch.dtl_ptr = dt;
731 /* order writing *dt vs. writing vpa->dtl_idx */
733 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
734 vcpu->arch.dtl.dirty = true;
737 /* See if there is a doorbell interrupt pending for a vcpu */
738 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
741 struct kvmppc_vcore *vc;
743 if (vcpu->arch.doorbell_request)
746 * Ensure that the read of vcore->dpdes comes after the read
747 * of vcpu->doorbell_request. This barrier matches the
748 * smp_wmb() in kvmppc_guest_entry_inject().
751 vc = vcpu->arch.vcore;
752 thr = vcpu->vcpu_id - vc->first_vcpuid;
753 return !!(vc->dpdes & (1 << thr));
756 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
758 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
760 if ((!vcpu->arch.vcore->arch_compat) &&
761 cpu_has_feature(CPU_FTR_ARCH_207S))
766 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
767 unsigned long resource, unsigned long value1,
768 unsigned long value2)
771 case H_SET_MODE_RESOURCE_SET_CIABR:
772 if (!kvmppc_power8_compatible(vcpu))
777 return H_UNSUPPORTED_FLAG_START;
778 /* Guests can't breakpoint the hypervisor */
779 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
781 vcpu->arch.ciabr = value1;
783 case H_SET_MODE_RESOURCE_SET_DAWR:
784 if (!kvmppc_power8_compatible(vcpu))
786 if (!ppc_breakpoint_available())
789 return H_UNSUPPORTED_FLAG_START;
790 if (value2 & DABRX_HYP)
792 vcpu->arch.dawr = value1;
793 vcpu->arch.dawrx = value2;
800 /* Copy guest memory in place - must reside within a single memslot */
801 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
804 struct kvm_memory_slot *to_memslot = NULL;
805 struct kvm_memory_slot *from_memslot = NULL;
806 unsigned long to_addr, from_addr;
809 /* Get HPA for from address */
810 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
813 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
816 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
817 if (kvm_is_error_hva(from_addr))
819 from_addr |= (from & (PAGE_SIZE - 1));
821 /* Get HPA for to address */
822 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
825 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
828 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
829 if (kvm_is_error_hva(to_addr))
831 to_addr |= (to & (PAGE_SIZE - 1));
834 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
838 mark_page_dirty(kvm, to >> PAGE_SHIFT);
842 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
843 unsigned long dest, unsigned long src)
845 u64 pg_sz = SZ_4K; /* 4K page size */
846 u64 pg_mask = SZ_4K - 1;
849 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
850 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
851 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
854 /* dest (and src if copy_page flag set) must be page aligned */
855 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
858 /* zero and/or copy the page as determined by the flags */
859 if (flags & H_COPY_PAGE) {
860 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
863 } else if (flags & H_ZERO_PAGE) {
864 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
869 /* We can ignore the remaining flags */
874 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
876 struct kvmppc_vcore *vcore = target->arch.vcore;
879 * We expect to have been called by the real mode handler
880 * (kvmppc_rm_h_confer()) which would have directly returned
881 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
882 * have useful work to do and should not confer) so we don't
886 spin_lock(&vcore->lock);
887 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
888 vcore->vcore_state != VCORE_INACTIVE &&
890 target = vcore->runner;
891 spin_unlock(&vcore->lock);
893 return kvm_vcpu_yield_to(target);
896 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
899 struct lppaca *lppaca;
901 spin_lock(&vcpu->arch.vpa_update_lock);
902 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
904 yield_count = be32_to_cpu(lppaca->yield_count);
905 spin_unlock(&vcpu->arch.vpa_update_lock);
909 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
911 unsigned long req = kvmppc_get_gpr(vcpu, 3);
912 unsigned long target, ret = H_SUCCESS;
914 struct kvm_vcpu *tvcpu;
917 if (req <= MAX_HCALL_OPCODE &&
918 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
925 target = kvmppc_get_gpr(vcpu, 4);
926 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
931 tvcpu->arch.prodded = 1;
933 if (tvcpu->arch.ceded)
934 kvmppc_fast_vcpu_kick_hv(tvcpu);
937 target = kvmppc_get_gpr(vcpu, 4);
940 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
945 yield_count = kvmppc_get_gpr(vcpu, 5);
946 if (kvmppc_get_yield_count(tvcpu) != yield_count)
948 kvm_arch_vcpu_yield_to(tvcpu);
951 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
952 kvmppc_get_gpr(vcpu, 5),
953 kvmppc_get_gpr(vcpu, 6));
956 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
959 idx = srcu_read_lock(&vcpu->kvm->srcu);
960 rc = kvmppc_rtas_hcall(vcpu);
961 srcu_read_unlock(&vcpu->kvm->srcu, idx);
968 /* Send the error out to userspace via KVM_RUN */
970 case H_LOGICAL_CI_LOAD:
971 ret = kvmppc_h_logical_ci_load(vcpu);
972 if (ret == H_TOO_HARD)
975 case H_LOGICAL_CI_STORE:
976 ret = kvmppc_h_logical_ci_store(vcpu);
977 if (ret == H_TOO_HARD)
981 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
982 kvmppc_get_gpr(vcpu, 5),
983 kvmppc_get_gpr(vcpu, 6),
984 kvmppc_get_gpr(vcpu, 7));
985 if (ret == H_TOO_HARD)
994 if (kvmppc_xics_enabled(vcpu)) {
995 if (xics_on_xive()) {
996 ret = H_NOT_AVAILABLE;
999 ret = kvmppc_xics_hcall(vcpu, req);
1004 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1007 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1008 kvmppc_get_gpr(vcpu, 5));
1010 #ifdef CONFIG_SPAPR_TCE_IOMMU
1012 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1013 kvmppc_get_gpr(vcpu, 5));
1014 if (ret == H_TOO_HARD)
1018 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1019 kvmppc_get_gpr(vcpu, 5),
1020 kvmppc_get_gpr(vcpu, 6));
1021 if (ret == H_TOO_HARD)
1024 case H_PUT_TCE_INDIRECT:
1025 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1026 kvmppc_get_gpr(vcpu, 5),
1027 kvmppc_get_gpr(vcpu, 6),
1028 kvmppc_get_gpr(vcpu, 7));
1029 if (ret == H_TOO_HARD)
1033 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1034 kvmppc_get_gpr(vcpu, 5),
1035 kvmppc_get_gpr(vcpu, 6),
1036 kvmppc_get_gpr(vcpu, 7));
1037 if (ret == H_TOO_HARD)
1042 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1046 case H_SET_PARTITION_TABLE:
1048 if (nesting_enabled(vcpu->kvm))
1049 ret = kvmhv_set_partition_table(vcpu);
1051 case H_ENTER_NESTED:
1053 if (!nesting_enabled(vcpu->kvm))
1055 ret = kvmhv_enter_nested_guest(vcpu);
1056 if (ret == H_INTERRUPT) {
1057 kvmppc_set_gpr(vcpu, 3, 0);
1058 vcpu->arch.hcall_needed = 0;
1060 } else if (ret == H_TOO_HARD) {
1061 kvmppc_set_gpr(vcpu, 3, 0);
1062 vcpu->arch.hcall_needed = 0;
1066 case H_TLB_INVALIDATE:
1068 if (nesting_enabled(vcpu->kvm))
1069 ret = kvmhv_do_nested_tlbie(vcpu);
1071 case H_COPY_TOFROM_GUEST:
1073 if (nesting_enabled(vcpu->kvm))
1074 ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1077 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1078 kvmppc_get_gpr(vcpu, 5),
1079 kvmppc_get_gpr(vcpu, 6));
1084 kvmppc_set_gpr(vcpu, 3, ret);
1085 vcpu->arch.hcall_needed = 0;
1086 return RESUME_GUEST;
1090 * Handle H_CEDE in the nested virtualization case where we haven't
1091 * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1092 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1093 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1095 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1097 vcpu->arch.shregs.msr |= MSR_EE;
1098 vcpu->arch.ceded = 1;
1100 if (vcpu->arch.prodded) {
1101 vcpu->arch.prodded = 0;
1103 vcpu->arch.ceded = 0;
1107 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1113 case H_REGISTER_VPA:
1115 case H_LOGICAL_CI_LOAD:
1116 case H_LOGICAL_CI_STORE:
1117 #ifdef CONFIG_KVM_XICS
1129 /* See if it's in the real-mode table */
1130 return kvmppc_hcall_impl_hv_realmode(cmd);
1133 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1134 struct kvm_vcpu *vcpu)
1138 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1141 * Fetch failed, so return to guest and
1142 * try executing it again.
1144 return RESUME_GUEST;
1147 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1148 run->exit_reason = KVM_EXIT_DEBUG;
1149 run->debug.arch.address = kvmppc_get_pc(vcpu);
1152 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1153 return RESUME_GUEST;
1157 static void do_nothing(void *x)
1161 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1163 int thr, cpu, pcpu, nthreads;
1165 unsigned long dpdes;
1167 nthreads = vcpu->kvm->arch.emul_smt_mode;
1169 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1170 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1171 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1175 * If the vcpu is currently running on a physical cpu thread,
1176 * interrupt it in order to pull it out of the guest briefly,
1177 * which will update its vcore->dpdes value.
1179 pcpu = READ_ONCE(v->cpu);
1181 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1182 if (kvmppc_doorbell_pending(v))
1189 * On POWER9, emulate doorbell-related instructions in order to
1190 * give the guest the illusion of running on a multi-threaded core.
1191 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1194 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1198 struct kvm *kvm = vcpu->kvm;
1199 struct kvm_vcpu *tvcpu;
1201 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1202 return RESUME_GUEST;
1203 if (get_op(inst) != 31)
1204 return EMULATE_FAIL;
1206 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1207 switch (get_xop(inst)) {
1208 case OP_31_XOP_MSGSNDP:
1209 arg = kvmppc_get_gpr(vcpu, rb);
1210 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1213 if (arg >= kvm->arch.emul_smt_mode)
1215 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1218 if (!tvcpu->arch.doorbell_request) {
1219 tvcpu->arch.doorbell_request = 1;
1220 kvmppc_fast_vcpu_kick_hv(tvcpu);
1223 case OP_31_XOP_MSGCLRP:
1224 arg = kvmppc_get_gpr(vcpu, rb);
1225 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1227 vcpu->arch.vcore->dpdes = 0;
1228 vcpu->arch.doorbell_request = 0;
1230 case OP_31_XOP_MFSPR:
1231 switch (get_sprn(inst)) {
1236 arg = kvmppc_read_dpdes(vcpu);
1239 return EMULATE_FAIL;
1241 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1244 return EMULATE_FAIL;
1246 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1247 return RESUME_GUEST;
1250 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1251 struct task_struct *tsk)
1253 int r = RESUME_HOST;
1255 vcpu->stat.sum_exits++;
1258 * This can happen if an interrupt occurs in the last stages
1259 * of guest entry or the first stages of guest exit (i.e. after
1260 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1261 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1262 * That can happen due to a bug, or due to a machine check
1263 * occurring at just the wrong time.
1265 if (vcpu->arch.shregs.msr & MSR_HV) {
1266 printk(KERN_EMERG "KVM trap in HV mode!\n");
1267 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1268 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1269 vcpu->arch.shregs.msr);
1270 kvmppc_dump_regs(vcpu);
1271 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1272 run->hw.hardware_exit_reason = vcpu->arch.trap;
1275 run->exit_reason = KVM_EXIT_UNKNOWN;
1276 run->ready_for_interrupt_injection = 1;
1277 switch (vcpu->arch.trap) {
1278 /* We're good on these - the host merely wanted to get our attention */
1279 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1280 vcpu->stat.dec_exits++;
1283 case BOOK3S_INTERRUPT_EXTERNAL:
1284 case BOOK3S_INTERRUPT_H_DOORBELL:
1285 case BOOK3S_INTERRUPT_H_VIRT:
1286 vcpu->stat.ext_intr_exits++;
1289 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1290 case BOOK3S_INTERRUPT_HMI:
1291 case BOOK3S_INTERRUPT_PERFMON:
1292 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1295 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1296 /* Print the MCE event to host console. */
1297 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1300 * If the guest can do FWNMI, exit to userspace so it can
1301 * deliver a FWNMI to the guest.
1302 * Otherwise we synthesize a machine check for the guest
1303 * so that it knows that the machine check occurred.
1305 if (!vcpu->kvm->arch.fwnmi_enabled) {
1306 ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1307 kvmppc_core_queue_machine_check(vcpu, flags);
1312 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1313 run->exit_reason = KVM_EXIT_NMI;
1314 run->hw.hardware_exit_reason = vcpu->arch.trap;
1315 /* Clear out the old NMI status from run->flags */
1316 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1317 /* Now set the NMI status */
1318 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1319 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1321 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1325 case BOOK3S_INTERRUPT_PROGRAM:
1329 * Normally program interrupts are delivered directly
1330 * to the guest by the hardware, but we can get here
1331 * as a result of a hypervisor emulation interrupt
1332 * (e40) getting turned into a 700 by BML RTAS.
1334 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1335 kvmppc_core_queue_program(vcpu, flags);
1339 case BOOK3S_INTERRUPT_SYSCALL:
1341 /* hcall - punt to userspace */
1344 /* hypercall with MSR_PR has already been handled in rmode,
1345 * and never reaches here.
1348 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1349 for (i = 0; i < 9; ++i)
1350 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1351 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1352 vcpu->arch.hcall_needed = 1;
1357 * We get these next two if the guest accesses a page which it thinks
1358 * it has mapped but which is not actually present, either because
1359 * it is for an emulated I/O device or because the corresonding
1360 * host page has been paged out. Any other HDSI/HISI interrupts
1361 * have been handled already.
1363 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1364 r = RESUME_PAGE_FAULT;
1366 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1367 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1368 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1369 DSISR_SRR1_MATCH_64S;
1370 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1371 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1372 r = RESUME_PAGE_FAULT;
1375 * This occurs if the guest executes an illegal instruction.
1376 * If the guest debug is disabled, generate a program interrupt
1377 * to the guest. If guest debug is enabled, we need to check
1378 * whether the instruction is a software breakpoint instruction.
1379 * Accordingly return to Guest or Host.
1381 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1382 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1383 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1384 swab32(vcpu->arch.emul_inst) :
1385 vcpu->arch.emul_inst;
1386 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1387 r = kvmppc_emulate_debug_inst(run, vcpu);
1389 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1394 * This occurs if the guest (kernel or userspace), does something that
1395 * is prohibited by HFSCR.
1396 * On POWER9, this could be a doorbell instruction that we need
1398 * Otherwise, we just generate a program interrupt to the guest.
