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>
80 #define CREATE_TRACE_POINTS
83 /* #define EXIT_DEBUG */
84 /* #define EXIT_DEBUG_SIMPLE */
85 /* #define EXIT_DEBUG_INT */
87 /* Used to indicate that a guest page fault needs to be handled */
88 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
89 /* Used to indicate that a guest passthrough interrupt needs to be handled */
90 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
92 /* Used as a "null" value for timebase values */
93 #define TB_NIL (~(u64)0)
95 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
97 static int dynamic_mt_modes = 6;
98 module_param(dynamic_mt_modes, int, 0644);
99 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
100 static int target_smt_mode;
101 module_param(target_smt_mode, int, 0644);
102 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
104 static bool indep_threads_mode = true;
105 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
106 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
108 static bool one_vm_per_core;
109 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
110 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
112 #ifdef CONFIG_KVM_XICS
113 static struct kernel_param_ops module_param_ops = {
114 .set = param_set_int,
115 .get = param_get_int,
118 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
119 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
121 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
122 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
125 /* If set, guests are allowed to create and control nested guests */
126 static bool nested = true;
127 module_param(nested, bool, S_IRUGO | S_IWUSR);
128 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
130 static inline bool nesting_enabled(struct kvm *kvm)
132 return kvm->arch.nested_enable && kvm_is_radix(kvm);
135 /* If set, the threads on each CPU core have to be in the same MMU mode */
136 static bool no_mixing_hpt_and_radix;
138 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
139 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
142 * RWMR values for POWER8. These control the rate at which PURR
143 * and SPURR count and should be set according to the number of
144 * online threads in the vcore being run.
146 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
147 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
148 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
149 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
150 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
151 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
152 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
153 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
155 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
167 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
171 struct kvm_vcpu *vcpu;
173 while (++i < MAX_SMT_THREADS) {
174 vcpu = READ_ONCE(vc->runnable_threads[i]);
183 /* Used to traverse the list of runnable threads for a given vcore */
184 #define for_each_runnable_thread(i, vcpu, vc) \
185 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
187 static bool kvmppc_ipi_thread(int cpu)
189 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
191 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
192 if (kvmhv_on_pseries())
195 /* On POWER9 we can use msgsnd to IPI any cpu */
196 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
197 msg |= get_hard_smp_processor_id(cpu);
199 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
203 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
204 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
206 if (cpu_first_thread_sibling(cpu) ==
207 cpu_first_thread_sibling(smp_processor_id())) {
208 msg |= cpu_thread_in_core(cpu);
210 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
217 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
218 if (cpu >= 0 && cpu < nr_cpu_ids) {
219 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
223 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
231 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
234 struct swait_queue_head *wqp;
236 wqp = kvm_arch_vcpu_wq(vcpu);
237 if (swq_has_sleeper(wqp)) {
239 ++vcpu->stat.halt_wakeup;
242 cpu = READ_ONCE(vcpu->arch.thread_cpu);
243 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
246 /* CPU points to the first thread of the core */
248 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
249 smp_send_reschedule(cpu);
253 * We use the vcpu_load/put functions to measure stolen time.
254 * Stolen time is counted as time when either the vcpu is able to
255 * run as part of a virtual core, but the task running the vcore
256 * is preempted or sleeping, or when the vcpu needs something done
257 * in the kernel by the task running the vcpu, but that task is
258 * preempted or sleeping. Those two things have to be counted
259 * separately, since one of the vcpu tasks will take on the job
260 * of running the core, and the other vcpu tasks in the vcore will
261 * sleep waiting for it to do that, but that sleep shouldn't count
264 * Hence we accumulate stolen time when the vcpu can run as part of
265 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
266 * needs its task to do other things in the kernel (for example,
267 * service a page fault) in busy_stolen. We don't accumulate
268 * stolen time for a vcore when it is inactive, or for a vcpu
269 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
270 * a misnomer; it means that the vcpu task is not executing in
271 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
272 * the kernel. We don't have any way of dividing up that time
273 * between time that the vcpu is genuinely stopped, time that
274 * the task is actively working on behalf of the vcpu, and time
275 * that the task is preempted, so we don't count any of it as
278 * Updates to busy_stolen are protected by arch.tbacct_lock;
279 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
280 * lock. The stolen times are measured in units of timebase ticks.
281 * (Note that the != TB_NIL checks below are purely defensive;
282 * they should never fail.)
285 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
289 spin_lock_irqsave(&vc->stoltb_lock, flags);
290 vc->preempt_tb = mftb();
291 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
294 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
298 spin_lock_irqsave(&vc->stoltb_lock, flags);
299 if (vc->preempt_tb != TB_NIL) {
300 vc->stolen_tb += mftb() - vc->preempt_tb;
301 vc->preempt_tb = TB_NIL;
303 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
306 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
308 struct kvmppc_vcore *vc = vcpu->arch.vcore;
312 * We can test vc->runner without taking the vcore lock,
313 * because only this task ever sets vc->runner to this
314 * vcpu, and once it is set to this vcpu, only this task
315 * ever sets it to NULL.
317 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
318 kvmppc_core_end_stolen(vc);
320 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
321 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
322 vcpu->arch.busy_preempt != TB_NIL) {
323 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
324 vcpu->arch.busy_preempt = TB_NIL;
326 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
329 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
331 struct kvmppc_vcore *vc = vcpu->arch.vcore;
334 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
335 kvmppc_core_start_stolen(vc);
337 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
338 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
339 vcpu->arch.busy_preempt = mftb();
340 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
343 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
346 * Check for illegal transactional state bit combination
347 * and if we find it, force the TS field to a safe state.
349 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
351 vcpu->arch.shregs.msr = msr;
352 kvmppc_end_cede(vcpu);
355 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
357 vcpu->arch.pvr = pvr;
360 /* Dummy value used in computing PCR value below */
361 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
363 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
365 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
366 struct kvmppc_vcore *vc = vcpu->arch.vcore;
368 /* We can (emulate) our own architecture version and anything older */
369 if (cpu_has_feature(CPU_FTR_ARCH_300))
370 host_pcr_bit = PCR_ARCH_300;
371 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
372 host_pcr_bit = PCR_ARCH_207;
373 else if (cpu_has_feature(CPU_FTR_ARCH_206))
374 host_pcr_bit = PCR_ARCH_206;
376 host_pcr_bit = PCR_ARCH_205;
378 /* Determine lowest PCR bit needed to run guest in given PVR level */
379 guest_pcr_bit = host_pcr_bit;
381 switch (arch_compat) {
383 guest_pcr_bit = PCR_ARCH_205;
387 guest_pcr_bit = PCR_ARCH_206;
390 guest_pcr_bit = PCR_ARCH_207;
393 guest_pcr_bit = PCR_ARCH_300;
400 /* Check requested PCR bits don't exceed our capabilities */
401 if (guest_pcr_bit > host_pcr_bit)
404 spin_lock(&vc->lock);
405 vc->arch_compat = arch_compat;
406 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
407 vc->pcr = host_pcr_bit - guest_pcr_bit;
408 spin_unlock(&vc->lock);
413 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
417 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
418 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
419 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
420 for (r = 0; r < 16; ++r)
421 pr_err("r%2d = %.16lx r%d = %.16lx\n",
422 r, kvmppc_get_gpr(vcpu, r),
423 r+16, kvmppc_get_gpr(vcpu, r+16));
424 pr_err("ctr = %.16lx lr = %.16lx\n",
425 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
426 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
427 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
428 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
429 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
430 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
431 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
432 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
433 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
434 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
435 pr_err("fault dar = %.16lx dsisr = %.8x\n",
436 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
437 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
438 for (r = 0; r < vcpu->arch.slb_max; ++r)
439 pr_err(" ESID = %.16llx VSID = %.16llx\n",
440 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
441 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
442 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
443 vcpu->arch.last_inst);
446 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
448 struct kvm_vcpu *ret;
450 mutex_lock(&kvm->lock);
451 ret = kvm_get_vcpu_by_id(kvm, id);
452 mutex_unlock(&kvm->lock);
456 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
458 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
459 vpa->yield_count = cpu_to_be32(1);
462 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
463 unsigned long addr, unsigned long len)
465 /* check address is cacheline aligned */
466 if (addr & (L1_CACHE_BYTES - 1))
468 spin_lock(&vcpu->arch.vpa_update_lock);
469 if (v->next_gpa != addr || v->len != len) {
471 v->len = addr ? len : 0;
472 v->update_pending = 1;
474 spin_unlock(&vcpu->arch.vpa_update_lock);
478 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
487 static int vpa_is_registered(struct kvmppc_vpa *vpap)
489 if (vpap->update_pending)
490 return vpap->next_gpa != 0;
491 return vpap->pinned_addr != NULL;
494 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
496 unsigned long vcpuid, unsigned long vpa)
498 struct kvm *kvm = vcpu->kvm;
499 unsigned long len, nb;
501 struct kvm_vcpu *tvcpu;
504 struct kvmppc_vpa *vpap;
506 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
510 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
511 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
512 subfunc == H_VPA_REG_SLB) {
513 /* Registering new area - address must be cache-line aligned */
514 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
517 /* convert logical addr to kernel addr and read length */
518 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
521 if (subfunc == H_VPA_REG_VPA)
522 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
524 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
525 kvmppc_unpin_guest_page(kvm, va, vpa, false);
528 if (len > nb || len < sizeof(struct reg_vpa))
537 spin_lock(&tvcpu->arch.vpa_update_lock);
540 case H_VPA_REG_VPA: /* register VPA */
542 * The size of our lppaca is 1kB because of the way we align
543 * it for the guest to avoid crossing a 4kB boundary. We only
544 * use 640 bytes of the structure though, so we should accept
545 * clients that set a size of 640.
547 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
548 if (len < sizeof(struct lppaca))
550 vpap = &tvcpu->arch.vpa;
554 case H_VPA_REG_DTL: /* register DTL */
555 if (len < sizeof(struct dtl_entry))
557 len -= len % sizeof(struct dtl_entry);
559 /* Check that they have previously registered a VPA */
561 if (!vpa_is_registered(&tvcpu->arch.vpa))
564 vpap = &tvcpu->arch.dtl;
568 case H_VPA_REG_SLB: /* register SLB shadow buffer */
569 /* Check that they have previously registered a VPA */
571 if (!vpa_is_registered(&tvcpu->arch.vpa))
574 vpap = &tvcpu->arch.slb_shadow;
578 case H_VPA_DEREG_VPA: /* deregister VPA */
579 /* Check they don't still have a DTL or SLB buf registered */
581 if (vpa_is_registered(&tvcpu->arch.dtl) ||
582 vpa_is_registered(&tvcpu->arch.slb_shadow))
585 vpap = &tvcpu->arch.vpa;
589 case H_VPA_DEREG_DTL: /* deregister DTL */
590 vpap = &tvcpu->arch.dtl;
594 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
595 vpap = &tvcpu->arch.slb_shadow;
601 vpap->next_gpa = vpa;
603 vpap->update_pending = 1;
606 spin_unlock(&tvcpu->arch.vpa_update_lock);
611 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
613 struct kvm *kvm = vcpu->kvm;
619 * We need to pin the page pointed to by vpap->next_gpa,
620 * but we can't call kvmppc_pin_guest_page under the lock
621 * as it does get_user_pages() and down_read(). So we
622 * have to drop the lock, pin the page, then get the lock
623 * again and check that a new area didn't get registered
627 gpa = vpap->next_gpa;
628 spin_unlock(&vcpu->arch.vpa_update_lock);
632 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
633 spin_lock(&vcpu->arch.vpa_update_lock);
634 if (gpa == vpap->next_gpa)
636 /* sigh... unpin that one and try again */
638 kvmppc_unpin_guest_page(kvm, va, gpa, false);
641 vpap->update_pending = 0;
642 if (va && nb < vpap->len) {
644 * If it's now too short, it must be that userspace
645 * has changed the mappings underlying guest memory,
646 * so unregister the region.
648 kvmppc_unpin_guest_page(kvm, va, gpa, false);
651 if (vpap->pinned_addr)
652 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
655 vpap->pinned_addr = va;
658 vpap->pinned_end = va + vpap->len;
661 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
663 if (!(vcpu->arch.vpa.update_pending ||
664 vcpu->arch.slb_shadow.update_pending ||
665 vcpu->arch.dtl.update_pending))
668 spin_lock(&vcpu->arch.vpa_update_lock);
669 if (vcpu->arch.vpa.update_pending) {
670 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
671 if (vcpu->arch.vpa.pinned_addr)
672 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
674 if (vcpu->arch.dtl.update_pending) {
675 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
676 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
677 vcpu->arch.dtl_index = 0;
679 if (vcpu->arch.slb_shadow.update_pending)
680 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
681 spin_unlock(&vcpu->arch.vpa_update_lock);
685 * Return the accumulated stolen time for the vcore up until `now'.
686 * The caller should hold the vcore lock.
688 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
693 spin_lock_irqsave(&vc->stoltb_lock, flags);
695 if (vc->vcore_state != VCORE_INACTIVE &&
696 vc->preempt_tb != TB_NIL)
697 p += now - vc->preempt_tb;
698 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
702 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
703 struct kvmppc_vcore *vc)
705 struct dtl_entry *dt;
707 unsigned long stolen;
708 unsigned long core_stolen;
712 dt = vcpu->arch.dtl_ptr;
713 vpa = vcpu->arch.vpa.pinned_addr;
715 core_stolen = vcore_stolen_time(vc, now);
716 stolen = core_stolen - vcpu->arch.stolen_logged;
717 vcpu->arch.stolen_logged = core_stolen;
718 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
719 stolen += vcpu->arch.busy_stolen;
720 vcpu->arch.busy_stolen = 0;
721 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
724 memset(dt, 0, sizeof(struct dtl_entry));
725 dt->dispatch_reason = 7;
726 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
727 dt->timebase = cpu_to_be64(now + vc->tb_offset);
728 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
729 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
730 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
732 if (dt == vcpu->arch.dtl.pinned_end)
733 dt = vcpu->arch.dtl.pinned_addr;
734 vcpu->arch.dtl_ptr = dt;
735 /* order writing *dt vs. writing vpa->dtl_idx */
737 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
738 vcpu->arch.dtl.dirty = true;
741 /* See if there is a doorbell interrupt pending for a vcpu */
742 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
745 struct kvmppc_vcore *vc;
747 if (vcpu->arch.doorbell_request)
750 * Ensure that the read of vcore->dpdes comes after the read
751 * of vcpu->doorbell_request. This barrier matches the
752 * smb_wmb() in kvmppc_guest_entry_inject().