1400 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1402 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1403 cpu_has_feature(CPU_FTR_ARCH_300))
1404 r = kvmppc_emulate_doorbell_instr(vcpu);
1405 if (r == EMULATE_FAIL) {
1406 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1411 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1412 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1414 * This occurs for various TM-related instructions that
1415 * we need to emulate on POWER9 DD2.2. We have already
1416 * handled the cases where the guest was in real-suspend
1417 * mode and was transitioning to transactional state.
1419 r = kvmhv_p9_tm_emulation(vcpu);
1423 case BOOK3S_INTERRUPT_HV_RM_HARD:
1424 r = RESUME_PASSTHROUGH;
1427 kvmppc_dump_regs(vcpu);
1428 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1429 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1430 vcpu->arch.shregs.msr);
1431 run->hw.hardware_exit_reason = vcpu->arch.trap;
1439 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1444 vcpu->stat.sum_exits++;
1447 * This can happen if an interrupt occurs in the last stages
1448 * of guest entry or the first stages of guest exit (i.e. after
1449 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1450 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1451 * That can happen due to a bug, or due to a machine check
1452 * occurring at just the wrong time.
1454 if (vcpu->arch.shregs.msr & MSR_HV) {
1455 pr_emerg("KVM trap in HV mode while nested!\n");
1456 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1457 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1458 vcpu->arch.shregs.msr);
1459 kvmppc_dump_regs(vcpu);
1462 switch (vcpu->arch.trap) {
1463 /* We're good on these - the host merely wanted to get our attention */
1464 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1465 vcpu->stat.dec_exits++;
1468 case BOOK3S_INTERRUPT_EXTERNAL:
1469 vcpu->stat.ext_intr_exits++;
1472 case BOOK3S_INTERRUPT_H_DOORBELL:
1473 case BOOK3S_INTERRUPT_H_VIRT:
1474 vcpu->stat.ext_intr_exits++;
1477 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1478 case BOOK3S_INTERRUPT_HMI:
1479 case BOOK3S_INTERRUPT_PERFMON:
1480 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1483 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1484 /* Pass the machine check to the L1 guest */
1486 /* Print the MCE event to host console. */
1487 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1490 * We get these next two if the guest accesses a page which it thinks
1491 * it has mapped but which is not actually present, either because
1492 * it is for an emulated I/O device or because the corresonding
1493 * host page has been paged out.
1495 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1496 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1497 r = kvmhv_nested_page_fault(run, vcpu);
1498 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1500 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1501 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1502 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1503 DSISR_SRR1_MATCH_64S;
1504 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1505 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1506 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1507 r = kvmhv_nested_page_fault(run, vcpu);
1508 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1511 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1512 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1514 * This occurs for various TM-related instructions that
1515 * we need to emulate on POWER9 DD2.2. We have already
1516 * handled the cases where the guest was in real-suspend
1517 * mode and was transitioning to transactional state.
1519 r = kvmhv_p9_tm_emulation(vcpu);
1523 case BOOK3S_INTERRUPT_HV_RM_HARD:
1524 vcpu->arch.trap = 0;
1526 if (!xics_on_xive())
1527 kvmppc_xics_rm_complete(vcpu, 0);
1537 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1538 struct kvm_sregs *sregs)
1542 memset(sregs, 0, sizeof(struct kvm_sregs));
1543 sregs->pvr = vcpu->arch.pvr;
1544 for (i = 0; i < vcpu->arch.slb_max; i++) {
1545 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1546 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1552 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1553 struct kvm_sregs *sregs)
1557 /* Only accept the same PVR as the host's, since we can't spoof it */
1558 if (sregs->pvr != vcpu->arch.pvr)
1562 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1563 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1564 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1565 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1569 vcpu->arch.slb_max = j;
1574 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1575 bool preserve_top32)
1577 struct kvm *kvm = vcpu->kvm;
1578 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1581 spin_lock(&vc->lock);
1583 * If ILE (interrupt little-endian) has changed, update the
1584 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1586 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1587 struct kvm_vcpu *vcpu;
1590 kvm_for_each_vcpu(i, vcpu, kvm) {
1591 if (vcpu->arch.vcore != vc)
1593 if (new_lpcr & LPCR_ILE)
1594 vcpu->arch.intr_msr |= MSR_LE;
1596 vcpu->arch.intr_msr &= ~MSR_LE;
1601 * Userspace can only modify DPFD (default prefetch depth),
1602 * ILE (interrupt little-endian) and TC (translation control).
1603 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1605 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1606 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1609 * On POWER9, allow userspace to enable large decrementer for the
1610 * guest, whether or not the host has it enabled.
1612 if (cpu_has_feature(CPU_FTR_ARCH_300))
1615 /* Broken 32-bit version of LPCR must not clear top bits */
1618 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1619 spin_unlock(&vc->lock);
1622 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1623 union kvmppc_one_reg *val)
1629 case KVM_REG_PPC_DEBUG_INST:
1630 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1632 case KVM_REG_PPC_HIOR:
1633 *val = get_reg_val(id, 0);
1635 case KVM_REG_PPC_DABR:
1636 *val = get_reg_val(id, vcpu->arch.dabr);
1638 case KVM_REG_PPC_DABRX:
1639 *val = get_reg_val(id, vcpu->arch.dabrx);
1641 case KVM_REG_PPC_DSCR:
1642 *val = get_reg_val(id, vcpu->arch.dscr);
1644 case KVM_REG_PPC_PURR:
1645 *val = get_reg_val(id, vcpu->arch.purr);
1647 case KVM_REG_PPC_SPURR:
1648 *val = get_reg_val(id, vcpu->arch.spurr);
1650 case KVM_REG_PPC_AMR:
1651 *val = get_reg_val(id, vcpu->arch.amr);
1653 case KVM_REG_PPC_UAMOR:
1654 *val = get_reg_val(id, vcpu->arch.uamor);
1656 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1657 i = id - KVM_REG_PPC_MMCR0;
1658 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1660 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1661 i = id - KVM_REG_PPC_PMC1;
1662 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1664 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1665 i = id - KVM_REG_PPC_SPMC1;
1666 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1668 case KVM_REG_PPC_SIAR:
1669 *val = get_reg_val(id, vcpu->arch.siar);
1671 case KVM_REG_PPC_SDAR:
1672 *val = get_reg_val(id, vcpu->arch.sdar);
1674 case KVM_REG_PPC_SIER:
1675 *val = get_reg_val(id, vcpu->arch.sier);
1677 case KVM_REG_PPC_IAMR:
1678 *val = get_reg_val(id, vcpu->arch.iamr);
1680 case KVM_REG_PPC_PSPB:
1681 *val = get_reg_val(id, vcpu->arch.pspb);
1683 case KVM_REG_PPC_DPDES:
1684 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1686 case KVM_REG_PPC_VTB:
1687 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1689 case KVM_REG_PPC_DAWR:
1690 *val = get_reg_val(id, vcpu->arch.dawr);
1692 case KVM_REG_PPC_DAWRX:
1693 *val = get_reg_val(id, vcpu->arch.dawrx);
1695 case KVM_REG_PPC_CIABR:
1696 *val = get_reg_val(id, vcpu->arch.ciabr);
1698 case KVM_REG_PPC_CSIGR:
1699 *val = get_reg_val(id, vcpu->arch.csigr);
1701 case KVM_REG_PPC_TACR:
1702 *val = get_reg_val(id, vcpu->arch.tacr);
1704 case KVM_REG_PPC_TCSCR:
1705 *val = get_reg_val(id, vcpu->arch.tcscr);
1707 case KVM_REG_PPC_PID:
1708 *val = get_reg_val(id, vcpu->arch.pid);
1710 case KVM_REG_PPC_ACOP:
1711 *val = get_reg_val(id, vcpu->arch.acop);
1713 case KVM_REG_PPC_WORT:
1714 *val = get_reg_val(id, vcpu->arch.wort);
1716 case KVM_REG_PPC_TIDR:
1717 *val = get_reg_val(id, vcpu->arch.tid);
1719 case KVM_REG_PPC_PSSCR:
1720 *val = get_reg_val(id, vcpu->arch.psscr);
1722 case KVM_REG_PPC_VPA_ADDR:
1723 spin_lock(&vcpu->arch.vpa_update_lock);
1724 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1725 spin_unlock(&vcpu->arch.vpa_update_lock);
1727 case KVM_REG_PPC_VPA_SLB:
1728 spin_lock(&vcpu->arch.vpa_update_lock);
1729 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1730 val->vpaval.length = vcpu->arch.slb_shadow.len;
1731 spin_unlock(&vcpu->arch.vpa_update_lock);
1733 case KVM_REG_PPC_VPA_DTL:
1734 spin_lock(&vcpu->arch.vpa_update_lock);
1735 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1736 val->vpaval.length = vcpu->arch.dtl.len;
1737 spin_unlock(&vcpu->arch.vpa_update_lock);
1739 case KVM_REG_PPC_TB_OFFSET:
1740 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1742 case KVM_REG_PPC_LPCR:
1743 case KVM_REG_PPC_LPCR_64:
1744 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1746 case KVM_REG_PPC_PPR:
1747 *val = get_reg_val(id, vcpu->arch.ppr);
1749 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1750 case KVM_REG_PPC_TFHAR:
1751 *val = get_reg_val(id, vcpu->arch.tfhar);
1753 case KVM_REG_PPC_TFIAR:
1754 *val = get_reg_val(id, vcpu->arch.tfiar);
1756 case KVM_REG_PPC_TEXASR:
1757 *val = get_reg_val(id, vcpu->arch.texasr);
1759 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1760 i = id - KVM_REG_PPC_TM_GPR0;
1761 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1763 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1766 i = id - KVM_REG_PPC_TM_VSR0;
1768 for (j = 0; j < TS_FPRWIDTH; j++)
1769 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1771 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1772 val->vval = vcpu->arch.vr_tm.vr[i-32];
1778 case KVM_REG_PPC_TM_CR:
1779 *val = get_reg_val(id, vcpu->arch.cr_tm);
1781 case KVM_REG_PPC_TM_XER:
1782 *val = get_reg_val(id, vcpu->arch.xer_tm);
1784 case KVM_REG_PPC_TM_LR:
1785 *val = get_reg_val(id, vcpu->arch.lr_tm);
1787 case KVM_REG_PPC_TM_CTR:
1788 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1790 case KVM_REG_PPC_TM_FPSCR:
1791 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1793 case KVM_REG_PPC_TM_AMR:
1794 *val = get_reg_val(id, vcpu->arch.amr_tm);
1796 case KVM_REG_PPC_TM_PPR:
1797 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1799 case KVM_REG_PPC_TM_VRSAVE:
1800 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1802 case KVM_REG_PPC_TM_VSCR:
1803 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1804 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1808 case KVM_REG_PPC_TM_DSCR:
1809 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1811 case KVM_REG_PPC_TM_TAR:
1812 *val = get_reg_val(id, vcpu->arch.tar_tm);
1815 case KVM_REG_PPC_ARCH_COMPAT:
1816 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1818 case KVM_REG_PPC_DEC_EXPIRY:
1819 *val = get_reg_val(id, vcpu->arch.dec_expires +
1820 vcpu->arch.vcore->tb_offset);
1822 case KVM_REG_PPC_ONLINE:
1823 *val = get_reg_val(id, vcpu->arch.online);
1825 case KVM_REG_PPC_PTCR:
1826 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1836 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1837 union kvmppc_one_reg *val)
1841 unsigned long addr, len;
1844 case KVM_REG_PPC_HIOR:
1845 /* Only allow this to be set to zero */
1846 if (set_reg_val(id, *val))
1849 case KVM_REG_PPC_DABR:
1850 vcpu->arch.dabr = set_reg_val(id, *val);
1852 case KVM_REG_PPC_DABRX:
1853 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1855 case KVM_REG_PPC_DSCR:
1856 vcpu->arch.dscr = set_reg_val(id, *val);
1858 case KVM_REG_PPC_PURR:
1859 vcpu->arch.purr = set_reg_val(id, *val);
1861 case KVM_REG_PPC_SPURR:
1862 vcpu->arch.spurr = set_reg_val(id, *val);
1864 case KVM_REG_PPC_AMR:
1865 vcpu->arch.amr = set_reg_val(id, *val);
1867 case KVM_REG_PPC_UAMOR:
1868 vcpu->arch.uamor = set_reg_val(id, *val);
1870 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1871 i = id - KVM_REG_PPC_MMCR0;
1872 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1874 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1875 i = id - KVM_REG_PPC_PMC1;
1876 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1878 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1879 i = id - KVM_REG_PPC_SPMC1;
1880 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1882 case KVM_REG_PPC_SIAR:
1883 vcpu->arch.siar = set_reg_val(id, *val);
1885 case KVM_REG_PPC_SDAR:
1886 vcpu->arch.sdar = set_reg_val(id, *val);
1888 case KVM_REG_PPC_SIER:
1889 vcpu->arch.sier = set_reg_val(id, *val);
1891 case KVM_REG_PPC_IAMR:
1892 vcpu->arch.iamr = set_reg_val(id, *val);
1894 case KVM_REG_PPC_PSPB:
1895 vcpu->arch.pspb = set_reg_val(id, *val);
1897 case KVM_REG_PPC_DPDES:
1898 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1900 case KVM_REG_PPC_VTB:
1901 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1903 case KVM_REG_PPC_DAWR:
1904 vcpu->arch.dawr = set_reg_val(id, *val);
1906 case KVM_REG_PPC_DAWRX:
1907 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1909 case KVM_REG_PPC_CIABR:
1910 vcpu->arch.ciabr = set_reg_val(id, *val);
1911 /* Don't allow setting breakpoints in hypervisor code */
1912 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1913 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1915 case KVM_REG_PPC_CSIGR:
1916 vcpu->arch.csigr = set_reg_val(id, *val);
1918 case KVM_REG_PPC_TACR:
1919 vcpu->arch.tacr = set_reg_val(id, *val);
1921 case KVM_REG_PPC_TCSCR:
1922 vcpu->arch.tcscr = set_reg_val(id, *val);
1924 case KVM_REG_PPC_PID:
1925 vcpu->arch.pid = set_reg_val(id, *val);
1927 case KVM_REG_PPC_ACOP:
1928 vcpu->arch.acop = set_reg_val(id, *val);
1930 case KVM_REG_PPC_WORT:
1931 vcpu->arch.wort = set_reg_val(id, *val);
1933 case KVM_REG_PPC_TIDR:
1934 vcpu->arch.tid = set_reg_val(id, *val);
1936 case KVM_REG_PPC_PSSCR:
1937 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1939 case KVM_REG_PPC_VPA_ADDR:
1940 addr = set_reg_val(id, *val);
1942 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1943 vcpu->arch.dtl.next_gpa))
1945 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1947 case KVM_REG_PPC_VPA_SLB:
1948 addr = val->vpaval.addr;
1949 len = val->vpaval.length;
1951 if (addr && !vcpu->arch.vpa.next_gpa)
1953 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1955 case KVM_REG_PPC_VPA_DTL:
1956 addr = val->vpaval.addr;
1957 len = val->vpaval.length;
1959 if (addr && (len < sizeof(struct dtl_entry) ||
1960 !vcpu->arch.vpa.next_gpa))
1962 len -= len % sizeof(struct dtl_entry);
1963 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1965 case KVM_REG_PPC_TB_OFFSET:
1966 /* round up to multiple of 2^24 */
1967 vcpu->arch.vcore->tb_offset =
1968 ALIGN(set_reg_val(id, *val), 1UL << 24);
1970 case KVM_REG_PPC_LPCR:
1971 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1973 case KVM_REG_PPC_LPCR_64:
1974 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1976 case KVM_REG_PPC_PPR:
1977 vcpu->arch.ppr = set_reg_val(id, *val);
1979 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1980 case KVM_REG_PPC_TFHAR:
1981 vcpu->arch.tfhar = set_reg_val(id, *val);
1983 case KVM_REG_PPC_TFIAR:
1984 vcpu->arch.tfiar = set_reg_val(id, *val);
1986 case KVM_REG_PPC_TEXASR:
1987 vcpu->arch.texasr = set_reg_val(id, *val);
1989 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1990 i = id - KVM_REG_PPC_TM_GPR0;
1991 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1993 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1996 i = id - KVM_REG_PPC_TM_VSR0;
1998 for (j = 0; j < TS_FPRWIDTH; j++)
1999 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2001 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2002 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2007 case KVM_REG_PPC_TM_CR:
2008 vcpu->arch.cr_tm = set_reg_val(id, *val);
2010 case KVM_REG_PPC_TM_XER:
2011 vcpu->arch.xer_tm = set_reg_val(id, *val);
2013 case KVM_REG_PPC_TM_LR:
2014 vcpu->arch.lr_tm = set_reg_val(id, *val);
2016 case KVM_REG_PPC_TM_CTR:
2017 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2019 case KVM_REG_PPC_TM_FPSCR:
2020 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2022 case KVM_REG_PPC_TM_AMR:
2023 vcpu->arch.amr_tm = set_reg_val(id, *val);
2025 case KVM_REG_PPC_TM_PPR:
2026 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2028 case KVM_REG_PPC_TM_VRSAVE:
2029 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2031 case KVM_REG_PPC_TM_VSCR:
2032 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2033 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2037 case KVM_REG_PPC_TM_DSCR:
2038 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2040 case KVM_REG_PPC_TM_TAR:
2041 vcpu->arch.tar_tm = set_reg_val(id, *val);
2044 case KVM_REG_PPC_ARCH_COMPAT:
2045 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2047 case KVM_REG_PPC_DEC_EXPIRY:
2048 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2049 vcpu->arch.vcore->tb_offset;
2051 case KVM_REG_PPC_ONLINE:
2052 i = set_reg_val(id, *val);
2053 if (i && !vcpu->arch.online)
2054 atomic_inc(&vcpu->arch.vcore->online_count);
2055 else if (!i && vcpu->arch.online)
2056 atomic_dec(&vcpu->arch.vcore->online_count);
2057 vcpu->arch.online = i;
2059 case KVM_REG_PPC_PTCR:
2060 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2071 * On POWER9, threads are independent and can be in different partitions.