755 vc = vcpu->arch.vcore;
756 thr = vcpu->vcpu_id - vc->first_vcpuid;
757 return !!(vc->dpdes & (1 << thr));
760 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
762 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
764 if ((!vcpu->arch.vcore->arch_compat) &&
765 cpu_has_feature(CPU_FTR_ARCH_207S))
770 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
771 unsigned long resource, unsigned long value1,
772 unsigned long value2)
775 case H_SET_MODE_RESOURCE_SET_CIABR:
776 if (!kvmppc_power8_compatible(vcpu))
781 return H_UNSUPPORTED_FLAG_START;
782 /* Guests can't breakpoint the hypervisor */
783 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
785 vcpu->arch.ciabr = value1;
787 case H_SET_MODE_RESOURCE_SET_DAWR:
788 if (!kvmppc_power8_compatible(vcpu))
790 if (!ppc_breakpoint_available())
793 return H_UNSUPPORTED_FLAG_START;
794 if (value2 & DABRX_HYP)
796 vcpu->arch.dawr = value1;
797 vcpu->arch.dawrx = value2;
804 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
806 struct kvmppc_vcore *vcore = target->arch.vcore;
809 * We expect to have been called by the real mode handler
810 * (kvmppc_rm_h_confer()) which would have directly returned
811 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
812 * have useful work to do and should not confer) so we don't
816 spin_lock(&vcore->lock);
817 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
818 vcore->vcore_state != VCORE_INACTIVE &&
820 target = vcore->runner;
821 spin_unlock(&vcore->lock);
823 return kvm_vcpu_yield_to(target);
826 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
829 struct lppaca *lppaca;
831 spin_lock(&vcpu->arch.vpa_update_lock);
832 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
834 yield_count = be32_to_cpu(lppaca->yield_count);
835 spin_unlock(&vcpu->arch.vpa_update_lock);
839 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
841 unsigned long req = kvmppc_get_gpr(vcpu, 3);
842 unsigned long target, ret = H_SUCCESS;
844 struct kvm_vcpu *tvcpu;
847 if (req <= MAX_HCALL_OPCODE &&
848 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
855 target = kvmppc_get_gpr(vcpu, 4);
856 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
861 tvcpu->arch.prodded = 1;
863 if (tvcpu->arch.ceded)
864 kvmppc_fast_vcpu_kick_hv(tvcpu);
867 target = kvmppc_get_gpr(vcpu, 4);
870 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
875 yield_count = kvmppc_get_gpr(vcpu, 5);
876 if (kvmppc_get_yield_count(tvcpu) != yield_count)
878 kvm_arch_vcpu_yield_to(tvcpu);
881 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
882 kvmppc_get_gpr(vcpu, 5),
883 kvmppc_get_gpr(vcpu, 6));
886 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
889 idx = srcu_read_lock(&vcpu->kvm->srcu);
890 rc = kvmppc_rtas_hcall(vcpu);
891 srcu_read_unlock(&vcpu->kvm->srcu, idx);
898 /* Send the error out to userspace via KVM_RUN */
900 case H_LOGICAL_CI_LOAD:
901 ret = kvmppc_h_logical_ci_load(vcpu);
902 if (ret == H_TOO_HARD)
905 case H_LOGICAL_CI_STORE:
906 ret = kvmppc_h_logical_ci_store(vcpu);
907 if (ret == H_TOO_HARD)
911 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
912 kvmppc_get_gpr(vcpu, 5),
913 kvmppc_get_gpr(vcpu, 6),
914 kvmppc_get_gpr(vcpu, 7));
915 if (ret == H_TOO_HARD)
924 if (kvmppc_xics_enabled(vcpu)) {
925 if (xics_on_xive()) {
926 ret = H_NOT_AVAILABLE;
929 ret = kvmppc_xics_hcall(vcpu, req);
934 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
937 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
938 kvmppc_get_gpr(vcpu, 5));
940 #ifdef CONFIG_SPAPR_TCE_IOMMU
942 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
943 kvmppc_get_gpr(vcpu, 5));
944 if (ret == H_TOO_HARD)
948 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
949 kvmppc_get_gpr(vcpu, 5),
950 kvmppc_get_gpr(vcpu, 6));
951 if (ret == H_TOO_HARD)
954 case H_PUT_TCE_INDIRECT:
955 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
956 kvmppc_get_gpr(vcpu, 5),
957 kvmppc_get_gpr(vcpu, 6),
958 kvmppc_get_gpr(vcpu, 7));
959 if (ret == H_TOO_HARD)
963 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
964 kvmppc_get_gpr(vcpu, 5),
965 kvmppc_get_gpr(vcpu, 6),
966 kvmppc_get_gpr(vcpu, 7));
967 if (ret == H_TOO_HARD)
972 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
976 case H_SET_PARTITION_TABLE:
978 if (nesting_enabled(vcpu->kvm))
979 ret = kvmhv_set_partition_table(vcpu);
983 if (!nesting_enabled(vcpu->kvm))
985 ret = kvmhv_enter_nested_guest(vcpu);
986 if (ret == H_INTERRUPT) {
987 kvmppc_set_gpr(vcpu, 3, 0);
988 vcpu->arch.hcall_needed = 0;
990 } else if (ret == H_TOO_HARD) {
991 kvmppc_set_gpr(vcpu, 3, 0);
992 vcpu->arch.hcall_needed = 0;
996 case H_TLB_INVALIDATE:
998 if (nesting_enabled(vcpu->kvm))
999 ret = kvmhv_do_nested_tlbie(vcpu);
1001 case H_COPY_TOFROM_GUEST:
1003 if (nesting_enabled(vcpu->kvm))
1004 ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1009 kvmppc_set_gpr(vcpu, 3, ret);
1010 vcpu->arch.hcall_needed = 0;
1011 return RESUME_GUEST;
1015 * Handle H_CEDE in the nested virtualization case where we haven't
1016 * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1017 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1018 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1020 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1022 vcpu->arch.shregs.msr |= MSR_EE;
1023 vcpu->arch.ceded = 1;
1025 if (vcpu->arch.prodded) {
1026 vcpu->arch.prodded = 0;
1028 vcpu->arch.ceded = 0;
1032 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1038 case H_REGISTER_VPA:
1040 case H_LOGICAL_CI_LOAD:
1041 case H_LOGICAL_CI_STORE:
1042 #ifdef CONFIG_KVM_XICS
1053 /* See if it's in the real-mode table */
1054 return kvmppc_hcall_impl_hv_realmode(cmd);
1057 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1058 struct kvm_vcpu *vcpu)
1062 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1065 * Fetch failed, so return to guest and
1066 * try executing it again.
1068 return RESUME_GUEST;
1071 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1072 run->exit_reason = KVM_EXIT_DEBUG;
1073 run->debug.arch.address = kvmppc_get_pc(vcpu);
1076 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1077 return RESUME_GUEST;
1081 static void do_nothing(void *x)
1085 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1087 int thr, cpu, pcpu, nthreads;
1089 unsigned long dpdes;
1091 nthreads = vcpu->kvm->arch.emul_smt_mode;
1093 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1094 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1095 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1099 * If the vcpu is currently running on a physical cpu thread,
1100 * interrupt it in order to pull it out of the guest briefly,
1101 * which will update its vcore->dpdes value.
1103 pcpu = READ_ONCE(v->cpu);
1105 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1106 if (kvmppc_doorbell_pending(v))
1113 * On POWER9, emulate doorbell-related instructions in order to
1114 * give the guest the illusion of running on a multi-threaded core.
1115 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1118 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1122 struct kvm *kvm = vcpu->kvm;
1123 struct kvm_vcpu *tvcpu;
1125 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1126 return RESUME_GUEST;
1127 if (get_op(inst) != 31)
1128 return EMULATE_FAIL;
1130 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1131 switch (get_xop(inst)) {
1132 case OP_31_XOP_MSGSNDP:
1133 arg = kvmppc_get_gpr(vcpu, rb);
1134 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1137 if (arg >= kvm->arch.emul_smt_mode)
1139 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1142 if (!tvcpu->arch.doorbell_request) {
1143 tvcpu->arch.doorbell_request = 1;
1144 kvmppc_fast_vcpu_kick_hv(tvcpu);
1147 case OP_31_XOP_MSGCLRP:
1148 arg = kvmppc_get_gpr(vcpu, rb);
1149 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1151 vcpu->arch.vcore->dpdes = 0;
1152 vcpu->arch.doorbell_request = 0;
1154 case OP_31_XOP_MFSPR:
1155 switch (get_sprn(inst)) {
1160 arg = kvmppc_read_dpdes(vcpu);
1163 return EMULATE_FAIL;
1165 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1168 return EMULATE_FAIL;
1170 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1171 return RESUME_GUEST;
1174 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1175 struct task_struct *tsk)
1177 int r = RESUME_HOST;
1179 vcpu->stat.sum_exits++;
1182 * This can happen if an interrupt occurs in the last stages
1183 * of guest entry or the first stages of guest exit (i.e. after
1184 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1185 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1186 * That can happen due to a bug, or due to a machine check
1187 * occurring at just the wrong time.
1189 if (vcpu->arch.shregs.msr & MSR_HV) {
1190 printk(KERN_EMERG "KVM trap in HV mode!\n");
1191 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1192 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1193 vcpu->arch.shregs.msr);
1194 kvmppc_dump_regs(vcpu);
1195 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1196 run->hw.hardware_exit_reason = vcpu->arch.trap;
1199 run->exit_reason = KVM_EXIT_UNKNOWN;
1200 run->ready_for_interrupt_injection = 1;
1201 switch (vcpu->arch.trap) {
1202 /* We're good on these - the host merely wanted to get our attention */
1203 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1204 vcpu->stat.dec_exits++;
1207 case BOOK3S_INTERRUPT_EXTERNAL:
1208 case BOOK3S_INTERRUPT_H_DOORBELL:
1209 case BOOK3S_INTERRUPT_H_VIRT:
1210 vcpu->stat.ext_intr_exits++;
1213 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1214 case BOOK3S_INTERRUPT_HMI:
1215 case BOOK3S_INTERRUPT_PERFMON:
1216 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1219 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1220 /* Print the MCE event to host console. */
1221 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1224 * If the guest can do FWNMI, exit to userspace so it can
1225 * deliver a FWNMI to the guest.
1226 * Otherwise we synthesize a machine check for the guest
1227 * so that it knows that the machine check occurred.
1229 if (!vcpu->kvm->arch.fwnmi_enabled) {
1230 ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1231 kvmppc_core_queue_machine_check(vcpu, flags);
1236 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1237 run->exit_reason = KVM_EXIT_NMI;
1238 run->hw.hardware_exit_reason = vcpu->arch.trap;
1239 /* Clear out the old NMI status from run->flags */
1240 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1241 /* Now set the NMI status */
1242 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1243 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1245 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1249 case BOOK3S_INTERRUPT_PROGRAM:
1253 * Normally program interrupts are delivered directly
1254 * to the guest by the hardware, but we can get here
1255 * as a result of a hypervisor emulation interrupt
1256 * (e40) getting turned into a 700 by BML RTAS.
1258 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1259 kvmppc_core_queue_program(vcpu, flags);
1263 case BOOK3S_INTERRUPT_SYSCALL:
1265 /* hcall - punt to userspace */
1268 /* hypercall with MSR_PR has already been handled in rmode,
1269 * and never reaches here.
1272 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1273 for (i = 0; i < 9; ++i)
1274 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1275 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1276 vcpu->arch.hcall_needed = 1;
1281 * We get these next two if the guest accesses a page which it thinks
1282 * it has mapped but which is not actually present, either because
1283 * it is for an emulated I/O device or because the corresonding
1284 * host page has been paged out. Any other HDSI/HISI interrupts
1285 * have been handled already.
1287 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1288 r = RESUME_PAGE_FAULT;
1290 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1291 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1292 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1293 DSISR_SRR1_MATCH_64S;
1294 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1295 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1296 r = RESUME_PAGE_FAULT;
1299 * This occurs if the guest executes an illegal instruction.
1300 * If the guest debug is disabled, generate a program interrupt
1301 * to the guest. If guest debug is enabled, we need to check
1302 * whether the instruction is a software breakpoint instruction.
1303 * Accordingly return to Guest or Host.
1305 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1306 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1307 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1308 swab32(vcpu->arch.emul_inst) :
1309 vcpu->arch.emul_inst;
1310 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1311 r = kvmppc_emulate_debug_inst(run, vcpu);
1313 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1318 * This occurs if the guest (kernel or userspace), does something that
1319 * is prohibited by HFSCR.
1320 * On POWER9, this could be a doorbell instruction that we need
1322 * Otherwise, we just generate a program interrupt to the guest.
1324 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1326 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1327 cpu_has_feature(CPU_FTR_ARCH_300))
1328 r = kvmppc_emulate_doorbell_instr(vcpu);
1329 if (r == EMULATE_FAIL) {
1330 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1335 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1336 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1338 * This occurs for various TM-related instructions that
1339 * we need to emulate on POWER9 DD2.2. We have already
1340 * handled the cases where the guest was in real-suspend
1341 * mode and was transitioning to transactional state.