2072 * Therefore we consider each thread to be a subcore.
2073 * There is a restriction that all threads have to be in the same
2074 * MMU mode (radix or HPT), unfortunately, but since we only support
2075 * HPT guests on a HPT host so far, that isn't an impediment yet.
2077 static int threads_per_vcore(struct kvm *kvm)
2079 if (kvm->arch.threads_indep)
2081 return threads_per_subcore;
2084 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2086 struct kvmppc_vcore *vcore;
2088 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2093 spin_lock_init(&vcore->lock);
2094 spin_lock_init(&vcore->stoltb_lock);
2095 init_swait_queue_head(&vcore->wq);
2096 vcore->preempt_tb = TB_NIL;
2097 vcore->lpcr = kvm->arch.lpcr;
2098 vcore->first_vcpuid = id;
2100 INIT_LIST_HEAD(&vcore->preempt_list);
2105 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2106 static struct debugfs_timings_element {
2110 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2111 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2112 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2113 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2114 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2117 #define N_TIMINGS (ARRAY_SIZE(timings))
2119 struct debugfs_timings_state {
2120 struct kvm_vcpu *vcpu;
2121 unsigned int buflen;
2122 char buf[N_TIMINGS * 100];
2125 static int debugfs_timings_open(struct inode *inode, struct file *file)
2127 struct kvm_vcpu *vcpu = inode->i_private;
2128 struct debugfs_timings_state *p;
2130 p = kzalloc(sizeof(*p), GFP_KERNEL);
2134 kvm_get_kvm(vcpu->kvm);
2136 file->private_data = p;
2138 return nonseekable_open(inode, file);
2141 static int debugfs_timings_release(struct inode *inode, struct file *file)
2143 struct debugfs_timings_state *p = file->private_data;
2145 kvm_put_kvm(p->vcpu->kvm);
2150 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2151 size_t len, loff_t *ppos)
2153 struct debugfs_timings_state *p = file->private_data;
2154 struct kvm_vcpu *vcpu = p->vcpu;
2156 struct kvmhv_tb_accumulator tb;
2165 buf_end = s + sizeof(p->buf);
2166 for (i = 0; i < N_TIMINGS; ++i) {
2167 struct kvmhv_tb_accumulator *acc;
2169 acc = (struct kvmhv_tb_accumulator *)
2170 ((unsigned long)vcpu + timings[i].offset);
2172 for (loops = 0; loops < 1000; ++loops) {
2173 count = acc->seqcount;
2178 if (count == acc->seqcount) {
2186 snprintf(s, buf_end - s, "%s: stuck\n",
2189 snprintf(s, buf_end - s,
2190 "%s: %llu %llu %llu %llu\n",
2191 timings[i].name, count / 2,
2192 tb_to_ns(tb.tb_total),
2193 tb_to_ns(tb.tb_min),
2194 tb_to_ns(tb.tb_max));
2197 p->buflen = s - p->buf;
2201 if (pos >= p->buflen)
2203 if (len > p->buflen - pos)
2204 len = p->buflen - pos;
2205 n = copy_to_user(buf, p->buf + pos, len);
2215 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2216 size_t len, loff_t *ppos)
2221 static const struct file_operations debugfs_timings_ops = {
2222 .owner = THIS_MODULE,
2223 .open = debugfs_timings_open,
2224 .release = debugfs_timings_release,
2225 .read = debugfs_timings_read,
2226 .write = debugfs_timings_write,
2227 .llseek = generic_file_llseek,
2230 /* Create a debugfs directory for the vcpu */
2231 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2234 struct kvm *kvm = vcpu->kvm;
2236 snprintf(buf, sizeof(buf), "vcpu%u", id);
2237 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2239 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2240 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2242 vcpu->arch.debugfs_timings =
2243 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2244 vcpu, &debugfs_timings_ops);
2247 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2248 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2251 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2253 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2256 struct kvm_vcpu *vcpu;
2259 struct kvmppc_vcore *vcore;
2262 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2266 err = kvm_vcpu_init(vcpu, kvm, id);
2270 vcpu->arch.shared = &vcpu->arch.shregs;
2271 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2273 * The shared struct is never shared on HV,
2274 * so we can always use host endianness
2276 #ifdef __BIG_ENDIAN__
2277 vcpu->arch.shared_big_endian = true;
2279 vcpu->arch.shared_big_endian = false;
2282 vcpu->arch.mmcr[0] = MMCR0_FC;
2283 vcpu->arch.ctrl = CTRL_RUNLATCH;
2284 /* default to host PVR, since we can't spoof it */
2285 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2286 spin_lock_init(&vcpu->arch.vpa_update_lock);
2287 spin_lock_init(&vcpu->arch.tbacct_lock);
2288 vcpu->arch.busy_preempt = TB_NIL;
2289 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2292 * Set the default HFSCR for the guest from the host value.
2293 * This value is only used on POWER9.
2294 * On POWER9, we want to virtualize the doorbell facility, so we
2295 * don't set the HFSCR_MSGP bit, and that causes those instructions
2296 * to trap and then we emulate them.
2298 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2299 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2300 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2301 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2302 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2303 vcpu->arch.hfscr |= HFSCR_TM;
2305 if (cpu_has_feature(CPU_FTR_TM_COMP))
2306 vcpu->arch.hfscr |= HFSCR_TM;
2308 kvmppc_mmu_book3s_hv_init(vcpu);
2310 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2312 init_waitqueue_head(&vcpu->arch.cpu_run);
2314 mutex_lock(&kvm->lock);
2317 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2318 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2319 pr_devel("KVM: VCPU ID too high\n");
2320 core = KVM_MAX_VCORES;
2322 BUG_ON(kvm->arch.smt_mode != 1);
2323 core = kvmppc_pack_vcpu_id(kvm, id);
2326 core = id / kvm->arch.smt_mode;
2328 if (core < KVM_MAX_VCORES) {
2329 vcore = kvm->arch.vcores[core];
2330 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2331 pr_devel("KVM: collision on id %u", id);
2333 } else if (!vcore) {
2335 * Take mmu_setup_lock for mutual exclusion
2336 * with kvmppc_update_lpcr().
2339 vcore = kvmppc_vcore_create(kvm,
2340 id & ~(kvm->arch.smt_mode - 1));
2341 mutex_lock(&kvm->arch.mmu_setup_lock);
2342 kvm->arch.vcores[core] = vcore;
2343 kvm->arch.online_vcores++;
2344 mutex_unlock(&kvm->arch.mmu_setup_lock);
2347 mutex_unlock(&kvm->lock);
2352 spin_lock(&vcore->lock);
2353 ++vcore->num_threads;
2354 spin_unlock(&vcore->lock);
2355 vcpu->arch.vcore = vcore;
2356 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2357 vcpu->arch.thread_cpu = -1;
2358 vcpu->arch.prev_cpu = -1;
2360 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2361 kvmppc_sanity_check(vcpu);
2363 debugfs_vcpu_init(vcpu, id);
2368 kmem_cache_free(kvm_vcpu_cache, vcpu);
2370 return ERR_PTR(err);
2373 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2374 unsigned long flags)
2381 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2383 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2385 * On POWER8 (or POWER7), the threading mode is "strict",
2386 * so we pack smt_mode vcpus per vcore.
2388 if (smt_mode > threads_per_subcore)
2392 * On POWER9, the threading mode is "loose",
2393 * so each vcpu gets its own vcore.
2398 mutex_lock(&kvm->lock);
2400 if (!kvm->arch.online_vcores) {
2401 kvm->arch.smt_mode = smt_mode;
2402 kvm->arch.emul_smt_mode = esmt;
2405 mutex_unlock(&kvm->lock);
2410 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2412 if (vpa->pinned_addr)
2413 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2417 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2419 spin_lock(&vcpu->arch.vpa_update_lock);
2420 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2421 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2422 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2423 spin_unlock(&vcpu->arch.vpa_update_lock);
2424 kvm_vcpu_uninit(vcpu);
2425 kmem_cache_free(kvm_vcpu_cache, vcpu);
2428 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2430 /* Indicate we want to get back into the guest */
2434 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2436 unsigned long dec_nsec, now;
2439 if (now > vcpu->arch.dec_expires) {
2440 /* decrementer has already gone negative */
2441 kvmppc_core_queue_dec(vcpu);
2442 kvmppc_core_prepare_to_enter(vcpu);
2445 dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2446 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2447 vcpu->arch.timer_running = 1;
2450 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2452 vcpu->arch.ceded = 0;
2453 if (vcpu->arch.timer_running) {
2454 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2455 vcpu->arch.timer_running = 0;
2459 extern int __kvmppc_vcore_entry(void);
2461 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2462 struct kvm_vcpu *vcpu)
2466 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2468 spin_lock_irq(&vcpu->arch.tbacct_lock);
2470 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2471 vcpu->arch.stolen_logged;
2472 vcpu->arch.busy_preempt = now;
2473 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2474 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2476 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2479 static int kvmppc_grab_hwthread(int cpu)
2481 struct paca_struct *tpaca;
2482 long timeout = 10000;
2484 tpaca = paca_ptrs[cpu];
2486 /* Ensure the thread won't go into the kernel if it wakes */
2487 tpaca->kvm_hstate.kvm_vcpu = NULL;
2488 tpaca->kvm_hstate.kvm_vcore = NULL;
2489 tpaca->kvm_hstate.napping = 0;
2491 tpaca->kvm_hstate.hwthread_req = 1;
2494 * If the thread is already executing in the kernel (e.g. handling
2495 * a stray interrupt), wait for it to get back to nap mode.
2496 * The smp_mb() is to ensure that our setting of hwthread_req
2497 * is visible before we look at hwthread_state, so if this
2498 * races with the code at system_reset_pSeries and the thread
2499 * misses our setting of hwthread_req, we are sure to see its
2500 * setting of hwthread_state, and vice versa.
2503 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2504 if (--timeout <= 0) {
2505 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2513 static void kvmppc_release_hwthread(int cpu)
2515 struct paca_struct *tpaca;
2517 tpaca = paca_ptrs[cpu];
2518 tpaca->kvm_hstate.hwthread_req = 0;
2519 tpaca->kvm_hstate.kvm_vcpu = NULL;
2520 tpaca->kvm_hstate.kvm_vcore = NULL;
2521 tpaca->kvm_hstate.kvm_split_mode = NULL;
2524 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2526 struct kvm_nested_guest *nested = vcpu->arch.nested;
2527 cpumask_t *cpu_in_guest;
2530 cpu = cpu_first_thread_sibling(cpu);
2532 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2533 cpu_in_guest = &nested->cpu_in_guest;
2535 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2536 cpu_in_guest = &kvm->arch.cpu_in_guest;
2539 * Make sure setting of bit in need_tlb_flush precedes
2540 * testing of cpu_in_guest bits. The matching barrier on
2541 * the other side is the first smp_mb() in kvmppc_run_core().
2544 for (i = 0; i < threads_per_core; ++i)
2545 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2546 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2549 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2551 struct kvm_nested_guest *nested = vcpu->arch.nested;
2552 struct kvm *kvm = vcpu->kvm;
2555 if (!cpu_has_feature(CPU_FTR_HVMODE))
2559 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2561 prev_cpu = vcpu->arch.prev_cpu;
2564 * With radix, the guest can do TLB invalidations itself,
2565 * and it could choose to use the local form (tlbiel) if
2566 * it is invalidating a translation that has only ever been
2567 * used on one vcpu. However, that doesn't mean it has
2568 * only ever been used on one physical cpu, since vcpus
2569 * can move around between pcpus. To cope with this, when
2570 * a vcpu moves from one pcpu to another, we need to tell
2571 * any vcpus running on the same core as this vcpu previously
2572 * ran to flush the TLB. The TLB is shared between threads,
2573 * so we use a single bit in .need_tlb_flush for all 4 threads.
2575 if (prev_cpu != pcpu) {
2576 if (prev_cpu >= 0 &&
2577 cpu_first_thread_sibling(prev_cpu) !=
2578 cpu_first_thread_sibling(pcpu))
2579 radix_flush_cpu(kvm, prev_cpu, vcpu);
2581 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2583 vcpu->arch.prev_cpu = pcpu;
2587 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2590 struct paca_struct *tpaca;
2591 struct kvm *kvm = vc->kvm;
2595 if (vcpu->arch.timer_running) {
2596 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2597 vcpu->arch.timer_running = 0;
2599 cpu += vcpu->arch.ptid;
2600 vcpu->cpu = vc->pcpu;
2601 vcpu->arch.thread_cpu = cpu;
2602 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2604 tpaca = paca_ptrs[cpu];
2605 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2606 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2607 tpaca->kvm_hstate.fake_suspend = 0;
2608 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2610 tpaca->kvm_hstate.kvm_vcore = vc;
2611 if (cpu != smp_processor_id())
2612 kvmppc_ipi_thread(cpu);
2615 static void kvmppc_wait_for_nap(int n_threads)
2617 int cpu = smp_processor_id();
2622 for (loops = 0; loops < 1000000; ++loops) {
2624 * Check if all threads are finished.
2625 * We set the vcore pointer when starting a thread
2626 * and the thread clears it when finished, so we look
2627 * for any threads that still have a non-NULL vcore ptr.
2629 for (i = 1; i < n_threads; ++i)
2630 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2632 if (i == n_threads) {
2639 for (i = 1; i < n_threads; ++i)
2640 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2641 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2645 * Check that we are on thread 0 and that any other threads in
2646 * this core are off-line. Then grab the threads so they can't
2649 static int on_primary_thread(void)
2651 int cpu = smp_processor_id();
2654 /* Are we on a primary subcore? */
2655 if (cpu_thread_in_subcore(cpu))
2659 while (++thr < threads_per_subcore)
2660 if (cpu_online(cpu + thr))
2663 /* Grab all hw threads so they can't go into the kernel */
2664 for (thr = 1; thr < threads_per_subcore; ++thr) {
2665 if (kvmppc_grab_hwthread(cpu + thr)) {
2666 /* Couldn't grab one; let the others go */
2668 kvmppc_release_hwthread(cpu + thr);
2669 } while (--thr > 0);
2677 * A list of virtual cores for each physical CPU.
2678 * These are vcores that could run but their runner VCPU tasks are
2679 * (or may be) preempted.
2681 struct preempted_vcore_list {
2682 struct list_head list;
2686 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2688 static void init_vcore_lists(void)
2692 for_each_possible_cpu(cpu) {
2693 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2694 spin_lock_init(&lp->lock);
2695 INIT_LIST_HEAD(&lp->list);
2699 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2701 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2703 vc->vcore_state = VCORE_PREEMPT;
2704 vc->pcpu = smp_processor_id();
2705 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2706 spin_lock(&lp->lock);
2707 list_add_tail(&vc->preempt_list, &lp->list);
2708 spin_unlock(&lp->lock);
2711 /* Start accumulating stolen time */
2712 kvmppc_core_start_stolen(vc);
2715 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2717 struct preempted_vcore_list *lp;
2719 kvmppc_core_end_stolen(vc);
2720 if (!list_empty(&vc->preempt_list)) {
2721 lp = &per_cpu(preempted_vcores, vc->pcpu);
2722 spin_lock(&lp->lock);
2723 list_del_init(&vc->preempt_list);
2724 spin_unlock(&lp->lock);
2726 vc->vcore_state = VCORE_INACTIVE;
2730 * This stores information about the virtual cores currently
2731 * assigned to a physical core.
2735 int max_subcore_threads;
2737 int subcore_threads[MAX_SUBCORES];
2738 struct kvmppc_vcore *vc[MAX_SUBCORES];
2742 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2743 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2745 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2747 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2749 memset(cip, 0, sizeof(*cip));
2750 cip->n_subcores = 1;
2751 cip->max_subcore_threads = vc->num_threads;
2752 cip->total_threads = vc->num_threads;
2753 cip->subcore_threads[0] = vc->num_threads;
2757 static bool subcore_config_ok(int n_subcores, int n_threads)
2760 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2761 * split-core mode, with one thread per subcore.
2763 if (cpu_has_feature(CPU_FTR_ARCH_300))
2764 return n_subcores <= 4 && n_threads == 1;
2766 /* On POWER8, can only dynamically split if unsplit to begin with */
2767 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2769 if (n_subcores > MAX_SUBCORES)
2771 if (n_subcores > 1) {
2772 if (!(dynamic_mt_modes & 2))
2774 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2778 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2781 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2783 vc->entry_exit_map = 0;
2785 vc->napping_threads = 0;
2786 vc->conferring_threads = 0;
2787 vc->tb_offset_applied = 0;
2790 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2792 int n_threads = vc->num_threads;
2795 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2798 /* In one_vm_per_core mode, require all vcores to be from the same vm */
2799 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2802 /* Some POWER9 chips require all threads to be in the same MMU mode */
2803 if (no_mixing_hpt_and_radix &&
2804 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2807 if (n_threads < cip->max_subcore_threads)
2808 n_threads = cip->max_subcore_threads;
2809 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2811 cip->max_subcore_threads = n_threads;
2813 sub = cip->n_subcores;
2815 cip->total_threads += vc->num_threads;
2816 cip->subcore_threads[sub] = vc->num_threads;
2818 init_vcore_to_run(vc);
2819 list_del_init(&vc->preempt_list);
2825 * Work out whether it is possible to piggyback the execution of
2826 * vcore *pvc onto the execution of the other vcores described in *cip.
2828 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2831 if (cip->total_threads + pvc->num_threads > target_threads)
2834 return can_dynamic_split(pvc, cip);
2837 static void prepare_threads(struct kvmppc_vcore *vc)
2840 struct kvm_vcpu *vcpu;
2842 for_each_runnable_thread(i, vcpu, vc) {
2843 if (signal_pending(vcpu->arch.run_task))
2844 vcpu->arch.ret = -EINTR;
2845 else if (vcpu->arch.vpa.update_pending ||
2846 vcpu->arch.slb_shadow.update_pending ||
2847 vcpu->arch.dtl.update_pending)
2848 vcpu->arch.ret = RESUME_GUEST;
2851 kvmppc_remove_runnable(vc, vcpu);
2852 wake_up(&vcpu->arch.cpu_run);
2856 static void collect_piggybacks(struct core_info *cip, int target_threads)
2858 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2859 struct kvmppc_vcore *pvc, *vcnext;
2861 spin_lock(&lp->lock);
2862 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2863 if (!spin_trylock(&pvc->lock))
2865 prepare_threads(pvc);
2866 if (!pvc->n_runnable) {
2867 list_del_init(&pvc->preempt_list);
2868 if (pvc->runner == NULL) {
2869 pvc->vcore_state = VCORE_INACTIVE;
2870 kvmppc_core_end_stolen(pvc);
2872 spin_unlock(&pvc->lock);
2875 if (!can_piggyback(pvc, cip, target_threads)) {
2876 spin_unlock(&pvc->lock);
2879 kvmppc_core_end_stolen(pvc);
2880 pvc->vcore_state = VCORE_PIGGYBACK;
2881 if (cip->total_threads >= target_threads)
2884 spin_unlock(&lp->lock);
2887 static bool recheck_signals(struct core_info *cip)
2890 struct kvm_vcpu *vcpu;
2892 for (sub = 0; sub < cip->n_subcores; ++sub)
2893 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2894 if (signal_pending(vcpu->arch.run_task))
2899 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2901 int still_running = 0, i;
2904 struct kvm_vcpu *vcpu;
2906 spin_lock(&vc->lock);
2908 for_each_runnable_thread(i, vcpu, vc) {
2910 * It's safe to unlock the vcore in the loop here, because
2911 * for_each_runnable_thread() is safe against removal of
2912 * the vcpu, and the vcore state is VCORE_EXITING here,
2913 * so any vcpus becoming runnable will have their arch.trap
2914 * set to zero and can't actually run in the guest.
2916 spin_unlock(&vc->lock);
2917 /* cancel pending dec exception if dec is positive */
2918 if (now < vcpu->arch.dec_expires &&
2919 kvmppc_core_pending_dec(vcpu))
2920 kvmppc_core_dequeue_dec(vcpu);
2922 trace_kvm_guest_exit(vcpu);
2925 if (vcpu->arch.trap)
2926 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2927 vcpu->arch.run_task);
2929 vcpu->arch.ret = ret;
2930 vcpu->arch.trap = 0;
2932 spin_lock(&vc->lock);
2933 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2934 if (vcpu->arch.pending_exceptions)
2935 kvmppc_core_prepare_to_enter(vcpu);
2936 if (vcpu->arch.ceded)
2937 kvmppc_set_timer(vcpu);
2941 kvmppc_remove_runnable(vc, vcpu);
2942 wake_up(&vcpu->arch.cpu_run);
2946 if (still_running > 0) {
2947 kvmppc_vcore_preempt(vc);
2948 } else if (vc->runner) {
2949 vc->vcore_state = VCORE_PREEMPT;
2950 kvmppc_core_start_stolen(vc);
2952 vc->vcore_state = VCORE_INACTIVE;
2954 if (vc->n_runnable > 0 && vc->runner == NULL) {
2955 /* make sure there's a candidate runner awake */
2957 vcpu = next_runnable_thread(vc, &i);
2958 wake_up(&vcpu->arch.cpu_run);
2961 spin_unlock(&vc->lock);
2965 * Clear core from the list of active host cores as we are about to
2966 * enter the guest. Only do this if it is the primary thread of the
2967 * core (not if a subcore) that is entering the guest.
2969 static inline int kvmppc_clear_host_core(unsigned int cpu)
2973 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2976 * Memory barrier can be omitted here as we will do a smp_wmb()
2977 * later in kvmppc_start_thread and we need ensure that state is
2978 * visible to other CPUs only after we enter guest.
2980 core = cpu >> threads_shift;
2981 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2986 * Advertise this core as an active host core since we exited the guest
2987 * Only need to do this if it is the primary thread of the core that is
2990 static inline int kvmppc_set_host_core(unsigned int cpu)
2994 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2998 * Memory barrier can be omitted here because we do a spin_unlock
2999 * immediately after this which provides the memory barrier.
3001 core = cpu >> threads_shift;
3002 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3006 static void set_irq_happened(int trap)
3009 case BOOK3S_INTERRUPT_EXTERNAL:
3010 local_paca->irq_happened |= PACA_IRQ_EE;
3012 case BOOK3S_INTERRUPT_H_DOORBELL:
3013 local_paca->irq_happened |= PACA_IRQ_DBELL;
3015 case BOOK3S_INTERRUPT_HMI:
3016 local_paca->irq_happened |= PACA_IRQ_HMI;
3018 case BOOK3S_INTERRUPT_SYSTEM_RESET:
3019 replay_system_reset();
3025 * Run a set of guest threads on a physical core.
3026 * Called with vc->lock held.
3028 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3030 struct kvm_vcpu *vcpu;
3033 struct core_info core_info;
3034 struct kvmppc_vcore *pvc;
3035 struct kvm_split_mode split_info, *sip;
3036 int split, subcore_size, active;
3039 unsigned long cmd_bit, stat_bit;
3042 int controlled_threads;
3048 * Remove from the list any threads that have a signal pending
3049 * or need a VPA update done
3051 prepare_threads(vc);
3053 /* if the runner is no longer runnable, let the caller pick a new one */
3054 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3060 init_vcore_to_run(vc);
3061 vc->preempt_tb = TB_NIL;
3064 * Number of threads that we will be controlling: the same as
3065 * the number of threads per subcore, except on POWER9,
3066 * where it's 1 because the threads are (mostly) independent.
3068 controlled_threads = threads_per_vcore(vc->kvm);
3071 * Make sure we are running on primary threads, and that secondary
3072 * threads are offline. Also check if the number of threads in this
3073 * guest are greater than the current system threads per guest.
3074 * On POWER9, we need to be not in independent-threads mode if
3075 * this is a HPT guest on a radix host machine where the
3076 * CPU threads may not be in different MMU modes.
3078 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3079 !kvm_is_radix(vc->kvm);
3080 if (((controlled_threads > 1) &&
3081 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3082 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3083 for_each_runnable_thread(i, vcpu, vc) {
3084 vcpu->arch.ret = -EBUSY;
3085 kvmppc_remove_runnable(vc, vcpu);
3086 wake_up(&vcpu->arch.cpu_run);
3092 * See if we could run any other vcores on the physical core
3093 * along with this one.
3095 init_core_info(&core_info, vc);
3096 pcpu = smp_processor_id();
3097 target_threads = controlled_threads;
3098 if (target_smt_mode && target_smt_mode < target_threads)
3099 target_threads = target_smt_mode;
3100 if (vc->num_threads < target_threads)
3101 collect_piggybacks(&core_info, target_threads);
3104 * On radix, arrange for TLB flushing if necessary.
3105 * This has to be done before disabling interrupts since
3106 * it uses smp_call_function().
3108 pcpu = smp_processor_id();
3109 if (kvm_is_radix(vc->kvm)) {
3110 for (sub = 0; sub < core_info.n_subcores; ++sub)
3111 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3112 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3116 * Hard-disable interrupts, and check resched flag and signals.
3117 * If we need to reschedule or deliver a signal, clean up
3118 * and return without going into the guest(s).
3119 * If the mmu_ready flag has been cleared, don't go into the
3120 * guest because that means a HPT resize operation is in progress.