1343 r = kvmhv_p9_tm_emulation(vcpu);
1347 case BOOK3S_INTERRUPT_HV_RM_HARD:
1348 r = RESUME_PASSTHROUGH;
1351 kvmppc_dump_regs(vcpu);
1352 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1353 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1354 vcpu->arch.shregs.msr);
1355 run->hw.hardware_exit_reason = vcpu->arch.trap;
1363 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1368 vcpu->stat.sum_exits++;
1371 * This can happen if an interrupt occurs in the last stages
1372 * of guest entry or the first stages of guest exit (i.e. after
1373 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1374 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1375 * That can happen due to a bug, or due to a machine check
1376 * occurring at just the wrong time.
1378 if (vcpu->arch.shregs.msr & MSR_HV) {
1379 pr_emerg("KVM trap in HV mode while nested!\n");
1380 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1381 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1382 vcpu->arch.shregs.msr);
1383 kvmppc_dump_regs(vcpu);
1386 switch (vcpu->arch.trap) {
1387 /* We're good on these - the host merely wanted to get our attention */
1388 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1389 vcpu->stat.dec_exits++;
1392 case BOOK3S_INTERRUPT_EXTERNAL:
1393 vcpu->stat.ext_intr_exits++;
1396 case BOOK3S_INTERRUPT_H_DOORBELL:
1397 case BOOK3S_INTERRUPT_H_VIRT:
1398 vcpu->stat.ext_intr_exits++;
1401 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1402 case BOOK3S_INTERRUPT_HMI:
1403 case BOOK3S_INTERRUPT_PERFMON:
1404 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1407 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1408 /* Pass the machine check to the L1 guest */
1410 /* Print the MCE event to host console. */
1411 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1414 * We get these next two if the guest accesses a page which it thinks
1415 * it has mapped but which is not actually present, either because
1416 * it is for an emulated I/O device or because the corresonding
1417 * host page has been paged out.
1419 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1420 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1421 r = kvmhv_nested_page_fault(run, vcpu);
1422 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1424 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1425 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1426 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1427 DSISR_SRR1_MATCH_64S;
1428 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1429 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1430 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1431 r = kvmhv_nested_page_fault(run, vcpu);
1432 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1435 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1436 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1438 * This occurs for various TM-related instructions that
1439 * we need to emulate on POWER9 DD2.2. We have already
1440 * handled the cases where the guest was in real-suspend
1441 * mode and was transitioning to transactional state.
1443 r = kvmhv_p9_tm_emulation(vcpu);
1447 case BOOK3S_INTERRUPT_HV_RM_HARD:
1448 vcpu->arch.trap = 0;
1450 if (!xics_on_xive())
1451 kvmppc_xics_rm_complete(vcpu, 0);
1461 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1462 struct kvm_sregs *sregs)
1466 memset(sregs, 0, sizeof(struct kvm_sregs));
1467 sregs->pvr = vcpu->arch.pvr;
1468 for (i = 0; i < vcpu->arch.slb_max; i++) {
1469 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1470 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1476 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1477 struct kvm_sregs *sregs)
1481 /* Only accept the same PVR as the host's, since we can't spoof it */
1482 if (sregs->pvr != vcpu->arch.pvr)
1486 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1487 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1488 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1489 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1493 vcpu->arch.slb_max = j;
1498 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1499 bool preserve_top32)
1501 struct kvm *kvm = vcpu->kvm;
1502 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1505 mutex_lock(&kvm->lock);
1506 spin_lock(&vc->lock);
1508 * If ILE (interrupt little-endian) has changed, update the
1509 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1511 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1512 struct kvm_vcpu *vcpu;
1515 kvm_for_each_vcpu(i, vcpu, kvm) {
1516 if (vcpu->arch.vcore != vc)
1518 if (new_lpcr & LPCR_ILE)
1519 vcpu->arch.intr_msr |= MSR_LE;
1521 vcpu->arch.intr_msr &= ~MSR_LE;
1526 * Userspace can only modify DPFD (default prefetch depth),
1527 * ILE (interrupt little-endian) and TC (translation control).
1528 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1530 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1531 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1534 * On POWER9, allow userspace to enable large decrementer for the
1535 * guest, whether or not the host has it enabled.
1537 if (cpu_has_feature(CPU_FTR_ARCH_300))
1540 /* Broken 32-bit version of LPCR must not clear top bits */
1543 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1544 spin_unlock(&vc->lock);
1545 mutex_unlock(&kvm->lock);
1548 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1549 union kvmppc_one_reg *val)
1555 case KVM_REG_PPC_DEBUG_INST:
1556 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1558 case KVM_REG_PPC_HIOR:
1559 *val = get_reg_val(id, 0);
1561 case KVM_REG_PPC_DABR:
1562 *val = get_reg_val(id, vcpu->arch.dabr);
1564 case KVM_REG_PPC_DABRX:
1565 *val = get_reg_val(id, vcpu->arch.dabrx);
1567 case KVM_REG_PPC_DSCR:
1568 *val = get_reg_val(id, vcpu->arch.dscr);
1570 case KVM_REG_PPC_PURR:
1571 *val = get_reg_val(id, vcpu->arch.purr);
1573 case KVM_REG_PPC_SPURR:
1574 *val = get_reg_val(id, vcpu->arch.spurr);
1576 case KVM_REG_PPC_AMR:
1577 *val = get_reg_val(id, vcpu->arch.amr);
1579 case KVM_REG_PPC_UAMOR:
1580 *val = get_reg_val(id, vcpu->arch.uamor);
1582 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1583 i = id - KVM_REG_PPC_MMCR0;
1584 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1586 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1587 i = id - KVM_REG_PPC_PMC1;
1588 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1590 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1591 i = id - KVM_REG_PPC_SPMC1;
1592 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1594 case KVM_REG_PPC_SIAR:
1595 *val = get_reg_val(id, vcpu->arch.siar);
1597 case KVM_REG_PPC_SDAR:
1598 *val = get_reg_val(id, vcpu->arch.sdar);
1600 case KVM_REG_PPC_SIER:
1601 *val = get_reg_val(id, vcpu->arch.sier);
1603 case KVM_REG_PPC_IAMR:
1604 *val = get_reg_val(id, vcpu->arch.iamr);
1606 case KVM_REG_PPC_PSPB:
1607 *val = get_reg_val(id, vcpu->arch.pspb);
1609 case KVM_REG_PPC_DPDES:
1610 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1612 case KVM_REG_PPC_VTB:
1613 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1615 case KVM_REG_PPC_DAWR:
1616 *val = get_reg_val(id, vcpu->arch.dawr);
1618 case KVM_REG_PPC_DAWRX:
1619 *val = get_reg_val(id, vcpu->arch.dawrx);
1621 case KVM_REG_PPC_CIABR:
1622 *val = get_reg_val(id, vcpu->arch.ciabr);
1624 case KVM_REG_PPC_CSIGR:
1625 *val = get_reg_val(id, vcpu->arch.csigr);
1627 case KVM_REG_PPC_TACR:
1628 *val = get_reg_val(id, vcpu->arch.tacr);
1630 case KVM_REG_PPC_TCSCR:
1631 *val = get_reg_val(id, vcpu->arch.tcscr);
1633 case KVM_REG_PPC_PID:
1634 *val = get_reg_val(id, vcpu->arch.pid);
1636 case KVM_REG_PPC_ACOP:
1637 *val = get_reg_val(id, vcpu->arch.acop);
1639 case KVM_REG_PPC_WORT:
1640 *val = get_reg_val(id, vcpu->arch.wort);
1642 case KVM_REG_PPC_TIDR:
1643 *val = get_reg_val(id, vcpu->arch.tid);
1645 case KVM_REG_PPC_PSSCR:
1646 *val = get_reg_val(id, vcpu->arch.psscr);
1648 case KVM_REG_PPC_VPA_ADDR:
1649 spin_lock(&vcpu->arch.vpa_update_lock);
1650 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1651 spin_unlock(&vcpu->arch.vpa_update_lock);
1653 case KVM_REG_PPC_VPA_SLB:
1654 spin_lock(&vcpu->arch.vpa_update_lock);
1655 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1656 val->vpaval.length = vcpu->arch.slb_shadow.len;
1657 spin_unlock(&vcpu->arch.vpa_update_lock);
1659 case KVM_REG_PPC_VPA_DTL:
1660 spin_lock(&vcpu->arch.vpa_update_lock);
1661 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1662 val->vpaval.length = vcpu->arch.dtl.len;
1663 spin_unlock(&vcpu->arch.vpa_update_lock);
1665 case KVM_REG_PPC_TB_OFFSET:
1666 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1668 case KVM_REG_PPC_LPCR:
1669 case KVM_REG_PPC_LPCR_64:
1670 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1672 case KVM_REG_PPC_PPR:
1673 *val = get_reg_val(id, vcpu->arch.ppr);
1675 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1676 case KVM_REG_PPC_TFHAR:
1677 *val = get_reg_val(id, vcpu->arch.tfhar);
1679 case KVM_REG_PPC_TFIAR:
1680 *val = get_reg_val(id, vcpu->arch.tfiar);
1682 case KVM_REG_PPC_TEXASR:
1683 *val = get_reg_val(id, vcpu->arch.texasr);
1685 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1686 i = id - KVM_REG_PPC_TM_GPR0;
1687 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1689 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1692 i = id - KVM_REG_PPC_TM_VSR0;
1694 for (j = 0; j < TS_FPRWIDTH; j++)
1695 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1697 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1698 val->vval = vcpu->arch.vr_tm.vr[i-32];
1704 case KVM_REG_PPC_TM_CR:
1705 *val = get_reg_val(id, vcpu->arch.cr_tm);
1707 case KVM_REG_PPC_TM_XER:
1708 *val = get_reg_val(id, vcpu->arch.xer_tm);
1710 case KVM_REG_PPC_TM_LR:
1711 *val = get_reg_val(id, vcpu->arch.lr_tm);
1713 case KVM_REG_PPC_TM_CTR:
1714 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1716 case KVM_REG_PPC_TM_FPSCR:
1717 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1719 case KVM_REG_PPC_TM_AMR:
1720 *val = get_reg_val(id, vcpu->arch.amr_tm);
1722 case KVM_REG_PPC_TM_PPR:
1723 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1725 case KVM_REG_PPC_TM_VRSAVE:
1726 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1728 case KVM_REG_PPC_TM_VSCR:
1729 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1730 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1734 case KVM_REG_PPC_TM_DSCR:
1735 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1737 case KVM_REG_PPC_TM_TAR:
1738 *val = get_reg_val(id, vcpu->arch.tar_tm);
1741 case KVM_REG_PPC_ARCH_COMPAT:
1742 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1744 case KVM_REG_PPC_DEC_EXPIRY:
1745 *val = get_reg_val(id, vcpu->arch.dec_expires +
1746 vcpu->arch.vcore->tb_offset);
1748 case KVM_REG_PPC_ONLINE:
1749 *val = get_reg_val(id, vcpu->arch.online);
1751 case KVM_REG_PPC_PTCR:
1752 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1762 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1763 union kvmppc_one_reg *val)
1767 unsigned long addr, len;
1770 case KVM_REG_PPC_HIOR:
1771 /* Only allow this to be set to zero */
1772 if (set_reg_val(id, *val))
1775 case KVM_REG_PPC_DABR:
1776 vcpu->arch.dabr = set_reg_val(id, *val);
1778 case KVM_REG_PPC_DABRX:
1779 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1781 case KVM_REG_PPC_DSCR:
1782 vcpu->arch.dscr = set_reg_val(id, *val);
1784 case KVM_REG_PPC_PURR:
1785 vcpu->arch.purr = set_reg_val(id, *val);
1787 case KVM_REG_PPC_SPURR:
1788 vcpu->arch.spurr = set_reg_val(id, *val);
1790 case KVM_REG_PPC_AMR:
1791 vcpu->arch.amr = set_reg_val(id, *val);
1793 case KVM_REG_PPC_UAMOR:
1794 vcpu->arch.uamor = set_reg_val(id, *val);
1796 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1797 i = id - KVM_REG_PPC_MMCR0;
1798 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1800 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1801 i = id - KVM_REG_PPC_PMC1;
1802 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1804 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1805 i = id - KVM_REG_PPC_SPMC1;
1806 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1808 case KVM_REG_PPC_SIAR:
1809 vcpu->arch.siar = set_reg_val(id, *val);
1811 case KVM_REG_PPC_SDAR:
1812 vcpu->arch.sdar = set_reg_val(id, *val);
1814 case KVM_REG_PPC_SIER:
1815 vcpu->arch.sier = set_reg_val(id, *val);
1817 case KVM_REG_PPC_IAMR:
1818 vcpu->arch.iamr = set_reg_val(id, *val);
1820 case KVM_REG_PPC_PSPB:
1821 vcpu->arch.pspb = set_reg_val(id, *val);
1823 case KVM_REG_PPC_DPDES:
1824 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1826 case KVM_REG_PPC_VTB:
1827 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1829 case KVM_REG_PPC_DAWR:
1830 vcpu->arch.dawr = set_reg_val(id, *val);
1832 case KVM_REG_PPC_DAWRX:
1833 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1835 case KVM_REG_PPC_CIABR:
1836 vcpu->arch.ciabr = set_reg_val(id, *val);
1837 /* Don't allow setting breakpoints in hypervisor code */
1838 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1839 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1841 case KVM_REG_PPC_CSIGR:
1842 vcpu->arch.csigr = set_reg_val(id, *val);
1844 case KVM_REG_PPC_TACR:
1845 vcpu->arch.tacr = set_reg_val(id, *val);
1847 case KVM_REG_PPC_TCSCR:
1848 vcpu->arch.tcscr = set_reg_val(id, *val);
1850 case KVM_REG_PPC_PID:
1851 vcpu->arch.pid = set_reg_val(id, *val);
1853 case KVM_REG_PPC_ACOP:
1854 vcpu->arch.acop = set_reg_val(id, *val);
1856 case KVM_REG_PPC_WORT:
1857 vcpu->arch.wort = set_reg_val(id, *val);
1859 case KVM_REG_PPC_TIDR:
1860 vcpu->arch.tid = set_reg_val(id, *val);
1862 case KVM_REG_PPC_PSSCR:
1863 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1865 case KVM_REG_PPC_VPA_ADDR:
1866 addr = set_reg_val(id, *val);
1868 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1869 vcpu->arch.dtl.next_gpa))
1871 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1873 case KVM_REG_PPC_VPA_SLB:
1874 addr = val->vpaval.addr;
1875 len = val->vpaval.length;
1877 if (addr && !vcpu->arch.vpa.next_gpa)
1879 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1881 case KVM_REG_PPC_VPA_DTL:
1882 addr = val->vpaval.addr;
1883 len = val->vpaval.length;
1885 if (addr && (len < sizeof(struct dtl_entry) ||
1886 !vcpu->arch.vpa.next_gpa))
1888 len -= len % sizeof(struct dtl_entry);
1889 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1891 case KVM_REG_PPC_TB_OFFSET:
1892 /* round up to multiple of 2^24 */
1893 vcpu->arch.vcore->tb_offset =
1894 ALIGN(set_reg_val(id, *val), 1UL << 24);
1896 case KVM_REG_PPC_LPCR:
1897 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1899 case KVM_REG_PPC_LPCR_64:
1900 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1902 case KVM_REG_PPC_PPR:
1903 vcpu->arch.ppr = set_reg_val(id, *val);
1905 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1906 case KVM_REG_PPC_TFHAR:
1907 vcpu->arch.tfhar = set_reg_val(id, *val);
1909 case KVM_REG_PPC_TFIAR:
1910 vcpu->arch.tfiar = set_reg_val(id, *val);
1912 case KVM_REG_PPC_TEXASR:
1913 vcpu->arch.texasr = set_reg_val(id, *val);
1915 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1916 i = id - KVM_REG_PPC_TM_GPR0;
1917 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1919 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1922 i = id - KVM_REG_PPC_TM_VSR0;
1924 for (j = 0; j < TS_FPRWIDTH; j++)
1925 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1927 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1928 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1933 case KVM_REG_PPC_TM_CR:
1934 vcpu->arch.cr_tm = set_reg_val(id, *val);
1936 case KVM_REG_PPC_TM_XER:
1937 vcpu->arch.xer_tm = set_reg_val(id, *val);
1939 case KVM_REG_PPC_TM_LR:
1940 vcpu->arch.lr_tm = set_reg_val(id, *val);
1942 case KVM_REG_PPC_TM_CTR:
1943 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1945 case KVM_REG_PPC_TM_FPSCR:
1946 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1948 case KVM_REG_PPC_TM_AMR:
1949 vcpu->arch.amr_tm = set_reg_val(id, *val);
1951 case KVM_REG_PPC_TM_PPR:
1952 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1954 case KVM_REG_PPC_TM_VRSAVE:
1955 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1957 case KVM_REG_PPC_TM_VSCR:
1958 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1959 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1963 case KVM_REG_PPC_TM_DSCR:
1964 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1966 case KVM_REG_PPC_TM_TAR:
1967 vcpu->arch.tar_tm = set_reg_val(id, *val);
1970 case KVM_REG_PPC_ARCH_COMPAT:
1971 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1973 case KVM_REG_PPC_DEC_EXPIRY:
1974 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1975 vcpu->arch.vcore->tb_offset;
1977 case KVM_REG_PPC_ONLINE:
1978 i = set_reg_val(id, *val);
1979 if (i && !vcpu->arch.online)
1980 atomic_inc(&vcpu->arch.vcore->online_count);
1981 else if (!i && vcpu->arch.online)
1982 atomic_dec(&vcpu->arch.vcore->online_count);
1983 vcpu->arch.online = i;
1985 case KVM_REG_PPC_PTCR:
1986 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
1997 * On POWER9, threads are independent and can be in different partitions.