3122 local_irq_disable();
3124 if (lazy_irq_pending() || need_resched() ||
3125 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3127 vc->vcore_state = VCORE_INACTIVE;
3128 /* Unlock all except the primary vcore */
3129 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3130 pvc = core_info.vc[sub];
3131 /* Put back on to the preempted vcores list */
3132 kvmppc_vcore_preempt(pvc);
3133 spin_unlock(&pvc->lock);
3135 for (i = 0; i < controlled_threads; ++i)
3136 kvmppc_release_hwthread(pcpu + i);
3140 kvmppc_clear_host_core(pcpu);
3142 /* Decide on micro-threading (split-core) mode */
3143 subcore_size = threads_per_subcore;
3144 cmd_bit = stat_bit = 0;
3145 split = core_info.n_subcores;
3147 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3148 && !cpu_has_feature(CPU_FTR_ARCH_300);
3150 if (split > 1 || hpt_on_radix) {
3152 memset(&split_info, 0, sizeof(split_info));
3153 for (sub = 0; sub < core_info.n_subcores; ++sub)
3154 split_info.vc[sub] = core_info.vc[sub];
3157 if (split == 2 && (dynamic_mt_modes & 2)) {
3158 cmd_bit = HID0_POWER8_1TO2LPAR;
3159 stat_bit = HID0_POWER8_2LPARMODE;
3162 cmd_bit = HID0_POWER8_1TO4LPAR;
3163 stat_bit = HID0_POWER8_4LPARMODE;
3165 subcore_size = MAX_SMT_THREADS / split;
3166 split_info.rpr = mfspr(SPRN_RPR);
3167 split_info.pmmar = mfspr(SPRN_PMMAR);
3168 split_info.ldbar = mfspr(SPRN_LDBAR);
3169 split_info.subcore_size = subcore_size;
3171 split_info.subcore_size = 1;
3173 /* Use the split_info for LPCR/LPIDR changes */
3174 split_info.lpcr_req = vc->lpcr;
3175 split_info.lpidr_req = vc->kvm->arch.lpid;
3176 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3177 split_info.do_set = 1;
3181 /* order writes to split_info before kvm_split_mode pointer */
3185 for (thr = 0; thr < controlled_threads; ++thr) {
3186 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3188 paca->kvm_hstate.tid = thr;
3189 paca->kvm_hstate.napping = 0;
3190 paca->kvm_hstate.kvm_split_mode = sip;
3193 /* Initiate micro-threading (split-core) on POWER8 if required */
3195 unsigned long hid0 = mfspr(SPRN_HID0);
3197 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3199 mtspr(SPRN_HID0, hid0);
3202 hid0 = mfspr(SPRN_HID0);
3203 if (hid0 & stat_bit)
3210 * On POWER8, set RWMR register.
3211 * Since it only affects PURR and SPURR, it doesn't affect
3212 * the host, so we don't save/restore the host value.
3215 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3216 int n_online = atomic_read(&vc->online_count);
3219 * Use the 8-thread value if we're doing split-core
3220 * or if the vcore's online count looks bogus.
3222 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3223 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3224 rwmr_val = p8_rwmr_values[n_online];
3225 mtspr(SPRN_RWMR, rwmr_val);
3228 /* Start all the threads */
3230 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3231 thr = is_power8 ? subcore_thread_map[sub] : sub;
3234 pvc = core_info.vc[sub];
3235 pvc->pcpu = pcpu + thr;
3236 for_each_runnable_thread(i, vcpu, pvc) {
3237 kvmppc_start_thread(vcpu, pvc);
3238 kvmppc_create_dtl_entry(vcpu, pvc);
3239 trace_kvm_guest_enter(vcpu);
3240 if (!vcpu->arch.ptid)
3242 active |= 1 << (thr + vcpu->arch.ptid);
3245 * We need to start the first thread of each subcore
3246 * even if it doesn't have a vcpu.
3249 kvmppc_start_thread(NULL, pvc);
3253 * Ensure that split_info.do_nap is set after setting
3254 * the vcore pointer in the PACA of the secondaries.
3259 * When doing micro-threading, poke the inactive threads as well.
3260 * This gets them to the nap instruction after kvm_do_nap,
3261 * which reduces the time taken to unsplit later.
3262 * For POWER9 HPT guest on radix host, we need all the secondary
3263 * threads woken up so they can do the LPCR/LPIDR change.
3265 if (cmd_bit || hpt_on_radix) {
3266 split_info.do_nap = 1; /* ask secondaries to nap when done */
3267 for (thr = 1; thr < threads_per_subcore; ++thr)
3268 if (!(active & (1 << thr)))
3269 kvmppc_ipi_thread(pcpu + thr);
3272 vc->vcore_state = VCORE_RUNNING;
3275 trace_kvmppc_run_core(vc, 0);
3277 for (sub = 0; sub < core_info.n_subcores; ++sub)
3278 spin_unlock(&core_info.vc[sub]->lock);
3280 guest_enter_irqoff();
3282 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3284 this_cpu_disable_ftrace();
3287 * Interrupts will be enabled once we get into the guest,
3288 * so tell lockdep that we're about to enable interrupts.
3290 trace_hardirqs_on();
3292 trap = __kvmppc_vcore_entry();
3294 trace_hardirqs_off();
3296 this_cpu_enable_ftrace();
3298 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3300 set_irq_happened(trap);
3302 spin_lock(&vc->lock);
3303 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3304 vc->vcore_state = VCORE_EXITING;
3306 /* wait for secondary threads to finish writing their state to memory */
3307 kvmppc_wait_for_nap(controlled_threads);
3309 /* Return to whole-core mode if we split the core earlier */
3311 unsigned long hid0 = mfspr(SPRN_HID0);
3312 unsigned long loops = 0;
3314 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3315 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3317 mtspr(SPRN_HID0, hid0);
3320 hid0 = mfspr(SPRN_HID0);
3321 if (!(hid0 & stat_bit))
3326 } else if (hpt_on_radix) {
3327 /* Wait for all threads to have seen final sync */
3328 for (thr = 1; thr < controlled_threads; ++thr) {
3329 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3331 while (paca->kvm_hstate.kvm_split_mode) {
3338 split_info.do_nap = 0;
3340 kvmppc_set_host_core(pcpu);
3345 /* Let secondaries go back to the offline loop */
3346 for (i = 0; i < controlled_threads; ++i) {
3347 kvmppc_release_hwthread(pcpu + i);
3348 if (sip && sip->napped[i])
3349 kvmppc_ipi_thread(pcpu + i);
3350 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3353 spin_unlock(&vc->lock);
3355 /* make sure updates to secondary vcpu structs are visible now */
3360 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3361 pvc = core_info.vc[sub];
3362 post_guest_process(pvc, pvc == vc);
3365 spin_lock(&vc->lock);
3368 vc->vcore_state = VCORE_INACTIVE;
3369 trace_kvmppc_run_core(vc, 1);
3373 * Load up hypervisor-mode registers on P9.
3375 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3378 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3380 u64 tb, purr, spurr;
3382 unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3383 unsigned long host_ciabr = mfspr(SPRN_CIABR);
3384 unsigned long host_dawr = mfspr(SPRN_DAWR);
3385 unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3386 unsigned long host_psscr = mfspr(SPRN_PSSCR);
3387 unsigned long host_pidr = mfspr(SPRN_PID);
3389 hdec = time_limit - mftb();
3391 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3392 mtspr(SPRN_HDEC, hdec);
3394 if (vc->tb_offset) {
3395 u64 new_tb = mftb() + vc->tb_offset;
3396 mtspr(SPRN_TBU40, new_tb);
3398 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3399 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3400 vc->tb_offset_applied = vc->tb_offset;
3404 mtspr(SPRN_PCR, vc->pcr);
3405 mtspr(SPRN_DPDES, vc->dpdes);
3406 mtspr(SPRN_VTB, vc->vtb);
3408 local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3409 local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3410 mtspr(SPRN_PURR, vcpu->arch.purr);
3411 mtspr(SPRN_SPURR, vcpu->arch.spurr);
3413 if (dawr_enabled()) {
3414 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3415 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3417 mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3418 mtspr(SPRN_IC, vcpu->arch.ic);
3419 mtspr(SPRN_PID, vcpu->arch.pid);
3421 mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3422 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3424 mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3426 mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3427 mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3428 mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3429 mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3431 mtspr(SPRN_AMOR, ~0UL);
3433 mtspr(SPRN_LPCR, lpcr);
3436 kvmppc_xive_push_vcpu(vcpu);
3438 mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3439 mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3441 trap = __kvmhv_vcpu_entry_p9(vcpu);
3443 /* Advance host PURR/SPURR by the amount used by guest */
3444 purr = mfspr(SPRN_PURR);
3445 spurr = mfspr(SPRN_SPURR);
3446 mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3447 purr - vcpu->arch.purr);
3448 mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3449 spurr - vcpu->arch.spurr);
3450 vcpu->arch.purr = purr;
3451 vcpu->arch.spurr = spurr;
3453 vcpu->arch.ic = mfspr(SPRN_IC);
3454 vcpu->arch.pid = mfspr(SPRN_PID);
3455 vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3457 vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3458 vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3459 vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3460 vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3462 /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3463 mtspr(SPRN_PSSCR, host_psscr |
3464 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3465 mtspr(SPRN_HFSCR, host_hfscr);
3466 mtspr(SPRN_CIABR, host_ciabr);
3467 mtspr(SPRN_DAWR, host_dawr);
3468 mtspr(SPRN_DAWRX, host_dawrx);
3469 mtspr(SPRN_PID, host_pidr);
3472 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3473 * case we interrupted the guest between a tlbie and a ptesync.
3475 asm volatile("eieio; tlbsync; ptesync");
3477 mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid); /* restore host LPID */
3480 vc->dpdes = mfspr(SPRN_DPDES);
3481 vc->vtb = mfspr(SPRN_VTB);
3482 mtspr(SPRN_DPDES, 0);
3486 if (vc->tb_offset_applied) {
3487 u64 new_tb = mftb() - vc->tb_offset_applied;
3488 mtspr(SPRN_TBU40, new_tb);
3490 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3491 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3492 vc->tb_offset_applied = 0;
3495 mtspr(SPRN_HDEC, 0x7fffffff);
3496 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3502 * Virtual-mode guest entry for POWER9 and later when the host and
3503 * guest are both using the radix MMU. The LPIDR has already been set.
3505 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3508 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3509 unsigned long host_dscr = mfspr(SPRN_DSCR);
3510 unsigned long host_tidr = mfspr(SPRN_TIDR);
3511 unsigned long host_iamr = mfspr(SPRN_IAMR);
3512 unsigned long host_amr = mfspr(SPRN_AMR);
3517 dec = mfspr(SPRN_DEC);
3520 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3521 local_paca->kvm_hstate.dec_expires = dec + tb;
3522 if (local_paca->kvm_hstate.dec_expires < time_limit)
3523 time_limit = local_paca->kvm_hstate.dec_expires;
3525 vcpu->arch.ceded = 0;
3527 kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */
3529 kvmppc_subcore_enter_guest();
3531 vc->entry_exit_map = 1;
3534 if (vcpu->arch.vpa.pinned_addr) {
3535 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3536 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3537 lp->yield_count = cpu_to_be32(yield_count);
3538 vcpu->arch.vpa.dirty = 1;
3541 if (cpu_has_feature(CPU_FTR_TM) ||
3542 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3543 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3545 kvmhv_load_guest_pmu(vcpu);
3547 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3548 load_fp_state(&vcpu->arch.fp);
3549 #ifdef CONFIG_ALTIVEC
3550 load_vr_state(&vcpu->arch.vr);
3552 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3554 mtspr(SPRN_DSCR, vcpu->arch.dscr);
3555 mtspr(SPRN_IAMR, vcpu->arch.iamr);
3556 mtspr(SPRN_PSPB, vcpu->arch.pspb);
3557 mtspr(SPRN_FSCR, vcpu->arch.fscr);
3558 mtspr(SPRN_TAR, vcpu->arch.tar);
3559 mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3560 mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3561 mtspr(SPRN_BESCR, vcpu->arch.bescr);
3562 mtspr(SPRN_WORT, vcpu->arch.wort);
3563 mtspr(SPRN_TIDR, vcpu->arch.tid);
3564 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3565 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3566 mtspr(SPRN_AMR, vcpu->arch.amr);
3567 mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3569 if (!(vcpu->arch.ctrl & 1))
3570 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3572 mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3574 if (kvmhv_on_pseries()) {
3575 /* call our hypervisor to load up HV regs and go */
3576 struct hv_guest_state hvregs;
3578 kvmhv_save_hv_regs(vcpu, &hvregs);
3580 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3581 hvregs.version = HV_GUEST_STATE_VERSION;
3582 if (vcpu->arch.nested) {
3583 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3584 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3586 hvregs.lpid = vcpu->kvm->arch.lpid;
3587 hvregs.vcpu_token = vcpu->vcpu_id;
3589 hvregs.hdec_expiry = time_limit;
3590 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3591 __pa(&vcpu->arch.regs));
3592 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3593 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3594 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3595 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3597 /* H_CEDE has to be handled now, not later */
3598 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3599 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3600 kvmppc_nested_cede(vcpu);
3604 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3607 vcpu->arch.slb_max = 0;
3608 dec = mfspr(SPRN_DEC);
3610 vcpu->arch.dec_expires = dec + tb;
3612 vcpu->arch.thread_cpu = -1;
3613 vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3615 vcpu->arch.iamr = mfspr(SPRN_IAMR);
3616 vcpu->arch.pspb = mfspr(SPRN_PSPB);
3617 vcpu->arch.fscr = mfspr(SPRN_FSCR);
3618 vcpu->arch.tar = mfspr(SPRN_TAR);
3619 vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3620 vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3621 vcpu->arch.bescr = mfspr(SPRN_BESCR);
3622 vcpu->arch.wort = mfspr(SPRN_WORT);
3623 vcpu->arch.tid = mfspr(SPRN_TIDR);
3624 vcpu->arch.amr = mfspr(SPRN_AMR);
3625 vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3626 vcpu->arch.dscr = mfspr(SPRN_DSCR);
3628 mtspr(SPRN_PSPB, 0);
3629 mtspr(SPRN_WORT, 0);
3630 mtspr(SPRN_UAMOR, 0);
3631 mtspr(SPRN_DSCR, host_dscr);
3632 mtspr(SPRN_TIDR, host_tidr);
3633 mtspr(SPRN_IAMR, host_iamr);
3634 mtspr(SPRN_PSPB, 0);
3636 if (host_amr != vcpu->arch.amr)
3637 mtspr(SPRN_AMR, host_amr);
3639 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3640 store_fp_state(&vcpu->arch.fp);
3641 #ifdef CONFIG_ALTIVEC
3642 store_vr_state(&vcpu->arch.vr);
3644 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3646 if (cpu_has_feature(CPU_FTR_TM) ||
3647 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3648 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3651 if (vcpu->arch.vpa.pinned_addr) {
3652 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3653 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3654 lp->yield_count = cpu_to_be32(yield_count);
3655 vcpu->arch.vpa.dirty = 1;
3656 save_pmu = lp->pmcregs_in_use;
3659 kvmhv_save_guest_pmu(vcpu, save_pmu);
3661 vc->entry_exit_map = 0x101;
3664 mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3665 mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3667 kvmhv_load_host_pmu();
3669 kvmppc_subcore_exit_guest();
3675 * Wait for some other vcpu thread to execute us, and
3676 * wake us up when we need to handle something in the host.
3678 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3679 struct kvm_vcpu *vcpu, int wait_state)
3683 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3684 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3685 spin_unlock(&vc->lock);
3687 spin_lock(&vc->lock);
3689 finish_wait(&vcpu->arch.cpu_run, &wait);
3692 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3694 if (!halt_poll_ns_grow)
3697 vc->halt_poll_ns *= halt_poll_ns_grow;
3698 if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3699 vc->halt_poll_ns = halt_poll_ns_grow_start;
3702 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3704 if (halt_poll_ns_shrink == 0)
3705 vc->halt_poll_ns = 0;
3707 vc->halt_poll_ns /= halt_poll_ns_shrink;
3710 #ifdef CONFIG_KVM_XICS
3711 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3713 if (!xics_on_xive())
3715 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3716 vcpu->arch.xive_saved_state.cppr;
3719 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3723 #endif /* CONFIG_KVM_XICS */
3725 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3727 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3728 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3735 * Check to see if any of the runnable vcpus on the vcore have pending
3736 * exceptions or are no longer ceded
3738 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3740 struct kvm_vcpu *vcpu;
3743 for_each_runnable_thread(i, vcpu, vc) {
3744 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3752 * All the vcpus in this vcore are idle, so wait for a decrementer
3753 * or external interrupt to one of the vcpus. vc->lock is held.
3755 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3757 ktime_t cur, start_poll, start_wait;
3760 DECLARE_SWAITQUEUE(wait);
3762 /* Poll for pending exceptions and ceded state */
3763 cur = start_poll = ktime_get();
3764 if (vc->halt_poll_ns) {
3765 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3766 ++vc->runner->stat.halt_attempted_poll;
3768 vc->vcore_state = VCORE_POLLING;
3769 spin_unlock(&vc->lock);
3772 if (kvmppc_vcore_check_block(vc)) {
3777 } while (single_task_running() && ktime_before(cur, stop));
3779 spin_lock(&vc->lock);
3780 vc->vcore_state = VCORE_INACTIVE;
3783 ++vc->runner->stat.halt_successful_poll;
3788 prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3790 if (kvmppc_vcore_check_block(vc)) {
3791 finish_swait(&vc->wq, &wait);
3793 /* If we polled, count this as a successful poll */
3794 if (vc->halt_poll_ns)
3795 ++vc->runner->stat.halt_successful_poll;
3799 start_wait = ktime_get();
3801 vc->vcore_state = VCORE_SLEEPING;
3802 trace_kvmppc_vcore_blocked(vc, 0);
3803 spin_unlock(&vc->lock);
3805 finish_swait(&vc->wq, &wait);
3806 spin_lock(&vc->lock);
3807 vc->vcore_state = VCORE_INACTIVE;
3808 trace_kvmppc_vcore_blocked(vc, 1);
3809 ++vc->runner->stat.halt_successful_wait;
3814 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3816 /* Attribute wait time */
3818 vc->runner->stat.halt_wait_ns +=
3819 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3820 /* Attribute failed poll time */
3821 if (vc->halt_poll_ns)
3822 vc->runner->stat.halt_poll_fail_ns +=
3823 ktime_to_ns(start_wait) -
3824 ktime_to_ns(start_poll);
3826 /* Attribute successful poll time */
3827 if (vc->halt_poll_ns)
3828 vc->runner->stat.halt_poll_success_ns +=
3830 ktime_to_ns(start_poll);
3833 /* Adjust poll time */
3835 if (block_ns <= vc->halt_poll_ns)
3837 /* We slept and blocked for longer than the max halt time */
3838 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3839 shrink_halt_poll_ns(vc);
3840 /* We slept and our poll time is too small */
3841 else if (vc->halt_poll_ns < halt_poll_ns &&
3842 block_ns < halt_poll_ns)
3843 grow_halt_poll_ns(vc);
3844 if (vc->halt_poll_ns > halt_poll_ns)
3845 vc->halt_poll_ns = halt_poll_ns;
3847 vc->halt_poll_ns = 0;
3849 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3853 * This never fails for a radix guest, as none of the operations it does
3854 * for a radix guest can fail or have a way to report failure.
3855 * kvmhv_run_single_vcpu() relies on this fact.
3857 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3860 struct kvm *kvm = vcpu->kvm;
3862 mutex_lock(&kvm->arch.mmu_setup_lock);
3863 if (!kvm->arch.mmu_ready) {
3864 if (!kvm_is_radix(kvm))
3865 r = kvmppc_hv_setup_htab_rma(vcpu);
3867 if (cpu_has_feature(CPU_FTR_ARCH_300))
3868 kvmppc_setup_partition_table(kvm);
3869 kvm->arch.mmu_ready = 1;
3872 mutex_unlock(&kvm->arch.mmu_setup_lock);
3876 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3879 struct kvmppc_vcore *vc;
3882 trace_kvmppc_run_vcpu_enter(vcpu);
3884 kvm_run->exit_reason = 0;
3885 vcpu->arch.ret = RESUME_GUEST;
3886 vcpu->arch.trap = 0;
3887 kvmppc_update_vpas(vcpu);
3890 * Synchronize with other threads in this virtual core
3892 vc = vcpu->arch.vcore;
3893 spin_lock(&vc->lock);
3894 vcpu->arch.ceded = 0;
3895 vcpu->arch.run_task = current;
3896 vcpu->arch.kvm_run = kvm_run;
3897 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3898 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3899 vcpu->arch.busy_preempt = TB_NIL;
3900 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3904 * This happens the first time this is called for a vcpu.
3905 * If the vcore is already running, we may be able to start
3906 * this thread straight away and have it join in.
3908 if (!signal_pending(current)) {
3909 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3910 vc->vcore_state == VCORE_RUNNING) &&
3911 !VCORE_IS_EXITING(vc)) {
3912 kvmppc_create_dtl_entry(vcpu, vc);
3913 kvmppc_start_thread(vcpu, vc);
3914 trace_kvm_guest_enter(vcpu);
3915 } else if (vc->vcore_state == VCORE_SLEEPING) {
3916 swake_up_one(&vc->wq);
3921 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3922 !signal_pending(current)) {
3923 /* See if the MMU is ready to go */
3924 if (!vcpu->kvm->arch.mmu_ready) {
3925 spin_unlock(&vc->lock);
3926 r = kvmhv_setup_mmu(vcpu);
3927 spin_lock(&vc->lock);
3929 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3930 kvm_run->fail_entry.
3931 hardware_entry_failure_reason = 0;
3937 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3938 kvmppc_vcore_end_preempt(vc);
3940 if (vc->vcore_state != VCORE_INACTIVE) {
3941 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3944 for_each_runnable_thread(i, v, vc) {
3945 kvmppc_core_prepare_to_enter(v);
3946 if (signal_pending(v->arch.run_task)) {
3947 kvmppc_remove_runnable(vc, v);
3948 v->stat.signal_exits++;
3949 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3950 v->arch.ret = -EINTR;
3951 wake_up(&v->arch.cpu_run);
3954 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3957 for_each_runnable_thread(i, v, vc) {
3958 if (!kvmppc_vcpu_woken(v))
3959 n_ceded += v->arch.ceded;
3964 if (n_ceded == vc->n_runnable) {
3965 kvmppc_vcore_blocked(vc);
3966 } else if (need_resched()) {
3967 kvmppc_vcore_preempt(vc);
3968 /* Let something else run */
3969 cond_resched_lock(&vc->lock);
3970 if (vc->vcore_state == VCORE_PREEMPT)
3971 kvmppc_vcore_end_preempt(vc);
3973 kvmppc_run_core(vc);
3978 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3979 (vc->vcore_state == VCORE_RUNNING ||
3980 vc->vcore_state == VCORE_EXITING ||
3981 vc->vcore_state == VCORE_PIGGYBACK))
3982 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3984 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3985 kvmppc_vcore_end_preempt(vc);
3987 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3988 kvmppc_remove_runnable(vc, vcpu);
3989 vcpu->stat.signal_exits++;
3990 kvm_run->exit_reason = KVM_EXIT_INTR;
3991 vcpu->arch.ret = -EINTR;
3994 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3995 /* Wake up some vcpu to run the core */
3997 v = next_runnable_thread(vc, &i);
3998 wake_up(&v->arch.cpu_run);
4001 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4002 spin_unlock(&vc->lock);
4003 return vcpu->arch.ret;
4006 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4007 struct kvm_vcpu *vcpu, u64 time_limit,
4012 struct kvmppc_vcore *vc;
4013 struct kvm *kvm = vcpu->kvm;
4014 struct kvm_nested_guest *nested = vcpu->arch.nested;
4016 trace_kvmppc_run_vcpu_enter(vcpu);
4018 kvm_run->exit_reason = 0;
4019 vcpu->arch.ret = RESUME_GUEST;
4020 vcpu->arch.trap = 0;
4022 vc = vcpu->arch.vcore;
4023 vcpu->arch.ceded = 0;
4024 vcpu->arch.run_task = current;
4025 vcpu->arch.kvm_run = kvm_run;
4026 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4027 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4028 vcpu->arch.busy_preempt = TB_NIL;
4029 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4030 vc->runnable_threads[0] = vcpu;
4034 /* See if the MMU is ready to go */
4035 if (!kvm->arch.mmu_ready)
4036 kvmhv_setup_mmu(vcpu);
4041 kvmppc_update_vpas(vcpu);
4043 init_vcore_to_run(vc);
4044 vc->preempt_tb = TB_NIL;
4047 pcpu = smp_processor_id();
4049 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4051 local_irq_disable();
4053 if (signal_pending(current))
4055 if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4059 kvmppc_core_prepare_to_enter(vcpu);
4060 if (vcpu->arch.doorbell_request) {
4063 vcpu->arch.doorbell_request = 0;
4065 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4066 &vcpu->arch.pending_exceptions))
4068 } else if (vcpu->arch.pending_exceptions ||
4069 vcpu->arch.doorbell_request ||
4070 xive_interrupt_pending(vcpu)) {
4071 vcpu->arch.ret = RESUME_HOST;
4075 kvmppc_clear_host_core(pcpu);
4077 local_paca->kvm_hstate.tid = 0;
4078 local_paca->kvm_hstate.napping = 0;
4079 local_paca->kvm_hstate.kvm_split_mode = NULL;
4080 kvmppc_start_thread(vcpu, vc);
4081 kvmppc_create_dtl_entry(vcpu, vc);
4082 trace_kvm_guest_enter(vcpu);
4084 vc->vcore_state = VCORE_RUNNING;
4085 trace_kvmppc_run_core(vc, 0);
4087 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4088 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4089 mtspr(SPRN_LPID, lpid);
4091 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4094 guest_enter_irqoff();
4096 srcu_idx = srcu_read_lock(&kvm->srcu);
4098 this_cpu_disable_ftrace();
4100 /* Tell lockdep that we're about to enable interrupts */
4101 trace_hardirqs_on();
4103 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4104 vcpu->arch.trap = trap;
4106 trace_hardirqs_off();
4108 this_cpu_enable_ftrace();
4110 srcu_read_unlock(&kvm->srcu, srcu_idx);
4112 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4113 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4117 set_irq_happened(trap);
4119 kvmppc_set_host_core(pcpu);
4124 cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4128 /* cancel pending decrementer exception if DEC is now positive */
4129 if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4130 kvmppc_core_dequeue_dec(vcpu);
4132 trace_kvm_guest_exit(vcpu);
4136 r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4138 r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4142 if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4143 !kvmppc_vcpu_woken(vcpu)) {
4144 kvmppc_set_timer(vcpu);
4145 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4146 if (signal_pending(current)) {
4147 vcpu->stat.signal_exits++;
4148 kvm_run->exit_reason = KVM_EXIT_INTR;
4149 vcpu->arch.ret = -EINTR;
4152 spin_lock(&vc->lock);
4153 kvmppc_vcore_blocked(vc);
4154 spin_unlock(&vc->lock);
4157 vcpu->arch.ceded = 0;
4159 vc->vcore_state = VCORE_INACTIVE;
4160 trace_kvmppc_run_core(vc, 1);
4163 kvmppc_remove_runnable(vc, vcpu);
4164 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4166 return vcpu->arch.ret;
4169 vcpu->stat.signal_exits++;
4170 kvm_run->exit_reason = KVM_EXIT_INTR;
4171 vcpu->arch.ret = -EINTR;
4178 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4182 unsigned long ebb_regs[3] = {}; /* shut up GCC */
4183 unsigned long user_tar = 0;
4184 unsigned int user_vrsave;
4187 if (!vcpu->arch.sane) {
4188 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4193 * Don't allow entry with a suspended transaction, because
4194 * the guest entry/exit code will lose it.
4195 * If the guest has TM enabled, save away their TM-related SPRs
4196 * (they will get restored by the TM unavailable interrupt).
4198 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4199 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4200 (current->thread.regs->msr & MSR_TM)) {
4201 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4202 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4203 run->fail_entry.hardware_entry_failure_reason = 0;
4206 /* Enable TM so we can read the TM SPRs */
4207 mtmsr(mfmsr() | MSR_TM);
4208 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4209 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4210 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4211 current->thread.regs->msr &= ~MSR_TM;
4216 * Force online to 1 for the sake of old userspace which doesn't
4219 if (!vcpu->arch.online) {
4220 atomic_inc(&vcpu->arch.vcore->online_count);
4221 vcpu->arch.online = 1;
4224 kvmppc_core_prepare_to_enter(vcpu);
4226 /* No need to go into the guest when all we'll do is come back out */
4227 if (signal_pending(current)) {
4228 run->exit_reason = KVM_EXIT_INTR;
4233 atomic_inc(&kvm->arch.vcpus_running);
4234 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4237 flush_all_to_thread(current);
4239 /* Save userspace EBB and other register values */
4240 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4241 ebb_regs[0] = mfspr(SPRN_EBBHR);
4242 ebb_regs[1] = mfspr(SPRN_EBBRR);
4243 ebb_regs[2] = mfspr(SPRN_BESCR);
4244 user_tar = mfspr(SPRN_TAR);
4246 user_vrsave = mfspr(SPRN_VRSAVE);
4248 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4249 vcpu->arch.pgdir = current->mm->pgd;
4250 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4254 * The early POWER9 chips that can't mix radix and HPT threads
4255 * on the same core also need the workaround for the problem
4256 * where the TLB would prefetch entries in the guest exit path
4257 * for radix guests using the guest PIDR value and LPID 0.