1998 * Therefore we consider each thread to be a subcore.
1999 * There is a restriction that all threads have to be in the same
2000 * MMU mode (radix or HPT), unfortunately, but since we only support
2001 * HPT guests on a HPT host so far, that isn't an impediment yet.
2003 static int threads_per_vcore(struct kvm *kvm)
2005 if (kvm->arch.threads_indep)
2007 return threads_per_subcore;
2010 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2012 struct kvmppc_vcore *vcore;
2014 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2019 spin_lock_init(&vcore->lock);
2020 spin_lock_init(&vcore->stoltb_lock);
2021 init_swait_queue_head(&vcore->wq);
2022 vcore->preempt_tb = TB_NIL;
2023 vcore->lpcr = kvm->arch.lpcr;
2024 vcore->first_vcpuid = id;
2026 INIT_LIST_HEAD(&vcore->preempt_list);
2031 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2032 static struct debugfs_timings_element {
2036 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2037 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2038 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2039 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2040 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2043 #define N_TIMINGS (ARRAY_SIZE(timings))
2045 struct debugfs_timings_state {
2046 struct kvm_vcpu *vcpu;
2047 unsigned int buflen;
2048 char buf[N_TIMINGS * 100];
2051 static int debugfs_timings_open(struct inode *inode, struct file *file)
2053 struct kvm_vcpu *vcpu = inode->i_private;
2054 struct debugfs_timings_state *p;
2056 p = kzalloc(sizeof(*p), GFP_KERNEL);
2060 kvm_get_kvm(vcpu->kvm);
2062 file->private_data = p;
2064 return nonseekable_open(inode, file);
2067 static int debugfs_timings_release(struct inode *inode, struct file *file)
2069 struct debugfs_timings_state *p = file->private_data;
2071 kvm_put_kvm(p->vcpu->kvm);
2076 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2077 size_t len, loff_t *ppos)
2079 struct debugfs_timings_state *p = file->private_data;
2080 struct kvm_vcpu *vcpu = p->vcpu;
2082 struct kvmhv_tb_accumulator tb;
2091 buf_end = s + sizeof(p->buf);
2092 for (i = 0; i < N_TIMINGS; ++i) {
2093 struct kvmhv_tb_accumulator *acc;
2095 acc = (struct kvmhv_tb_accumulator *)
2096 ((unsigned long)vcpu + timings[i].offset);
2098 for (loops = 0; loops < 1000; ++loops) {
2099 count = acc->seqcount;
2104 if (count == acc->seqcount) {
2112 snprintf(s, buf_end - s, "%s: stuck\n",
2115 snprintf(s, buf_end - s,
2116 "%s: %llu %llu %llu %llu\n",
2117 timings[i].name, count / 2,
2118 tb_to_ns(tb.tb_total),
2119 tb_to_ns(tb.tb_min),
2120 tb_to_ns(tb.tb_max));
2123 p->buflen = s - p->buf;
2127 if (pos >= p->buflen)
2129 if (len > p->buflen - pos)
2130 len = p->buflen - pos;
2131 n = copy_to_user(buf, p->buf + pos, len);
2141 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2142 size_t len, loff_t *ppos)
2147 static const struct file_operations debugfs_timings_ops = {
2148 .owner = THIS_MODULE,
2149 .open = debugfs_timings_open,
2150 .release = debugfs_timings_release,
2151 .read = debugfs_timings_read,
2152 .write = debugfs_timings_write,
2153 .llseek = generic_file_llseek,
2156 /* Create a debugfs directory for the vcpu */
2157 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2160 struct kvm *kvm = vcpu->kvm;
2162 snprintf(buf, sizeof(buf), "vcpu%u", id);
2163 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2165 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2166 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2168 vcpu->arch.debugfs_timings =
2169 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2170 vcpu, &debugfs_timings_ops);
2173 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2174 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2177 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2179 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2182 struct kvm_vcpu *vcpu;
2185 struct kvmppc_vcore *vcore;
2188 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2192 err = kvm_vcpu_init(vcpu, kvm, id);
2196 vcpu->arch.shared = &vcpu->arch.shregs;
2197 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2199 * The shared struct is never shared on HV,
2200 * so we can always use host endianness
2202 #ifdef __BIG_ENDIAN__
2203 vcpu->arch.shared_big_endian = true;
2205 vcpu->arch.shared_big_endian = false;
2208 vcpu->arch.mmcr[0] = MMCR0_FC;
2209 vcpu->arch.ctrl = CTRL_RUNLATCH;
2210 /* default to host PVR, since we can't spoof it */
2211 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2212 spin_lock_init(&vcpu->arch.vpa_update_lock);
2213 spin_lock_init(&vcpu->arch.tbacct_lock);
2214 vcpu->arch.busy_preempt = TB_NIL;
2215 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2218 * Set the default HFSCR for the guest from the host value.
2219 * This value is only used on POWER9.
2220 * On POWER9, we want to virtualize the doorbell facility, so we
2221 * don't set the HFSCR_MSGP bit, and that causes those instructions
2222 * to trap and then we emulate them.
2224 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2225 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2226 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2227 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2228 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2229 vcpu->arch.hfscr |= HFSCR_TM;
2231 if (cpu_has_feature(CPU_FTR_TM_COMP))
2232 vcpu->arch.hfscr |= HFSCR_TM;
2234 kvmppc_mmu_book3s_hv_init(vcpu);
2236 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2238 init_waitqueue_head(&vcpu->arch.cpu_run);
2240 mutex_lock(&kvm->lock);
2243 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2244 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2245 pr_devel("KVM: VCPU ID too high\n");
2246 core = KVM_MAX_VCORES;
2248 BUG_ON(kvm->arch.smt_mode != 1);
2249 core = kvmppc_pack_vcpu_id(kvm, id);
2252 core = id / kvm->arch.smt_mode;
2254 if (core < KVM_MAX_VCORES) {
2255 vcore = kvm->arch.vcores[core];
2256 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2257 pr_devel("KVM: collision on id %u", id);
2259 } else if (!vcore) {
2261 vcore = kvmppc_vcore_create(kvm,
2262 id & ~(kvm->arch.smt_mode - 1));
2263 kvm->arch.vcores[core] = vcore;
2264 kvm->arch.online_vcores++;
2267 mutex_unlock(&kvm->lock);
2272 spin_lock(&vcore->lock);
2273 ++vcore->num_threads;
2274 spin_unlock(&vcore->lock);
2275 vcpu->arch.vcore = vcore;
2276 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2277 vcpu->arch.thread_cpu = -1;
2278 vcpu->arch.prev_cpu = -1;
2280 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2281 kvmppc_sanity_check(vcpu);
2283 debugfs_vcpu_init(vcpu, id);
2288 kmem_cache_free(kvm_vcpu_cache, vcpu);
2290 return ERR_PTR(err);
2293 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2294 unsigned long flags)
2301 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2303 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2305 * On POWER8 (or POWER7), the threading mode is "strict",
2306 * so we pack smt_mode vcpus per vcore.
2308 if (smt_mode > threads_per_subcore)
2312 * On POWER9, the threading mode is "loose",
2313 * so each vcpu gets its own vcore.
2318 mutex_lock(&kvm->lock);
2320 if (!kvm->arch.online_vcores) {
2321 kvm->arch.smt_mode = smt_mode;
2322 kvm->arch.emul_smt_mode = esmt;
2325 mutex_unlock(&kvm->lock);
2330 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2332 if (vpa->pinned_addr)
2333 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2337 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2339 spin_lock(&vcpu->arch.vpa_update_lock);
2340 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2341 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2342 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2343 spin_unlock(&vcpu->arch.vpa_update_lock);
2344 kvm_vcpu_uninit(vcpu);
2345 kmem_cache_free(kvm_vcpu_cache, vcpu);
2348 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2350 /* Indicate we want to get back into the guest */
2354 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2356 unsigned long dec_nsec, now;
2359 if (now > vcpu->arch.dec_expires) {
2360 /* decrementer has already gone negative */
2361 kvmppc_core_queue_dec(vcpu);
2362 kvmppc_core_prepare_to_enter(vcpu);
2365 dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2366 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2367 vcpu->arch.timer_running = 1;
2370 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2372 vcpu->arch.ceded = 0;
2373 if (vcpu->arch.timer_running) {
2374 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2375 vcpu->arch.timer_running = 0;
2379 extern int __kvmppc_vcore_entry(void);
2381 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2382 struct kvm_vcpu *vcpu)
2386 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2388 spin_lock_irq(&vcpu->arch.tbacct_lock);
2390 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2391 vcpu->arch.stolen_logged;
2392 vcpu->arch.busy_preempt = now;
2393 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2394 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2396 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2399 static int kvmppc_grab_hwthread(int cpu)
2401 struct paca_struct *tpaca;
2402 long timeout = 10000;
2404 tpaca = paca_ptrs[cpu];
2406 /* Ensure the thread won't go into the kernel if it wakes */
2407 tpaca->kvm_hstate.kvm_vcpu = NULL;
2408 tpaca->kvm_hstate.kvm_vcore = NULL;
2409 tpaca->kvm_hstate.napping = 0;
2411 tpaca->kvm_hstate.hwthread_req = 1;
2414 * If the thread is already executing in the kernel (e.g. handling
2415 * a stray interrupt), wait for it to get back to nap mode.
2416 * The smp_mb() is to ensure that our setting of hwthread_req
2417 * is visible before we look at hwthread_state, so if this
2418 * races with the code at system_reset_pSeries and the thread
2419 * misses our setting of hwthread_req, we are sure to see its
2420 * setting of hwthread_state, and vice versa.
2423 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2424 if (--timeout <= 0) {
2425 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2433 static void kvmppc_release_hwthread(int cpu)
2435 struct paca_struct *tpaca;
2437 tpaca = paca_ptrs[cpu];
2438 tpaca->kvm_hstate.hwthread_req = 0;
2439 tpaca->kvm_hstate.kvm_vcpu = NULL;
2440 tpaca->kvm_hstate.kvm_vcore = NULL;
2441 tpaca->kvm_hstate.kvm_split_mode = NULL;
2444 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2446 struct kvm_nested_guest *nested = vcpu->arch.nested;
2447 cpumask_t *cpu_in_guest;
2450 cpu = cpu_first_thread_sibling(cpu);
2452 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2453 cpu_in_guest = &nested->cpu_in_guest;
2455 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2456 cpu_in_guest = &kvm->arch.cpu_in_guest;
2459 * Make sure setting of bit in need_tlb_flush precedes
2460 * testing of cpu_in_guest bits. The matching barrier on
2461 * the other side is the first smp_mb() in kvmppc_run_core().
2464 for (i = 0; i < threads_per_core; ++i)
2465 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2466 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2469 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2471 struct kvm_nested_guest *nested = vcpu->arch.nested;
2472 struct kvm *kvm = vcpu->kvm;
2475 if (!cpu_has_feature(CPU_FTR_HVMODE))
2479 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2481 prev_cpu = vcpu->arch.prev_cpu;
2484 * With radix, the guest can do TLB invalidations itself,
2485 * and it could choose to use the local form (tlbiel) if
2486 * it is invalidating a translation that has only ever been
2487 * used on one vcpu. However, that doesn't mean it has
2488 * only ever been used on one physical cpu, since vcpus
2489 * can move around between pcpus. To cope with this, when
2490 * a vcpu moves from one pcpu to another, we need to tell
2491 * any vcpus running on the same core as this vcpu previously
2492 * ran to flush the TLB. The TLB is shared between threads,
2493 * so we use a single bit in .need_tlb_flush for all 4 threads.
2495 if (prev_cpu != pcpu) {
2496 if (prev_cpu >= 0 &&
2497 cpu_first_thread_sibling(prev_cpu) !=
2498 cpu_first_thread_sibling(pcpu))
2499 radix_flush_cpu(kvm, prev_cpu, vcpu);
2501 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2503 vcpu->arch.prev_cpu = pcpu;
2507 static void kvmppc_radix_check_need_tlb_flush(struct kvm *kvm, int pcpu,
2508 struct kvm_nested_guest *nested)
2510 cpumask_t *need_tlb_flush;
2513 if (!cpu_has_feature(CPU_FTR_HVMODE))
2516 if (cpu_has_feature(CPU_FTR_ARCH_300))
2520 lpid = nested->shadow_lpid;
2521 need_tlb_flush = &nested->need_tlb_flush;
2523 lpid = kvm->arch.lpid;
2524 need_tlb_flush = &kvm->arch.need_tlb_flush;
2527 mtspr(SPRN_LPID, lpid);
2531 if (cpumask_test_cpu(pcpu, need_tlb_flush)) {
2532 radix__local_flush_tlb_lpid_guest(lpid);
2533 /* Clear the bit after the TLB flush */
2534 cpumask_clear_cpu(pcpu, need_tlb_flush);
2538 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2541 struct paca_struct *tpaca;
2542 struct kvm *kvm = vc->kvm;
2546 if (vcpu->arch.timer_running) {
2547 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2548 vcpu->arch.timer_running = 0;
2550 cpu += vcpu->arch.ptid;
2551 vcpu->cpu = vc->pcpu;
2552 vcpu->arch.thread_cpu = cpu;
2553 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2555 tpaca = paca_ptrs[cpu];
2556 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2557 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2558 tpaca->kvm_hstate.fake_suspend = 0;
2559 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2561 tpaca->kvm_hstate.kvm_vcore = vc;
2562 if (cpu != smp_processor_id())
2563 kvmppc_ipi_thread(cpu);
2566 static void kvmppc_wait_for_nap(int n_threads)
2568 int cpu = smp_processor_id();
2573 for (loops = 0; loops < 1000000; ++loops) {
2575 * Check if all threads are finished.
2576 * We set the vcore pointer when starting a thread
2577 * and the thread clears it when finished, so we look
2578 * for any threads that still have a non-NULL vcore ptr.
2580 for (i = 1; i < n_threads; ++i)
2581 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2583 if (i == n_threads) {
2590 for (i = 1; i < n_threads; ++i)
2591 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2592 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2596 * Check that we are on thread 0 and that any other threads in
2597 * this core are off-line. Then grab the threads so they can't
2600 static int on_primary_thread(void)
2602 int cpu = smp_processor_id();
2605 /* Are we on a primary subcore? */
2606 if (cpu_thread_in_subcore(cpu))
2610 while (++thr < threads_per_subcore)
2611 if (cpu_online(cpu + thr))
2614 /* Grab all hw threads so they can't go into the kernel */
2615 for (thr = 1; thr < threads_per_subcore; ++thr) {
2616 if (kvmppc_grab_hwthread(cpu + thr)) {
2617 /* Couldn't grab one; let the others go */
2619 kvmppc_release_hwthread(cpu + thr);
2620 } while (--thr > 0);
2628 * A list of virtual cores for each physical CPU.
2629 * These are vcores that could run but their runner VCPU tasks are
2630 * (or may be) preempted.
2632 struct preempted_vcore_list {
2633 struct list_head list;
2637 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2639 static void init_vcore_lists(void)
2643 for_each_possible_cpu(cpu) {
2644 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2645 spin_lock_init(&lp->lock);
2646 INIT_LIST_HEAD(&lp->list);
2650 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2652 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2654 vc->vcore_state = VCORE_PREEMPT;
2655 vc->pcpu = smp_processor_id();
2656 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2657 spin_lock(&lp->lock);
2658 list_add_tail(&vc->preempt_list, &lp->list);
2659 spin_unlock(&lp->lock);
2662 /* Start accumulating stolen time */
2663 kvmppc_core_start_stolen(vc);
2666 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2668 struct preempted_vcore_list *lp;
2670 kvmppc_core_end_stolen(vc);
2671 if (!list_empty(&vc->preempt_list)) {
2672 lp = &per_cpu(preempted_vcores, vc->pcpu);
2673 spin_lock(&lp->lock);
2674 list_del_init(&vc->preempt_list);
2675 spin_unlock(&lp->lock);
2677 vc->vcore_state = VCORE_INACTIVE;
2681 * This stores information about the virtual cores currently
2682 * assigned to a physical core.
2686 int max_subcore_threads;
2688 int subcore_threads[MAX_SUBCORES];
2689 struct kvmppc_vcore *vc[MAX_SUBCORES];
2693 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2694 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2696 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2698 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2700 memset(cip, 0, sizeof(*cip));
2701 cip->n_subcores = 1;
2702 cip->max_subcore_threads = vc->num_threads;
2703 cip->total_threads = vc->num_threads;
2704 cip->subcore_threads[0] = vc->num_threads;
2708 static bool subcore_config_ok(int n_subcores, int n_threads)
2711 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2712 * split-core mode, with one thread per subcore.
2714 if (cpu_has_feature(CPU_FTR_ARCH_300))
2715 return n_subcores <= 4 && n_threads == 1;
2717 /* On POWER8, can only dynamically split if unsplit to begin with */
2718 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2720 if (n_subcores > MAX_SUBCORES)
2722 if (n_subcores > 1) {
2723 if (!(dynamic_mt_modes & 2))
2725 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2729 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2732 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2734 vc->entry_exit_map = 0;
2736 vc->napping_threads = 0;
2737 vc->conferring_threads = 0;
2738 vc->tb_offset_applied = 0;
2741 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2743 int n_threads = vc->num_threads;
2746 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2749 /* In one_vm_per_core mode, require all vcores to be from the same vm */
2750 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2753 /* Some POWER9 chips require all threads to be in the same MMU mode */
2754 if (no_mixing_hpt_and_radix &&
2755 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2758 if (n_threads < cip->max_subcore_threads)
2759 n_threads = cip->max_subcore_threads;
2760 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2762 cip->max_subcore_threads = n_threads;
2764 sub = cip->n_subcores;
2766 cip->total_threads += vc->num_threads;
2767 cip->subcore_threads[sub] = vc->num_threads;
2769 init_vcore_to_run(vc);
2770 list_del_init(&vc->preempt_list);
2776 * Work out whether it is possible to piggyback the execution of
2777 * vcore *pvc onto the execution of the other vcores described in *cip.
2779 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2782 if (cip->total_threads + pvc->num_threads > target_threads)
2785 return can_dynamic_split(pvc, cip);
2788 static void prepare_threads(struct kvmppc_vcore *vc)
2791 struct kvm_vcpu *vcpu;
2793 for_each_runnable_thread(i, vcpu, vc) {
2794 if (signal_pending(vcpu->arch.run_task))
2795 vcpu->arch.ret = -EINTR;
2796 else if (vcpu->arch.vpa.update_pending ||
2797 vcpu->arch.slb_shadow.update_pending ||
2798 vcpu->arch.dtl.update_pending)
2799 vcpu->arch.ret = RESUME_GUEST;
2802 kvmppc_remove_runnable(vc, vcpu);
2803 wake_up(&vcpu->arch.cpu_run);
2807 static void collect_piggybacks(struct core_info *cip, int target_threads)
2809 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2810 struct kvmppc_vcore *pvc, *vcnext;
2812 spin_lock(&lp->lock);
2813 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2814 if (!spin_trylock(&pvc->lock))
2816 prepare_threads(pvc);
2817 if (!pvc->n_runnable) {
2818 list_del_init(&pvc->preempt_list);
2819 if (pvc->runner == NULL) {
2820 pvc->vcore_state = VCORE_INACTIVE;
2821 kvmppc_core_end_stolen(pvc);
2823 spin_unlock(&pvc->lock);
2826 if (!can_piggyback(pvc, cip, target_threads)) {
2827 spin_unlock(&pvc->lock);
2830 kvmppc_core_end_stolen(pvc);
2831 pvc->vcore_state = VCORE_PIGGYBACK;
2832 if (cip->total_threads >= target_threads)
2835 spin_unlock(&lp->lock);
2838 static bool recheck_signals(struct core_info *cip)
2841 struct kvm_vcpu *vcpu;
2843 for (sub = 0; sub < cip->n_subcores; ++sub)
2844 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2845 if (signal_pending(vcpu->arch.run_task))
2850 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2852 int still_running = 0, i;
2855 struct kvm_vcpu *vcpu;
2857 spin_lock(&vc->lock);
2859 for_each_runnable_thread(i, vcpu, vc) {
2861 * It's safe to unlock the vcore in the loop here, because
2862 * for_each_runnable_thread() is safe against removal of
2863 * the vcpu, and the vcore state is VCORE_EXITING here,
2864 * so any vcpus becoming runnable will have their arch.trap
2865 * set to zero and can't actually run in the guest.
2867 spin_unlock(&vc->lock);
2868 /* cancel pending dec exception if dec is positive */
2869 if (now < vcpu->arch.dec_expires &&
2870 kvmppc_core_pending_dec(vcpu))
2871 kvmppc_core_dequeue_dec(vcpu);
2873 trace_kvm_guest_exit(vcpu);
2876 if (vcpu->arch.trap)
2877 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2878 vcpu->arch.run_task);
2880 vcpu->arch.ret = ret;
2881 vcpu->arch.trap = 0;
2883 spin_lock(&vc->lock);
2884 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2885 if (vcpu->arch.pending_exceptions)
2886 kvmppc_core_prepare_to_enter(vcpu);
2887 if (vcpu->arch.ceded)
2888 kvmppc_set_timer(vcpu);
2892 kvmppc_remove_runnable(vc, vcpu);
2893 wake_up(&vcpu->arch.cpu_run);
2897 if (still_running > 0) {
2898 kvmppc_vcore_preempt(vc);
2899 } else if (vc->runner) {
2900 vc->vcore_state = VCORE_PREEMPT;
2901 kvmppc_core_start_stolen(vc);
2903 vc->vcore_state = VCORE_INACTIVE;
2905 if (vc->n_runnable > 0 && vc->runner == NULL) {
2906 /* make sure there's a candidate runner awake */
2908 vcpu = next_runnable_thread(vc, &i);
2909 wake_up(&vcpu->arch.cpu_run);
2912 spin_unlock(&vc->lock);
2916 * Clear core from the list of active host cores as we are about to
2917 * enter the guest. Only do this if it is the primary thread of the
2918 * core (not if a subcore) that is entering the guest.
2920 static inline int kvmppc_clear_host_core(unsigned int cpu)
2924 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2927 * Memory barrier can be omitted here as we will do a smp_wmb()
2928 * later in kvmppc_start_thread and we need ensure that state is
2929 * visible to other CPUs only after we enter guest.
2931 core = cpu >> threads_shift;
2932 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2937 * Advertise this core as an active host core since we exited the guest
2938 * Only need to do this if it is the primary thread of the core that is
2941 static inline int kvmppc_set_host_core(unsigned int cpu)
2945 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2949 * Memory barrier can be omitted here because we do a spin_unlock
2950 * immediately after this which provides the memory barrier.
2952 core = cpu >> threads_shift;
2953 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2957 static void set_irq_happened(int trap)
2960 case BOOK3S_INTERRUPT_EXTERNAL:
2961 local_paca->irq_happened |= PACA_IRQ_EE;
2963 case BOOK3S_INTERRUPT_H_DOORBELL:
2964 local_paca->irq_happened |= PACA_IRQ_DBELL;
2966 case BOOK3S_INTERRUPT_HMI:
2967 local_paca->irq_happened |= PACA_IRQ_HMI;
2969 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2970 replay_system_reset();
2976 * Run a set of guest threads on a physical core.
2977 * Called with vc->lock held.
2979 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2981 struct kvm_vcpu *vcpu;
2984 struct core_info core_info;
2985 struct kvmppc_vcore *pvc;
2986 struct kvm_split_mode split_info, *sip;
2987 int split, subcore_size, active;
2990 unsigned long cmd_bit, stat_bit;
2993 int controlled_threads;
2999 * Remove from the list any threads that have a signal pending
3000 * or need a VPA update done
3002 prepare_threads(vc);
3004 /* if the runner is no longer runnable, let the caller pick a new one */
3005 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3011 init_vcore_to_run(vc);
3012 vc->preempt_tb = TB_NIL;
3015 * Number of threads that we will be controlling: the same as
3016 * the number of threads per subcore, except on POWER9,
3017 * where it's 1 because the threads are (mostly) independent.
3019 controlled_threads = threads_per_vcore(vc->kvm);
3022 * Make sure we are running on primary threads, and that secondary
3023 * threads are offline. Also check if the number of threads in this
3024 * guest are greater than the current system threads per guest.