4258 * The workaround is in the old path (kvmppc_run_vcpu())
4259 * but not the new path (kvmhv_run_single_vcpu()).
4261 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4262 !no_mixing_hpt_and_radix)
4263 r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4264 vcpu->arch.vcore->lpcr);
4266 r = kvmppc_run_vcpu(run, vcpu);
4268 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4269 !(vcpu->arch.shregs.msr & MSR_PR)) {
4270 trace_kvm_hcall_enter(vcpu);
4271 r = kvmppc_pseries_do_hcall(vcpu);
4272 trace_kvm_hcall_exit(vcpu, r);
4273 kvmppc_core_prepare_to_enter(vcpu);
4274 } else if (r == RESUME_PAGE_FAULT) {
4275 srcu_idx = srcu_read_lock(&kvm->srcu);
4276 r = kvmppc_book3s_hv_page_fault(run, vcpu,
4277 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4278 srcu_read_unlock(&kvm->srcu, srcu_idx);
4279 } else if (r == RESUME_PASSTHROUGH) {
4280 if (WARN_ON(xics_on_xive()))
4283 r = kvmppc_xics_rm_complete(vcpu, 0);
4285 } while (is_kvmppc_resume_guest(r));
4287 /* Restore userspace EBB and other register values */
4288 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4289 mtspr(SPRN_EBBHR, ebb_regs[0]);
4290 mtspr(SPRN_EBBRR, ebb_regs[1]);
4291 mtspr(SPRN_BESCR, ebb_regs[2]);
4292 mtspr(SPRN_TAR, user_tar);
4293 mtspr(SPRN_FSCR, current->thread.fscr);
4295 mtspr(SPRN_VRSAVE, user_vrsave);
4297 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4298 atomic_dec(&kvm->arch.vcpus_running);
4302 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4303 int shift, int sllp)
4305 (*sps)->page_shift = shift;
4306 (*sps)->slb_enc = sllp;
4307 (*sps)->enc[0].page_shift = shift;
4308 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4310 * Add 16MB MPSS support (may get filtered out by userspace)
4313 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4315 (*sps)->enc[1].page_shift = 24;
4316 (*sps)->enc[1].pte_enc = penc;
4322 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4323 struct kvm_ppc_smmu_info *info)
4325 struct kvm_ppc_one_seg_page_size *sps;
4328 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4329 * POWER7 doesn't support keys for instruction accesses,
4330 * POWER8 and POWER9 do.
4332 info->data_keys = 32;
4333 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4335 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4336 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4337 info->slb_size = 32;
4339 /* We only support these sizes for now, and no muti-size segments */
4340 sps = &info->sps[0];
4341 kvmppc_add_seg_page_size(&sps, 12, 0);
4342 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4343 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4345 /* If running as a nested hypervisor, we don't support HPT guests */
4346 if (kvmhv_on_pseries())
4347 info->flags |= KVM_PPC_NO_HASH;
4353 * Get (and clear) the dirty memory log for a memory slot.
4355 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4356 struct kvm_dirty_log *log)
4358 struct kvm_memslots *slots;
4359 struct kvm_memory_slot *memslot;
4362 unsigned long *buf, *p;
4363 struct kvm_vcpu *vcpu;
4365 mutex_lock(&kvm->slots_lock);
4368 if (log->slot >= KVM_USER_MEM_SLOTS)
4371 slots = kvm_memslots(kvm);
4372 memslot = id_to_memslot(slots, log->slot);
4374 if (!memslot->dirty_bitmap)
4378 * Use second half of bitmap area because both HPT and radix
4379 * accumulate bits in the first half.
4381 n = kvm_dirty_bitmap_bytes(memslot);
4382 buf = memslot->dirty_bitmap + n / sizeof(long);
4385 if (kvm_is_radix(kvm))
4386 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4388 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4393 * We accumulate dirty bits in the first half of the
4394 * memslot's dirty_bitmap area, for when pages are paged
4395 * out or modified by the host directly. Pick up these
4396 * bits and add them to the map.
4398 p = memslot->dirty_bitmap;
4399 for (i = 0; i < n / sizeof(long); ++i)
4400 buf[i] |= xchg(&p[i], 0);
4402 /* Harvest dirty bits from VPA and DTL updates */
4403 /* Note: we never modify the SLB shadow buffer areas */
4404 kvm_for_each_vcpu(i, vcpu, kvm) {
4405 spin_lock(&vcpu->arch.vpa_update_lock);
4406 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4407 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4408 spin_unlock(&vcpu->arch.vpa_update_lock);
4412 if (copy_to_user(log->dirty_bitmap, buf, n))
4417 mutex_unlock(&kvm->slots_lock);
4421 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4422 struct kvm_memory_slot *dont)
4424 if (!dont || free->arch.rmap != dont->arch.rmap) {
4425 vfree(free->arch.rmap);
4426 free->arch.rmap = NULL;
4430 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4431 unsigned long npages)
4433 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4434 if (!slot->arch.rmap)
4440 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4441 struct kvm_memory_slot *memslot,
4442 const struct kvm_userspace_memory_region *mem)
4447 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4448 const struct kvm_userspace_memory_region *mem,
4449 const struct kvm_memory_slot *old,
4450 const struct kvm_memory_slot *new,
4451 enum kvm_mr_change change)
4453 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4456 * If we are making a new memslot, it might make
4457 * some address that was previously cached as emulated
4458 * MMIO be no longer emulated MMIO, so invalidate
4459 * all the caches of emulated MMIO translations.
4462 atomic64_inc(&kvm->arch.mmio_update);
4465 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4466 * have already called kvm_arch_flush_shadow_memslot() to
4467 * flush shadow mappings. For KVM_MR_CREATE we have no
4468 * previous mappings. So the only case to handle is
4469 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4471 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4472 * to get rid of any THP PTEs in the partition-scoped page tables
4473 * so we can track dirtiness at the page level; we flush when
4474 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4477 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4478 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4479 kvmppc_radix_flush_memslot(kvm, old);
4483 * Update LPCR values in kvm->arch and in vcores.
4484 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4485 * of kvm->arch.lpcr update).
4487 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4492 if ((kvm->arch.lpcr & mask) == lpcr)
4495 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4497 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4498 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4501 spin_lock(&vc->lock);
4502 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4503 spin_unlock(&vc->lock);
4504 if (++cores_done >= kvm->arch.online_vcores)
4509 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4514 void kvmppc_setup_partition_table(struct kvm *kvm)
4516 unsigned long dw0, dw1;
4518 if (!kvm_is_radix(kvm)) {
4519 /* PS field - page size for VRMA */
4520 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4521 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4522 /* HTABSIZE and HTABORG fields */
4523 dw0 |= kvm->arch.sdr1;
4525 /* Second dword as set by userspace */
4526 dw1 = kvm->arch.process_table;
4528 dw0 = PATB_HR | radix__get_tree_size() |
4529 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4530 dw1 = PATB_GR | kvm->arch.process_table;
4532 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4536 * Set up HPT (hashed page table) and RMA (real-mode area).
4537 * Must be called with kvm->arch.mmu_setup_lock held.
4539 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4542 struct kvm *kvm = vcpu->kvm;
4544 struct kvm_memory_slot *memslot;
4545 struct vm_area_struct *vma;
4546 unsigned long lpcr = 0, senc;
4547 unsigned long psize, porder;
4550 /* Allocate hashed page table (if not done already) and reset it */
4551 if (!kvm->arch.hpt.virt) {
4552 int order = KVM_DEFAULT_HPT_ORDER;
4553 struct kvm_hpt_info info;
4555 err = kvmppc_allocate_hpt(&info, order);
4556 /* If we get here, it means userspace didn't specify a
4557 * size explicitly. So, try successively smaller
4558 * sizes if the default failed. */
4559 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4560 err = kvmppc_allocate_hpt(&info, order);
4563 pr_err("KVM: Couldn't alloc HPT\n");
4567 kvmppc_set_hpt(kvm, &info);
4570 /* Look up the memslot for guest physical address 0 */
4571 srcu_idx = srcu_read_lock(&kvm->srcu);
4572 memslot = gfn_to_memslot(kvm, 0);
4574 /* We must have some memory at 0 by now */
4576 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4579 /* Look up the VMA for the start of this memory slot */
4580 hva = memslot->userspace_addr;
4581 down_read(¤t->mm->mmap_sem);
4582 vma = find_vma(current->mm, hva);
4583 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4586 psize = vma_kernel_pagesize(vma);
4588 up_read(¤t->mm->mmap_sem);
4590 /* We can handle 4k, 64k or 16M pages in the VRMA */
4591 if (psize >= 0x1000000)
4593 else if (psize >= 0x10000)
4597 porder = __ilog2(psize);
4599 senc = slb_pgsize_encoding(psize);
4600 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4601 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4602 /* Create HPTEs in the hash page table for the VRMA */
4603 kvmppc_map_vrma(vcpu, memslot, porder);
4605 /* Update VRMASD field in the LPCR */
4606 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4607 /* the -4 is to account for senc values starting at 0x10 */
4608 lpcr = senc << (LPCR_VRMASD_SH - 4);
4609 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4612 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4616 srcu_read_unlock(&kvm->srcu, srcu_idx);
4621 up_read(¤t->mm->mmap_sem);
4626 * Must be called with kvm->arch.mmu_setup_lock held and
4627 * mmu_ready = 0 and no vcpus running.
4629 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4631 if (nesting_enabled(kvm))
4632 kvmhv_release_all_nested(kvm);
4633 kvmppc_rmap_reset(kvm);
4634 kvm->arch.process_table = 0;
4635 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4636 spin_lock(&kvm->mmu_lock);
4637 kvm->arch.radix = 0;
4638 spin_unlock(&kvm->mmu_lock);
4639 kvmppc_free_radix(kvm);
4640 kvmppc_update_lpcr(kvm, LPCR_VPM1,
4641 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4646 * Must be called with kvm->arch.mmu_setup_lock held and
4647 * mmu_ready = 0 and no vcpus running.
4649 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4653 err = kvmppc_init_vm_radix(kvm);
4656 kvmppc_rmap_reset(kvm);
4657 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4658 spin_lock(&kvm->mmu_lock);
4659 kvm->arch.radix = 1;
4660 spin_unlock(&kvm->mmu_lock);
4661 kvmppc_free_hpt(&kvm->arch.hpt);
4662 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4663 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4667 #ifdef CONFIG_KVM_XICS
4669 * Allocate a per-core structure for managing state about which cores are
4670 * running in the host versus the guest and for exchanging data between
4671 * real mode KVM and CPU running in the host.
4672 * This is only done for the first VM.
4673 * The allocated structure stays even if all VMs have stopped.
4674 * It is only freed when the kvm-hv module is unloaded.
4675 * It's OK for this routine to fail, we just don't support host
4676 * core operations like redirecting H_IPI wakeups.
4678 void kvmppc_alloc_host_rm_ops(void)
4680 struct kvmppc_host_rm_ops *ops;
4681 unsigned long l_ops;
4685 /* Not the first time here ? */
4686 if (kvmppc_host_rm_ops_hv != NULL)
4689 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4693 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4694 ops->rm_core = kzalloc(size, GFP_KERNEL);
4696 if (!ops->rm_core) {
4703 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4704 if (!cpu_online(cpu))
4707 core = cpu >> threads_shift;
4708 ops->rm_core[core].rm_state.in_host = 1;
4711 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4714 * Make the contents of the kvmppc_host_rm_ops structure visible
4715 * to other CPUs before we assign it to the global variable.
4716 * Do an atomic assignment (no locks used here), but if someone
4717 * beats us to it, just free our copy and return.
4720 l_ops = (unsigned long) ops;
4722 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4724 kfree(ops->rm_core);
4729 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4730 "ppc/kvm_book3s:prepare",
4731 kvmppc_set_host_core,
4732 kvmppc_clear_host_core);
4736 void kvmppc_free_host_rm_ops(void)
4738 if (kvmppc_host_rm_ops_hv) {
4739 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4740 kfree(kvmppc_host_rm_ops_hv->rm_core);
4741 kfree(kvmppc_host_rm_ops_hv);
4742 kvmppc_host_rm_ops_hv = NULL;
4747 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4749 unsigned long lpcr, lpid;
4753 mutex_init(&kvm->arch.mmu_setup_lock);
4755 /* Allocate the guest's logical partition ID */
4757 lpid = kvmppc_alloc_lpid();
4760 kvm->arch.lpid = lpid;
4762 kvmppc_alloc_host_rm_ops();
4764 kvmhv_vm_nested_init(kvm);
4767 * Since we don't flush the TLB when tearing down a VM,
4768 * and this lpid might have previously been used,
4769 * make sure we flush on each core before running the new VM.
4770 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4771 * does this flush for us.
4773 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4774 cpumask_setall(&kvm->arch.need_tlb_flush);
4776 /* Start out with the default set of hcalls enabled */
4777 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4778 sizeof(kvm->arch.enabled_hcalls));
4780 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4781 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4783 /* Init LPCR for virtual RMA mode */
4784 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4785 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4786 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4787 lpcr &= LPCR_PECE | LPCR_LPES;
4791 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4792 LPCR_VPM0 | LPCR_VPM1;
4793 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4794 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4795 /* On POWER8 turn on online bit to enable PURR/SPURR */
4796 if (cpu_has_feature(CPU_FTR_ARCH_207S))
4799 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4800 * Set HVICE bit to enable hypervisor virtualization interrupts.
4801 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4802 * be unnecessary but better safe than sorry in case we re-enable
4803 * EE in HV mode with this LPCR still set)
4805 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4807 lpcr |= LPCR_HVICE | LPCR_HEIC;
4810 * If xive is enabled, we route 0x500 interrupts directly
4818 * If the host uses radix, the guest starts out as radix.