3025 * On POWER9, we need to be not in independent-threads mode if
3026 * this is a HPT guest on a radix host machine where the
3027 * CPU threads may not be in different MMU modes.
3029 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3030 !kvm_is_radix(vc->kvm);
3031 if (((controlled_threads > 1) &&
3032 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3033 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3034 for_each_runnable_thread(i, vcpu, vc) {
3035 vcpu->arch.ret = -EBUSY;
3036 kvmppc_remove_runnable(vc, vcpu);
3037 wake_up(&vcpu->arch.cpu_run);
3043 * See if we could run any other vcores on the physical core
3044 * along with this one.
3046 init_core_info(&core_info, vc);
3047 pcpu = smp_processor_id();
3048 target_threads = controlled_threads;
3049 if (target_smt_mode && target_smt_mode < target_threads)
3050 target_threads = target_smt_mode;
3051 if (vc->num_threads < target_threads)
3052 collect_piggybacks(&core_info, target_threads);
3055 * On radix, arrange for TLB flushing if necessary.
3056 * This has to be done before disabling interrupts since
3057 * it uses smp_call_function().
3059 pcpu = smp_processor_id();
3060 if (kvm_is_radix(vc->kvm)) {
3061 for (sub = 0; sub < core_info.n_subcores; ++sub)
3062 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3063 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3067 * Hard-disable interrupts, and check resched flag and signals.
3068 * If we need to reschedule or deliver a signal, clean up
3069 * and return without going into the guest(s).
3070 * If the mmu_ready flag has been cleared, don't go into the
3071 * guest because that means a HPT resize operation is in progress.
3073 local_irq_disable();
3075 if (lazy_irq_pending() || need_resched() ||
3076 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3078 vc->vcore_state = VCORE_INACTIVE;
3079 /* Unlock all except the primary vcore */
3080 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3081 pvc = core_info.vc[sub];
3082 /* Put back on to the preempted vcores list */
3083 kvmppc_vcore_preempt(pvc);
3084 spin_unlock(&pvc->lock);
3086 for (i = 0; i < controlled_threads; ++i)
3087 kvmppc_release_hwthread(pcpu + i);
3091 kvmppc_clear_host_core(pcpu);
3093 /* Decide on micro-threading (split-core) mode */
3094 subcore_size = threads_per_subcore;
3095 cmd_bit = stat_bit = 0;
3096 split = core_info.n_subcores;
3098 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3099 && !cpu_has_feature(CPU_FTR_ARCH_300);
3101 if (split > 1 || hpt_on_radix) {
3103 memset(&split_info, 0, sizeof(split_info));
3104 for (sub = 0; sub < core_info.n_subcores; ++sub)
3105 split_info.vc[sub] = core_info.vc[sub];
3108 if (split == 2 && (dynamic_mt_modes & 2)) {
3109 cmd_bit = HID0_POWER8_1TO2LPAR;
3110 stat_bit = HID0_POWER8_2LPARMODE;
3113 cmd_bit = HID0_POWER8_1TO4LPAR;
3114 stat_bit = HID0_POWER8_4LPARMODE;
3116 subcore_size = MAX_SMT_THREADS / split;
3117 split_info.rpr = mfspr(SPRN_RPR);
3118 split_info.pmmar = mfspr(SPRN_PMMAR);
3119 split_info.ldbar = mfspr(SPRN_LDBAR);
3120 split_info.subcore_size = subcore_size;
3122 split_info.subcore_size = 1;
3124 /* Use the split_info for LPCR/LPIDR changes */
3125 split_info.lpcr_req = vc->lpcr;
3126 split_info.lpidr_req = vc->kvm->arch.lpid;
3127 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3128 split_info.do_set = 1;
3132 /* order writes to split_info before kvm_split_mode pointer */
3136 for (thr = 0; thr < controlled_threads; ++thr) {
3137 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3139 paca->kvm_hstate.tid = thr;
3140 paca->kvm_hstate.napping = 0;
3141 paca->kvm_hstate.kvm_split_mode = sip;
3144 /* Initiate micro-threading (split-core) on POWER8 if required */
3146 unsigned long hid0 = mfspr(SPRN_HID0);
3148 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3150 mtspr(SPRN_HID0, hid0);
3153 hid0 = mfspr(SPRN_HID0);
3154 if (hid0 & stat_bit)
3161 * On POWER8, set RWMR register.
3162 * Since it only affects PURR and SPURR, it doesn't affect
3163 * the host, so we don't save/restore the host value.
3166 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3167 int n_online = atomic_read(&vc->online_count);
3170 * Use the 8-thread value if we're doing split-core
3171 * or if the vcore's online count looks bogus.
3173 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3174 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3175 rwmr_val = p8_rwmr_values[n_online];
3176 mtspr(SPRN_RWMR, rwmr_val);
3179 /* Start all the threads */
3181 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3182 thr = is_power8 ? subcore_thread_map[sub] : sub;
3185 pvc = core_info.vc[sub];
3186 pvc->pcpu = pcpu + thr;
3187 for_each_runnable_thread(i, vcpu, pvc) {
3188 kvmppc_start_thread(vcpu, pvc);
3189 kvmppc_create_dtl_entry(vcpu, pvc);
3190 trace_kvm_guest_enter(vcpu);
3191 if (!vcpu->arch.ptid)
3193 active |= 1 << (thr + vcpu->arch.ptid);
3196 * We need to start the first thread of each subcore
3197 * even if it doesn't have a vcpu.
3200 kvmppc_start_thread(NULL, pvc);
3204 * Ensure that split_info.do_nap is set after setting
3205 * the vcore pointer in the PACA of the secondaries.
3210 * When doing micro-threading, poke the inactive threads as well.
3211 * This gets them to the nap instruction after kvm_do_nap,
3212 * which reduces the time taken to unsplit later.
3213 * For POWER9 HPT guest on radix host, we need all the secondary
3214 * threads woken up so they can do the LPCR/LPIDR change.
3216 if (cmd_bit || hpt_on_radix) {
3217 split_info.do_nap = 1; /* ask secondaries to nap when done */
3218 for (thr = 1; thr < threads_per_subcore; ++thr)
3219 if (!(active & (1 << thr)))
3220 kvmppc_ipi_thread(pcpu + thr);
3223 vc->vcore_state = VCORE_RUNNING;
3226 trace_kvmppc_run_core(vc, 0);
3228 for (sub = 0; sub < core_info.n_subcores; ++sub)
3229 spin_unlock(&core_info.vc[sub]->lock);
3231 if (kvm_is_radix(vc->kvm)) {
3233 * Do we need to flush the process scoped TLB for the LPAR?
3235 * On POWER9, individual threads can come in here, but the
3236 * TLB is shared between the 4 threads in a core, hence
3237 * invalidating on one thread invalidates for all.
3238 * Thus we make all 4 threads use the same bit here.
3240 * Hash must be flushed in realmode in order to use tlbiel.
3242 kvmppc_radix_check_need_tlb_flush(vc->kvm, pcpu, NULL);
3246 * Interrupts will be enabled once we get into the guest,
3247 * so tell lockdep that we're about to enable interrupts.
3249 trace_hardirqs_on();
3251 guest_enter_irqoff();
3253 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3255 this_cpu_disable_ftrace();
3257 trap = __kvmppc_vcore_entry();
3259 this_cpu_enable_ftrace();
3261 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3263 trace_hardirqs_off();
3264 set_irq_happened(trap);
3266 spin_lock(&vc->lock);
3267 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3268 vc->vcore_state = VCORE_EXITING;
3270 /* wait for secondary threads to finish writing their state to memory */
3271 kvmppc_wait_for_nap(controlled_threads);
3273 /* Return to whole-core mode if we split the core earlier */
3275 unsigned long hid0 = mfspr(SPRN_HID0);
3276 unsigned long loops = 0;
3278 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3279 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3281 mtspr(SPRN_HID0, hid0);
3284 hid0 = mfspr(SPRN_HID0);
3285 if (!(hid0 & stat_bit))
3290 } else if (hpt_on_radix) {
3291 /* Wait for all threads to have seen final sync */
3292 for (thr = 1; thr < controlled_threads; ++thr) {
3293 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3295 while (paca->kvm_hstate.kvm_split_mode) {
3302 split_info.do_nap = 0;
3304 kvmppc_set_host_core(pcpu);
3309 /* Let secondaries go back to the offline loop */
3310 for (i = 0; i < controlled_threads; ++i) {
3311 kvmppc_release_hwthread(pcpu + i);
3312 if (sip && sip->napped[i])
3313 kvmppc_ipi_thread(pcpu + i);
3314 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3317 spin_unlock(&vc->lock);
3319 /* make sure updates to secondary vcpu structs are visible now */
3324 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3325 pvc = core_info.vc[sub];
3326 post_guest_process(pvc, pvc == vc);
3329 spin_lock(&vc->lock);
3332 vc->vcore_state = VCORE_INACTIVE;
3333 trace_kvmppc_run_core(vc, 1);
3337 * Load up hypervisor-mode registers on P9.
3339 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3342 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3344 u64 tb, purr, spurr;
3346 unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3347 unsigned long host_ciabr = mfspr(SPRN_CIABR);
3348 unsigned long host_dawr = mfspr(SPRN_DAWR);
3349 unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3350 unsigned long host_psscr = mfspr(SPRN_PSSCR);
3351 unsigned long host_pidr = mfspr(SPRN_PID);
3353 hdec = time_limit - mftb();
3355 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3356 mtspr(SPRN_HDEC, hdec);
3358 if (vc->tb_offset) {
3359 u64 new_tb = mftb() + vc->tb_offset;
3360 mtspr(SPRN_TBU40, new_tb);
3362 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3363 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3364 vc->tb_offset_applied = vc->tb_offset;
3368 mtspr(SPRN_PCR, vc->pcr);
3369 mtspr(SPRN_DPDES, vc->dpdes);
3370 mtspr(SPRN_VTB, vc->vtb);
3372 local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3373 local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3374 mtspr(SPRN_PURR, vcpu->arch.purr);
3375 mtspr(SPRN_SPURR, vcpu->arch.spurr);
3377 if (cpu_has_feature(CPU_FTR_DAWR)) {
3378 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3379 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3381 mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3382 mtspr(SPRN_IC, vcpu->arch.ic);
3383 mtspr(SPRN_PID, vcpu->arch.pid);
3385 mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3386 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3388 mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3390 mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3391 mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3392 mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3393 mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3395 mtspr(SPRN_AMOR, ~0UL);
3397 mtspr(SPRN_LPCR, lpcr);
3400 kvmppc_xive_push_vcpu(vcpu);
3402 mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3403 mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3405 trap = __kvmhv_vcpu_entry_p9(vcpu);
3407 /* Advance host PURR/SPURR by the amount used by guest */
3408 purr = mfspr(SPRN_PURR);
3409 spurr = mfspr(SPRN_SPURR);
3410 mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3411 purr - vcpu->arch.purr);
3412 mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3413 spurr - vcpu->arch.spurr);
3414 vcpu->arch.purr = purr;
3415 vcpu->arch.spurr = spurr;
3417 vcpu->arch.ic = mfspr(SPRN_IC);
3418 vcpu->arch.pid = mfspr(SPRN_PID);
3419 vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3421 vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3422 vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3423 vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3424 vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3426 /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3427 mtspr(SPRN_PSSCR, host_psscr |
3428 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3429 mtspr(SPRN_HFSCR, host_hfscr);
3430 mtspr(SPRN_CIABR, host_ciabr);
3431 mtspr(SPRN_DAWR, host_dawr);
3432 mtspr(SPRN_DAWRX, host_dawrx);
3433 mtspr(SPRN_PID, host_pidr);
3436 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3437 * case we interrupted the guest between a tlbie and a ptesync.
3439 asm volatile("eieio; tlbsync; ptesync");
3441 mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid); /* restore host LPID */
3444 vc->dpdes = mfspr(SPRN_DPDES);
3445 vc->vtb = mfspr(SPRN_VTB);
3446 mtspr(SPRN_DPDES, 0);
3450 if (vc->tb_offset_applied) {
3451 u64 new_tb = mftb() - vc->tb_offset_applied;
3452 mtspr(SPRN_TBU40, new_tb);
3454 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3455 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3456 vc->tb_offset_applied = 0;
3459 mtspr(SPRN_HDEC, 0x7fffffff);
3460 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3466 * Virtual-mode guest entry for POWER9 and later when the host and
3467 * guest are both using the radix MMU. The LPIDR has already been set.
3469 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3472 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3473 unsigned long host_dscr = mfspr(SPRN_DSCR);
3474 unsigned long host_tidr = mfspr(SPRN_TIDR);
3475 unsigned long host_iamr = mfspr(SPRN_IAMR);
3476 unsigned long host_amr = mfspr(SPRN_AMR);
3481 dec = mfspr(SPRN_DEC);
3484 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3485 local_paca->kvm_hstate.dec_expires = dec + tb;
3486 if (local_paca->kvm_hstate.dec_expires < time_limit)
3487 time_limit = local_paca->kvm_hstate.dec_expires;
3489 vcpu->arch.ceded = 0;
3491 kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */
3493 kvmppc_subcore_enter_guest();
3495 vc->entry_exit_map = 1;
3498 if (vcpu->arch.vpa.pinned_addr) {
3499 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3500 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3501 lp->yield_count = cpu_to_be32(yield_count);
3502 vcpu->arch.vpa.dirty = 1;
3505 if (cpu_has_feature(CPU_FTR_TM) ||
3506 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3507 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3509 kvmhv_load_guest_pmu(vcpu);
3511 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3512 load_fp_state(&vcpu->arch.fp);
3513 #ifdef CONFIG_ALTIVEC
3514 load_vr_state(&vcpu->arch.vr);
3517 mtspr(SPRN_DSCR, vcpu->arch.dscr);
3518 mtspr(SPRN_IAMR, vcpu->arch.iamr);
3519 mtspr(SPRN_PSPB, vcpu->arch.pspb);
3520 mtspr(SPRN_FSCR, vcpu->arch.fscr);
3521 mtspr(SPRN_TAR, vcpu->arch.tar);
3522 mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3523 mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3524 mtspr(SPRN_BESCR, vcpu->arch.bescr);
3525 mtspr(SPRN_WORT, vcpu->arch.wort);
3526 mtspr(SPRN_TIDR, vcpu->arch.tid);
3527 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3528 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3529 mtspr(SPRN_AMR, vcpu->arch.amr);
3530 mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3532 if (!(vcpu->arch.ctrl & 1))
3533 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3535 mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3537 if (kvmhv_on_pseries()) {
3538 /* call our hypervisor to load up HV regs and go */
3539 struct hv_guest_state hvregs;
3541 kvmhv_save_hv_regs(vcpu, &hvregs);
3543 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3544 hvregs.version = HV_GUEST_STATE_VERSION;
3545 if (vcpu->arch.nested) {
3546 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3547 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3549 hvregs.lpid = vcpu->kvm->arch.lpid;
3550 hvregs.vcpu_token = vcpu->vcpu_id;
3552 hvregs.hdec_expiry = time_limit;
3553 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3554 __pa(&vcpu->arch.regs));
3555 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3556 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3557 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3558 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3560 /* H_CEDE has to be handled now, not later */
3561 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3562 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3563 kvmppc_nested_cede(vcpu);
3567 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3570 vcpu->arch.slb_max = 0;
3571 dec = mfspr(SPRN_DEC);
3573 vcpu->arch.dec_expires = dec + tb;
3575 vcpu->arch.thread_cpu = -1;
3576 vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3578 vcpu->arch.iamr = mfspr(SPRN_IAMR);
3579 vcpu->arch.pspb = mfspr(SPRN_PSPB);
3580 vcpu->arch.fscr = mfspr(SPRN_FSCR);
3581 vcpu->arch.tar = mfspr(SPRN_TAR);
3582 vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3583 vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3584 vcpu->arch.bescr = mfspr(SPRN_BESCR);
3585 vcpu->arch.wort = mfspr(SPRN_WORT);
3586 vcpu->arch.tid = mfspr(SPRN_TIDR);
3587 vcpu->arch.amr = mfspr(SPRN_AMR);
3588 vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3589 vcpu->arch.dscr = mfspr(SPRN_DSCR);
3591 mtspr(SPRN_PSPB, 0);
3592 mtspr(SPRN_WORT, 0);
3593 mtspr(SPRN_UAMOR, 0);
3594 mtspr(SPRN_DSCR, host_dscr);
3595 mtspr(SPRN_TIDR, host_tidr);
3596 mtspr(SPRN_IAMR, host_iamr);
3597 mtspr(SPRN_PSPB, 0);
3599 if (host_amr != vcpu->arch.amr)
3600 mtspr(SPRN_AMR, host_amr);
3602 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3603 store_fp_state(&vcpu->arch.fp);
3604 #ifdef CONFIG_ALTIVEC
3605 store_vr_state(&vcpu->arch.vr);
3608 if (cpu_has_feature(CPU_FTR_TM) ||
3609 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3610 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3613 if (vcpu->arch.vpa.pinned_addr) {
3614 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3615 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3616 lp->yield_count = cpu_to_be32(yield_count);
3617 vcpu->arch.vpa.dirty = 1;
3618 save_pmu = lp->pmcregs_in_use;
3621 kvmhv_save_guest_pmu(vcpu, save_pmu);
3623 vc->entry_exit_map = 0x101;
3626 mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3628 kvmhv_load_host_pmu();
3630 kvmppc_subcore_exit_guest();
3636 * Wait for some other vcpu thread to execute us, and
3637 * wake us up when we need to handle something in the host.
3639 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3640 struct kvm_vcpu *vcpu, int wait_state)
3644 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3645 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3646 spin_unlock(&vc->lock);
3648 spin_lock(&vc->lock);
3650 finish_wait(&vcpu->arch.cpu_run, &wait);
3653 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3655 if (!halt_poll_ns_grow)
3658 vc->halt_poll_ns *= halt_poll_ns_grow;
3659 if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3660 vc->halt_poll_ns = halt_poll_ns_grow_start;
3663 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3665 if (halt_poll_ns_shrink == 0)
3666 vc->halt_poll_ns = 0;
3668 vc->halt_poll_ns /= halt_poll_ns_shrink;
3671 #ifdef CONFIG_KVM_XICS
3672 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3674 if (!xics_on_xive())
3676 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3677 vcpu->arch.xive_saved_state.cppr;
3680 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3684 #endif /* CONFIG_KVM_XICS */
3686 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3688 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3689 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3696 * Check to see if any of the runnable vcpus on the vcore have pending
3697 * exceptions or are no longer ceded
3699 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3701 struct kvm_vcpu *vcpu;
3704 for_each_runnable_thread(i, vcpu, vc) {
3705 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3713 * All the vcpus in this vcore are idle, so wait for a decrementer
3714 * or external interrupt to one of the vcpus. vc->lock is held.
3716 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3718 ktime_t cur, start_poll, start_wait;
3721 DECLARE_SWAITQUEUE(wait);
3723 /* Poll for pending exceptions and ceded state */
3724 cur = start_poll = ktime_get();
3725 if (vc->halt_poll_ns) {
3726 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3727 ++vc->runner->stat.halt_attempted_poll;
3729 vc->vcore_state = VCORE_POLLING;
3730 spin_unlock(&vc->lock);
3733 if (kvmppc_vcore_check_block(vc)) {
3738 } while (single_task_running() && ktime_before(cur, stop));
3740 spin_lock(&vc->lock);
3741 vc->vcore_state = VCORE_INACTIVE;
3744 ++vc->runner->stat.halt_successful_poll;
3749 prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3751 if (kvmppc_vcore_check_block(vc)) {
3752 finish_swait(&vc->wq, &wait);
3754 /* If we polled, count this as a successful poll */
3755 if (vc->halt_poll_ns)
3756 ++vc->runner->stat.halt_successful_poll;
3760 start_wait = ktime_get();
3762 vc->vcore_state = VCORE_SLEEPING;
3763 trace_kvmppc_vcore_blocked(vc, 0);
3764 spin_unlock(&vc->lock);
3766 finish_swait(&vc->wq, &wait);
3767 spin_lock(&vc->lock);
3768 vc->vcore_state = VCORE_INACTIVE;
3769 trace_kvmppc_vcore_blocked(vc, 1);
3770 ++vc->runner->stat.halt_successful_wait;
3775 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3777 /* Attribute wait time */
3779 vc->runner->stat.halt_wait_ns +=
3780 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3781 /* Attribute failed poll time */
3782 if (vc->halt_poll_ns)
3783 vc->runner->stat.halt_poll_fail_ns +=
3784 ktime_to_ns(start_wait) -
3785 ktime_to_ns(start_poll);
3787 /* Attribute successful poll time */
3788 if (vc->halt_poll_ns)
3789 vc->runner->stat.halt_poll_success_ns +=
3791 ktime_to_ns(start_poll);
3794 /* Adjust poll time */
3796 if (block_ns <= vc->halt_poll_ns)
3798 /* We slept and blocked for longer than the max halt time */
3799 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3800 shrink_halt_poll_ns(vc);
3801 /* We slept and our poll time is too small */
3802 else if (vc->halt_poll_ns < halt_poll_ns &&
3803 block_ns < halt_poll_ns)
3804 grow_halt_poll_ns(vc);
3805 if (vc->halt_poll_ns > halt_poll_ns)
3806 vc->halt_poll_ns = halt_poll_ns;
3808 vc->halt_poll_ns = 0;
3810 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3814 * This never fails for a radix guest, as none of the operations it does
3815 * for a radix guest can fail or have a way to report failure.
3816 * kvmhv_run_single_vcpu() relies on this fact.
3818 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3821 struct kvm *kvm = vcpu->kvm;
3823 mutex_lock(&kvm->lock);
3824 if (!kvm->arch.mmu_ready) {
3825 if (!kvm_is_radix(kvm))
3826 r = kvmppc_hv_setup_htab_rma(vcpu);
3828 if (cpu_has_feature(CPU_FTR_ARCH_300))
3829 kvmppc_setup_partition_table(kvm);
3830 kvm->arch.mmu_ready = 1;
3833 mutex_unlock(&kvm->lock);
3837 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3840 struct kvmppc_vcore *vc;
3843 trace_kvmppc_run_vcpu_enter(vcpu);
3845 kvm_run->exit_reason = 0;
3846 vcpu->arch.ret = RESUME_GUEST;
3847 vcpu->arch.trap = 0;
3848 kvmppc_update_vpas(vcpu);
3851 * Synchronize with other threads in this virtual core
3853 vc = vcpu->arch.vcore;
3854 spin_lock(&vc->lock);
3855 vcpu->arch.ceded = 0;
3856 vcpu->arch.run_task = current;
3857 vcpu->arch.kvm_run = kvm_run;
3858 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3859 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3860 vcpu->arch.busy_preempt = TB_NIL;
3861 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3865 * This happens the first time this is called for a vcpu.
3866 * If the vcore is already running, we may be able to start
3867 * this thread straight away and have it join in.
3869 if (!signal_pending(current)) {
3870 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3871 vc->vcore_state == VCORE_RUNNING) &&
3872 !VCORE_IS_EXITING(vc)) {
3873 kvmppc_create_dtl_entry(vcpu, vc);
3874 kvmppc_start_thread(vcpu, vc);
3875 trace_kvm_guest_enter(vcpu);
3876 } else if (vc->vcore_state == VCORE_SLEEPING) {
3877 swake_up_one(&vc->wq);
3882 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3883 !signal_pending(current)) {
3884 /* See if the MMU is ready to go */
3885 if (!vcpu->kvm->arch.mmu_ready) {
3886 spin_unlock(&vc->lock);
3887 r = kvmhv_setup_mmu(vcpu);
3888 spin_lock(&vc->lock);
3890 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3891 kvm_run->fail_entry.
3892 hardware_entry_failure_reason = 0;
3898 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3899 kvmppc_vcore_end_preempt(vc);
3901 if (vc->vcore_state != VCORE_INACTIVE) {
3902 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3905 for_each_runnable_thread(i, v, vc) {
3906 kvmppc_core_prepare_to_enter(v);
3907 if (signal_pending(v->arch.run_task)) {
3908 kvmppc_remove_runnable(vc, v);
3909 v->stat.signal_exits++;
3910 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3911 v->arch.ret = -EINTR;
3912 wake_up(&v->arch.cpu_run);
3915 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3918 for_each_runnable_thread(i, v, vc) {
3919 if (!kvmppc_vcpu_woken(v))
3920 n_ceded += v->arch.ceded;
3925 if (n_ceded == vc->n_runnable) {
3926 kvmppc_vcore_blocked(vc);
3927 } else if (need_resched()) {
3928 kvmppc_vcore_preempt(vc);
3929 /* Let something else run */
3930 cond_resched_lock(&vc->lock);
3931 if (vc->vcore_state == VCORE_PREEMPT)
3932 kvmppc_vcore_end_preempt(vc);
3934 kvmppc_run_core(vc);
3939 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3940 (vc->vcore_state == VCORE_RUNNING ||
3941 vc->vcore_state == VCORE_EXITING ||
3942 vc->vcore_state == VCORE_PIGGYBACK))
3943 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3945 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3946 kvmppc_vcore_end_preempt(vc);
3948 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3949 kvmppc_remove_runnable(vc, vcpu);
3950 vcpu->stat.signal_exits++;
3951 kvm_run->exit_reason = KVM_EXIT_INTR;
3952 vcpu->arch.ret = -EINTR;
3955 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3956 /* Wake up some vcpu to run the core */
3958 v = next_runnable_thread(vc, &i);
3959 wake_up(&v->arch.cpu_run);
3962 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3963 spin_unlock(&vc->lock);
3964 return vcpu->arch.ret;
3967 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
3968 struct kvm_vcpu *vcpu, u64 time_limit,
3973 struct kvmppc_vcore *vc;
3974 struct kvm *kvm = vcpu->kvm;
3975 struct kvm_nested_guest *nested = vcpu->arch.nested;
3977 trace_kvmppc_run_vcpu_enter(vcpu);
3979 kvm_run->exit_reason = 0;
3980 vcpu->arch.ret = RESUME_GUEST;
3981 vcpu->arch.trap = 0;
3983 vc = vcpu->arch.vcore;
3984 vcpu->arch.ceded = 0;
3985 vcpu->arch.run_task = current;
3986 vcpu->arch.kvm_run = kvm_run;
3987 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3988 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3989 vcpu->arch.busy_preempt = TB_NIL;
3990 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
3991 vc->runnable_threads[0] = vcpu;
3995 /* See if the MMU is ready to go */
3996 if (!kvm->arch.mmu_ready)
3997 kvmhv_setup_mmu(vcpu);
4002 kvmppc_update_vpas(vcpu);
4004 init_vcore_to_run(vc);
4005 vc->preempt_tb = TB_NIL;
4008 pcpu = smp_processor_id();
4010 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4012 local_irq_disable();
4014 if (signal_pending(current))
4016 if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4020 kvmppc_core_prepare_to_enter(vcpu);
4021 if (vcpu->arch.doorbell_request) {
4024 vcpu->arch.doorbell_request = 0;
4026 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4027 &vcpu->arch.pending_exceptions))
4029 } else if (vcpu->arch.pending_exceptions ||
4030 vcpu->arch.doorbell_request ||
4031 xive_interrupt_pending(vcpu)) {
4032 vcpu->arch.ret = RESUME_HOST;
4036 kvmppc_clear_host_core(pcpu);
4038 local_paca->kvm_hstate.tid = 0;
4039 local_paca->kvm_hstate.napping = 0;
4040 local_paca->kvm_hstate.kvm_split_mode = NULL;
4041 kvmppc_start_thread(vcpu, vc);
4042 kvmppc_create_dtl_entry(vcpu, vc);
4043 trace_kvm_guest_enter(vcpu);
4045 vc->vcore_state = VCORE_RUNNING;
4046 trace_kvmppc_run_core(vc, 0);
4048 if (cpu_has_feature(CPU_FTR_HVMODE))
4049 kvmppc_radix_check_need_tlb_flush(kvm, pcpu, nested);
4051 trace_hardirqs_on();
4052 guest_enter_irqoff();
4054 srcu_idx = srcu_read_lock(&kvm->srcu);
4056 this_cpu_disable_ftrace();
4058 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4059 vcpu->arch.trap = trap;
4061 this_cpu_enable_ftrace();
4063 srcu_read_unlock(&kvm->srcu, srcu_idx);
4065 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4066 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4070 trace_hardirqs_off();
4071 set_irq_happened(trap);
4073 kvmppc_set_host_core(pcpu);
4078 cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4082 /* cancel pending decrementer exception if DEC is now positive */
4083 if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4084 kvmppc_core_dequeue_dec(vcpu);
4086 trace_kvm_guest_exit(vcpu);
4090 r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4092 r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4096 if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4097 !kvmppc_vcpu_woken(vcpu)) {
4098 kvmppc_set_timer(vcpu);
4099 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4100 if (signal_pending(current)) {
4101 vcpu->stat.signal_exits++;
4102 kvm_run->exit_reason = KVM_EXIT_INTR;
4103 vcpu->arch.ret = -EINTR;
4106 spin_lock(&vc->lock);
4107 kvmppc_vcore_blocked(vc);
4108 spin_unlock(&vc->lock);
4111 vcpu->arch.ceded = 0;
4113 vc->vcore_state = VCORE_INACTIVE;
4114 trace_kvmppc_run_core(vc, 1);
4117 kvmppc_remove_runnable(vc, vcpu);
4118 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4120 return vcpu->arch.ret;
4123 vcpu->stat.signal_exits++;
4124 kvm_run->exit_reason = KVM_EXIT_INTR;
4125 vcpu->arch.ret = -EINTR;
4132 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4136 unsigned long ebb_regs[3] = {}; /* shut up GCC */
4137 unsigned long user_tar = 0;
4138 unsigned int user_vrsave;
4141 if (!vcpu->arch.sane) {
4142 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4147 * Don't allow entry with a suspended transaction, because
4148 * the guest entry/exit code will lose it.