4820 if (radix_enabled()) {
4821 kvm->arch.radix = 1;
4822 kvm->arch.mmu_ready = 1;
4824 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4825 ret = kvmppc_init_vm_radix(kvm);
4827 kvmppc_free_lpid(kvm->arch.lpid);
4830 kvmppc_setup_partition_table(kvm);
4833 kvm->arch.lpcr = lpcr;
4835 /* Initialization for future HPT resizes */
4836 kvm->arch.resize_hpt = NULL;
4839 * Work out how many sets the TLB has, for the use of
4840 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4842 if (radix_enabled())
4843 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
4844 else if (cpu_has_feature(CPU_FTR_ARCH_300))
4845 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
4846 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4847 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
4849 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
4852 * Track that we now have a HV mode VM active. This blocks secondary
4853 * CPU threads from coming online.
4854 * On POWER9, we only need to do this if the "indep_threads_mode"
4855 * module parameter has been set to N.
4857 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4858 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4859 pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4860 kvm->arch.threads_indep = true;
4862 kvm->arch.threads_indep = indep_threads_mode;
4865 if (!kvm->arch.threads_indep)
4866 kvm_hv_vm_activated();
4869 * Initialize smt_mode depending on processor.
4870 * POWER8 and earlier have to use "strict" threading, where
4871 * all vCPUs in a vcore have to run on the same (sub)core,
4872 * whereas on POWER9 the threads can each run a different
4875 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4876 kvm->arch.smt_mode = threads_per_subcore;
4878 kvm->arch.smt_mode = 1;
4879 kvm->arch.emul_smt_mode = 1;
4882 * Create a debugfs directory for the VM
4884 snprintf(buf, sizeof(buf), "vm%d", current->pid);
4885 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4886 kvmppc_mmu_debugfs_init(kvm);
4887 if (radix_enabled())
4888 kvmhv_radix_debugfs_init(kvm);
4893 static void kvmppc_free_vcores(struct kvm *kvm)
4897 for (i = 0; i < KVM_MAX_VCORES; ++i)
4898 kfree(kvm->arch.vcores[i]);
4899 kvm->arch.online_vcores = 0;
4902 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4904 debugfs_remove_recursive(kvm->arch.debugfs_dir);
4906 if (!kvm->arch.threads_indep)
4907 kvm_hv_vm_deactivated();
4909 kvmppc_free_vcores(kvm);
4912 if (kvm_is_radix(kvm))
4913 kvmppc_free_radix(kvm);
4915 kvmppc_free_hpt(&kvm->arch.hpt);
4917 /* Perform global invalidation and return lpid to the pool */
4918 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4919 if (nesting_enabled(kvm))
4920 kvmhv_release_all_nested(kvm);
4921 kvm->arch.process_table = 0;
4922 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4924 kvmppc_free_lpid(kvm->arch.lpid);
4926 kvmppc_free_pimap(kvm);
4929 /* We don't need to emulate any privileged instructions or dcbz */
4930 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4931 unsigned int inst, int *advance)
4933 return EMULATE_FAIL;
4936 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4939 return EMULATE_FAIL;
4942 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4945 return EMULATE_FAIL;
4948 static int kvmppc_core_check_processor_compat_hv(void)
4950 if (cpu_has_feature(CPU_FTR_HVMODE) &&
4951 cpu_has_feature(CPU_FTR_ARCH_206))
4954 /* POWER9 in radix mode is capable of being a nested hypervisor. */
4955 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4961 #ifdef CONFIG_KVM_XICS
4963 void kvmppc_free_pimap(struct kvm *kvm)
4965 kfree(kvm->arch.pimap);
4968 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4970 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4973 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4975 struct irq_desc *desc;
4976 struct kvmppc_irq_map *irq_map;
4977 struct kvmppc_passthru_irqmap *pimap;
4978 struct irq_chip *chip;
4981 if (!kvm_irq_bypass)
4984 desc = irq_to_desc(host_irq);
4988 mutex_lock(&kvm->lock);
4990 pimap = kvm->arch.pimap;
4991 if (pimap == NULL) {
4992 /* First call, allocate structure to hold IRQ map */
4993 pimap = kvmppc_alloc_pimap();
4994 if (pimap == NULL) {
4995 mutex_unlock(&kvm->lock);
4998 kvm->arch.pimap = pimap;
5002 * For now, we only support interrupts for which the EOI operation
5003 * is an OPAL call followed by a write to XIRR, since that's
5004 * what our real-mode EOI code does, or a XIVE interrupt
5006 chip = irq_data_get_irq_chip(&desc->irq_data);
5007 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5008 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5009 host_irq, guest_gsi);
5010 mutex_unlock(&kvm->lock);
5015 * See if we already have an entry for this guest IRQ number.
5016 * If it's mapped to a hardware IRQ number, that's an error,
5017 * otherwise re-use this entry.
5019 for (i = 0; i < pimap->n_mapped; i++) {
5020 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5021 if (pimap->mapped[i].r_hwirq) {
5022 mutex_unlock(&kvm->lock);
5029 if (i == KVMPPC_PIRQ_MAPPED) {
5030 mutex_unlock(&kvm->lock);
5031 return -EAGAIN; /* table is full */
5034 irq_map = &pimap->mapped[i];
5036 irq_map->v_hwirq = guest_gsi;
5037 irq_map->desc = desc;
5040 * Order the above two stores before the next to serialize with
5041 * the KVM real mode handler.
5044 irq_map->r_hwirq = desc->irq_data.hwirq;
5046 if (i == pimap->n_mapped)
5050 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5052 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5054 irq_map->r_hwirq = 0;
5056 mutex_unlock(&kvm->lock);
5061 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5063 struct irq_desc *desc;
5064 struct kvmppc_passthru_irqmap *pimap;
5067 if (!kvm_irq_bypass)
5070 desc = irq_to_desc(host_irq);
5074 mutex_lock(&kvm->lock);
5075 if (!kvm->arch.pimap)
5078 pimap = kvm->arch.pimap;
5080 for (i = 0; i < pimap->n_mapped; i++) {
5081 if (guest_gsi == pimap->mapped[i].v_hwirq)
5085 if (i == pimap->n_mapped) {
5086 mutex_unlock(&kvm->lock);
5091 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5093 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5095 /* invalidate the entry (what do do on error from the above ?) */
5096 pimap->mapped[i].r_hwirq = 0;
5099 * We don't free this structure even when the count goes to
5100 * zero. The structure is freed when we destroy the VM.
5103 mutex_unlock(&kvm->lock);
5107 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5108 struct irq_bypass_producer *prod)
5111 struct kvm_kernel_irqfd *irqfd =
5112 container_of(cons, struct kvm_kernel_irqfd, consumer);
5114 irqfd->producer = prod;
5116 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5118 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5119 prod->irq, irqfd->gsi, ret);
5124 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5125 struct irq_bypass_producer *prod)
5128 struct kvm_kernel_irqfd *irqfd =
5129 container_of(cons, struct kvm_kernel_irqfd, consumer);
5131 irqfd->producer = NULL;
5134 * When producer of consumer is unregistered, we change back to
5135 * default external interrupt handling mode - KVM real mode
5136 * will switch back to host.
5138 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5140 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5141 prod->irq, irqfd->gsi, ret);
5145 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5146 unsigned int ioctl, unsigned long arg)
5148 struct kvm *kvm __maybe_unused = filp->private_data;
5149 void __user *argp = (void __user *)arg;
5154 case KVM_PPC_ALLOCATE_HTAB: {
5158 if (get_user(htab_order, (u32 __user *)argp))
5160 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5167 case KVM_PPC_GET_HTAB_FD: {
5168 struct kvm_get_htab_fd ghf;
5171 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5173 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5177 case KVM_PPC_RESIZE_HPT_PREPARE: {
5178 struct kvm_ppc_resize_hpt rhpt;
5181 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5184 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5188 case KVM_PPC_RESIZE_HPT_COMMIT: {
5189 struct kvm_ppc_resize_hpt rhpt;
5192 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5195 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5207 * List of hcall numbers to enable by default.
5208 * For compatibility with old userspace, we enable by default
5209 * all hcalls that were implemented before the hcall-enabling
5210 * facility was added. Note this list should not include H_RTAS.
5212 static unsigned int default_hcall_list[] = {
5226 #ifdef CONFIG_KVM_XICS
5237 static void init_default_hcalls(void)
5242 for (i = 0; default_hcall_list[i]; ++i) {
5243 hcall = default_hcall_list[i];
5244 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5245 __set_bit(hcall / 4, default_enabled_hcalls);
5249 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5255 /* If not on a POWER9, reject it */
5256 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5259 /* If any unknown flags set, reject it */
5260 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5263 /* GR (guest radix) bit in process_table field must match */
5264 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5265 if (!!(cfg->process_table & PATB_GR) != radix)
5268 /* Process table size field must be reasonable, i.e. <= 24 */
5269 if ((cfg->process_table & PRTS_MASK) > 24)
5272 /* We can change a guest to/from radix now, if the host is radix */
5273 if (radix && !radix_enabled())
5276 /* If we're a nested hypervisor, we currently only support radix */
5277 if (kvmhv_on_pseries() && !radix)
5280 mutex_lock(&kvm->arch.mmu_setup_lock);
5281 if (radix != kvm_is_radix(kvm)) {
5282 if (kvm->arch.mmu_ready) {
5283 kvm->arch.mmu_ready = 0;
5284 /* order mmu_ready vs. vcpus_running */
5286 if (atomic_read(&kvm->arch.vcpus_running)) {
5287 kvm->arch.mmu_ready = 1;
5293 err = kvmppc_switch_mmu_to_radix(kvm);
5295 err = kvmppc_switch_mmu_to_hpt(kvm);
5300 kvm->arch.process_table = cfg->process_table;
5301 kvmppc_setup_partition_table(kvm);
5303 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5304 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5308 mutex_unlock(&kvm->arch.mmu_setup_lock);
5312 static int kvmhv_enable_nested(struct kvm *kvm)
5316 if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5319 /* kvm == NULL means the caller is testing if the capability exists */
5321 kvm->arch.nested_enable = true;
5325 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5330 if (kvmhv_vcpu_is_radix(vcpu)) {
5331 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5337 /* For now quadrants are the only way to access nested guest memory */
5338 if (rc && vcpu->arch.nested)
5344 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5349 if (kvmhv_vcpu_is_radix(vcpu)) {
5350 rc = kvmhv_copy_to_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 struct kvmppc_ops kvm_ops_hv = {
5364 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5365 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5366 .get_one_reg = kvmppc_get_one_reg_hv,
5367 .set_one_reg = kvmppc_set_one_reg_hv,
5368 .vcpu_load = kvmppc_core_vcpu_load_hv,
5369 .vcpu_put = kvmppc_core_vcpu_put_hv,
5370 .set_msr = kvmppc_set_msr_hv,
5371 .vcpu_run = kvmppc_vcpu_run_hv,
5372 .vcpu_create = kvmppc_core_vcpu_create_hv,
5373 .vcpu_free = kvmppc_core_vcpu_free_hv,
5374 .check_requests = kvmppc_core_check_requests_hv,
5375 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
5376 .flush_memslot = kvmppc_core_flush_memslot_hv,
5377 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5378 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
5379 .unmap_hva_range = kvm_unmap_hva_range_hv,
5380 .age_hva = kvm_age_hva_hv,
5381 .test_age_hva = kvm_test_age_hva_hv,
5382 .set_spte_hva = kvm_set_spte_hva_hv,
5383 .mmu_destroy = kvmppc_mmu_destroy_hv,
5384 .free_memslot = kvmppc_core_free_memslot_hv,
5385 .create_memslot = kvmppc_core_create_memslot_hv,
5386 .init_vm = kvmppc_core_init_vm_hv,
5387 .destroy_vm = kvmppc_core_destroy_vm_hv,
5388 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5389 .emulate_op = kvmppc_core_emulate_op_hv,
5390 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5391 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5392 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5393 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
5394 .hcall_implemented = kvmppc_hcall_impl_hv,
5395 #ifdef CONFIG_KVM_XICS
5396 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5397 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5399 .configure_mmu = kvmhv_configure_mmu,
5400 .get_rmmu_info = kvmhv_get_rmmu_info,
5401 .set_smt_mode = kvmhv_set_smt_mode,
5402 .enable_nested = kvmhv_enable_nested,
5403 .load_from_eaddr = kvmhv_load_from_eaddr,
5404 .store_to_eaddr = kvmhv_store_to_eaddr,
5407 static int kvm_init_subcore_bitmap(void)
5410 int nr_cores = cpu_nr_cores();
5411 struct sibling_subcore_state *sibling_subcore_state;
5413 for (i = 0; i < nr_cores; i++) {
5414 int first_cpu = i * threads_per_core;
5415 int node = cpu_to_node(first_cpu);
5417 /* Ignore if it is already allocated. */
5418 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5421 sibling_subcore_state =
5422 kzalloc_node(sizeof(struct sibling_subcore_state),
5424 if (!sibling_subcore_state)
5428 for (j = 0; j < threads_per_core; j++) {
5429 int cpu = first_cpu + j;
5431 paca_ptrs[cpu]->sibling_subcore_state =
5432 sibling_subcore_state;
5438 static int kvmppc_radix_possible(void)
5440 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5443 static int kvmppc_book3s_init_hv(void)
5447 * FIXME!! Do we need to check on all cpus ?
5449 r = kvmppc_core_check_processor_compat_hv();
5453 r = kvmhv_nested_init();
5457 r = kvm_init_subcore_bitmap();
5462 * We need a way of accessing the XICS interrupt controller,
5463 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5464 * indirectly, via OPAL.
5467 if (!xics_on_xive() && !kvmhv_on_pseries() &&
5468 !local_paca->kvm_hstate.xics_phys) {
5469 struct device_node *np;
5471 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5473 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5476 /* presence of intc confirmed - node can be dropped again */
5481 kvm_ops_hv.owner = THIS_MODULE;
5482 kvmppc_hv_ops = &kvm_ops_hv;
5484 init_default_hcalls();
5488 r = kvmppc_mmu_hv_init();
5492 if (kvmppc_radix_possible())
5493 r = kvmppc_radix_init();
5496 * POWER9 chips before version 2.02 can't have some threads in
5497 * HPT mode and some in radix mode on the same core.
5499 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5500 unsigned int pvr = mfspr(SPRN_PVR);
5501 if ((pvr >> 16) == PVR_POWER9 &&
5502 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5503 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5504 no_mixing_hpt_and_radix = true;
5510 static void kvmppc_book3s_exit_hv(void)
5512 kvmppc_free_host_rm_ops();
5513 if (kvmppc_radix_possible())
5514 kvmppc_radix_exit();
5515 kvmppc_hv_ops = NULL;
5516 kvmhv_nested_exit();
5519 module_init(kvmppc_book3s_init_hv);
5520 module_exit(kvmppc_book3s_exit_hv);
5521 MODULE_LICENSE("GPL");
5522 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5523 MODULE_ALIAS("devname:kvm");