4149 * If the guest has TM enabled, save away their TM-related SPRs
4150 * (they will get restored by the TM unavailable interrupt).
4152 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4153 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4154 (current->thread.regs->msr & MSR_TM)) {
4155 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4156 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4157 run->fail_entry.hardware_entry_failure_reason = 0;
4160 /* Enable TM so we can read the TM SPRs */
4161 mtmsr(mfmsr() | MSR_TM);
4162 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4163 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4164 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4165 current->thread.regs->msr &= ~MSR_TM;
4170 * Force online to 1 for the sake of old userspace which doesn't
4173 if (!vcpu->arch.online) {
4174 atomic_inc(&vcpu->arch.vcore->online_count);
4175 vcpu->arch.online = 1;
4178 kvmppc_core_prepare_to_enter(vcpu);
4180 /* No need to go into the guest when all we'll do is come back out */
4181 if (signal_pending(current)) {
4182 run->exit_reason = KVM_EXIT_INTR;
4187 atomic_inc(&kvm->arch.vcpus_running);
4188 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4191 flush_all_to_thread(current);
4193 /* Save userspace EBB and other register values */
4194 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4195 ebb_regs[0] = mfspr(SPRN_EBBHR);
4196 ebb_regs[1] = mfspr(SPRN_EBBRR);
4197 ebb_regs[2] = mfspr(SPRN_BESCR);
4198 user_tar = mfspr(SPRN_TAR);
4200 user_vrsave = mfspr(SPRN_VRSAVE);
4202 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4203 vcpu->arch.pgdir = current->mm->pgd;
4204 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4208 * The early POWER9 chips that can't mix radix and HPT threads
4209 * on the same core also need the workaround for the problem
4210 * where the TLB would prefetch entries in the guest exit path
4211 * for radix guests using the guest PIDR value and LPID 0.
4212 * The workaround is in the old path (kvmppc_run_vcpu())
4213 * but not the new path (kvmhv_run_single_vcpu()).
4215 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4216 !no_mixing_hpt_and_radix)
4217 r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4218 vcpu->arch.vcore->lpcr);
4220 r = kvmppc_run_vcpu(run, vcpu);
4222 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4223 !(vcpu->arch.shregs.msr & MSR_PR)) {
4224 trace_kvm_hcall_enter(vcpu);
4225 r = kvmppc_pseries_do_hcall(vcpu);
4226 trace_kvm_hcall_exit(vcpu, r);
4227 kvmppc_core_prepare_to_enter(vcpu);
4228 } else if (r == RESUME_PAGE_FAULT) {
4229 srcu_idx = srcu_read_lock(&kvm->srcu);
4230 r = kvmppc_book3s_hv_page_fault(run, vcpu,
4231 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4232 srcu_read_unlock(&kvm->srcu, srcu_idx);
4233 } else if (r == RESUME_PASSTHROUGH) {
4234 if (WARN_ON(xics_on_xive()))
4237 r = kvmppc_xics_rm_complete(vcpu, 0);
4239 } while (is_kvmppc_resume_guest(r));
4241 /* Restore userspace EBB and other register values */
4242 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4243 mtspr(SPRN_EBBHR, ebb_regs[0]);
4244 mtspr(SPRN_EBBRR, ebb_regs[1]);
4245 mtspr(SPRN_BESCR, ebb_regs[2]);
4246 mtspr(SPRN_TAR, user_tar);
4247 mtspr(SPRN_FSCR, current->thread.fscr);
4249 mtspr(SPRN_VRSAVE, user_vrsave);
4251 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4252 atomic_dec(&kvm->arch.vcpus_running);
4256 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4257 int shift, int sllp)
4259 (*sps)->page_shift = shift;
4260 (*sps)->slb_enc = sllp;
4261 (*sps)->enc[0].page_shift = shift;
4262 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4264 * Add 16MB MPSS support (may get filtered out by userspace)
4267 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4269 (*sps)->enc[1].page_shift = 24;
4270 (*sps)->enc[1].pte_enc = penc;
4276 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4277 struct kvm_ppc_smmu_info *info)
4279 struct kvm_ppc_one_seg_page_size *sps;
4282 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4283 * POWER7 doesn't support keys for instruction accesses,
4284 * POWER8 and POWER9 do.
4286 info->data_keys = 32;
4287 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4289 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4290 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4291 info->slb_size = 32;
4293 /* We only support these sizes for now, and no muti-size segments */
4294 sps = &info->sps[0];
4295 kvmppc_add_seg_page_size(&sps, 12, 0);
4296 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4297 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4299 /* If running as a nested hypervisor, we don't support HPT guests */
4300 if (kvmhv_on_pseries())
4301 info->flags |= KVM_PPC_NO_HASH;
4307 * Get (and clear) the dirty memory log for a memory slot.
4309 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4310 struct kvm_dirty_log *log)
4312 struct kvm_memslots *slots;
4313 struct kvm_memory_slot *memslot;
4316 unsigned long *buf, *p;
4317 struct kvm_vcpu *vcpu;
4319 mutex_lock(&kvm->slots_lock);
4322 if (log->slot >= KVM_USER_MEM_SLOTS)
4325 slots = kvm_memslots(kvm);
4326 memslot = id_to_memslot(slots, log->slot);
4328 if (!memslot->dirty_bitmap)
4332 * Use second half of bitmap area because both HPT and radix
4333 * accumulate bits in the first half.
4335 n = kvm_dirty_bitmap_bytes(memslot);
4336 buf = memslot->dirty_bitmap + n / sizeof(long);
4339 if (kvm_is_radix(kvm))
4340 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4342 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4347 * We accumulate dirty bits in the first half of the
4348 * memslot's dirty_bitmap area, for when pages are paged
4349 * out or modified by the host directly. Pick up these
4350 * bits and add them to the map.
4352 p = memslot->dirty_bitmap;
4353 for (i = 0; i < n / sizeof(long); ++i)
4354 buf[i] |= xchg(&p[i], 0);
4356 /* Harvest dirty bits from VPA and DTL updates */
4357 /* Note: we never modify the SLB shadow buffer areas */
4358 kvm_for_each_vcpu(i, vcpu, kvm) {
4359 spin_lock(&vcpu->arch.vpa_update_lock);
4360 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4361 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4362 spin_unlock(&vcpu->arch.vpa_update_lock);
4366 if (copy_to_user(log->dirty_bitmap, buf, n))
4371 mutex_unlock(&kvm->slots_lock);
4375 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4376 struct kvm_memory_slot *dont)
4378 if (!dont || free->arch.rmap != dont->arch.rmap) {
4379 vfree(free->arch.rmap);
4380 free->arch.rmap = NULL;
4384 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4385 unsigned long npages)
4387 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4388 if (!slot->arch.rmap)
4394 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4395 struct kvm_memory_slot *memslot,
4396 const struct kvm_userspace_memory_region *mem)
4401 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4402 const struct kvm_userspace_memory_region *mem,
4403 const struct kvm_memory_slot *old,
4404 const struct kvm_memory_slot *new,
4405 enum kvm_mr_change change)
4407 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4410 * If we are making a new memslot, it might make
4411 * some address that was previously cached as emulated
4412 * MMIO be no longer emulated MMIO, so invalidate
4413 * all the caches of emulated MMIO translations.
4416 atomic64_inc(&kvm->arch.mmio_update);
4419 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4420 * have already called kvm_arch_flush_shadow_memslot() to
4421 * flush shadow mappings. For KVM_MR_CREATE we have no
4422 * previous mappings. So the only case to handle is
4423 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4425 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4426 * to get rid of any THP PTEs in the partition-scoped page tables
4427 * so we can track dirtiness at the page level; we flush when
4428 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4431 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4432 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4433 kvmppc_radix_flush_memslot(kvm, old);
4437 * Update LPCR values in kvm->arch and in vcores.
4438 * Caller must hold kvm->lock.
4440 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4445 if ((kvm->arch.lpcr & mask) == lpcr)
4448 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4450 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4451 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4454 spin_lock(&vc->lock);
4455 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4456 spin_unlock(&vc->lock);
4457 if (++cores_done >= kvm->arch.online_vcores)
4462 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4467 void kvmppc_setup_partition_table(struct kvm *kvm)
4469 unsigned long dw0, dw1;
4471 if (!kvm_is_radix(kvm)) {
4472 /* PS field - page size for VRMA */
4473 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4474 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4475 /* HTABSIZE and HTABORG fields */
4476 dw0 |= kvm->arch.sdr1;
4478 /* Second dword as set by userspace */
4479 dw1 = kvm->arch.process_table;
4481 dw0 = PATB_HR | radix__get_tree_size() |
4482 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4483 dw1 = PATB_GR | kvm->arch.process_table;
4485 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4489 * Set up HPT (hashed page table) and RMA (real-mode area).
4490 * Must be called with kvm->lock held.
4492 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4495 struct kvm *kvm = vcpu->kvm;
4497 struct kvm_memory_slot *memslot;
4498 struct vm_area_struct *vma;
4499 unsigned long lpcr = 0, senc;
4500 unsigned long psize, porder;
4503 /* Allocate hashed page table (if not done already) and reset it */
4504 if (!kvm->arch.hpt.virt) {
4505 int order = KVM_DEFAULT_HPT_ORDER;
4506 struct kvm_hpt_info info;
4508 err = kvmppc_allocate_hpt(&info, order);
4509 /* If we get here, it means userspace didn't specify a
4510 * size explicitly. So, try successively smaller
4511 * sizes if the default failed. */
4512 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4513 err = kvmppc_allocate_hpt(&info, order);
4516 pr_err("KVM: Couldn't alloc HPT\n");
4520 kvmppc_set_hpt(kvm, &info);
4523 /* Look up the memslot for guest physical address 0 */
4524 srcu_idx = srcu_read_lock(&kvm->srcu);
4525 memslot = gfn_to_memslot(kvm, 0);
4527 /* We must have some memory at 0 by now */
4529 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4532 /* Look up the VMA for the start of this memory slot */
4533 hva = memslot->userspace_addr;
4534 down_read(¤t->mm->mmap_sem);
4535 vma = find_vma(current->mm, hva);
4536 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4539 psize = vma_kernel_pagesize(vma);
4541 up_read(¤t->mm->mmap_sem);
4543 /* We can handle 4k, 64k or 16M pages in the VRMA */
4544 if (psize >= 0x1000000)
4546 else if (psize >= 0x10000)
4550 porder = __ilog2(psize);
4552 senc = slb_pgsize_encoding(psize);
4553 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4554 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4555 /* Create HPTEs in the hash page table for the VRMA */
4556 kvmppc_map_vrma(vcpu, memslot, porder);
4558 /* Update VRMASD field in the LPCR */
4559 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4560 /* the -4 is to account for senc values starting at 0x10 */
4561 lpcr = senc << (LPCR_VRMASD_SH - 4);
4562 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4565 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4569 srcu_read_unlock(&kvm->srcu, srcu_idx);
4574 up_read(¤t->mm->mmap_sem);
4578 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4579 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4581 if (nesting_enabled(kvm))
4582 kvmhv_release_all_nested(kvm);
4583 kvmppc_rmap_reset(kvm);
4584 kvm->arch.process_table = 0;
4585 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4586 spin_lock(&kvm->mmu_lock);
4587 kvm->arch.radix = 0;
4588 spin_unlock(&kvm->mmu_lock);
4589 kvmppc_free_radix(kvm);
4590 kvmppc_update_lpcr(kvm, LPCR_VPM1,
4591 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4595 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4596 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4600 err = kvmppc_init_vm_radix(kvm);
4603 kvmppc_rmap_reset(kvm);
4604 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4605 spin_lock(&kvm->mmu_lock);
4606 kvm->arch.radix = 1;
4607 spin_unlock(&kvm->mmu_lock);
4608 kvmppc_free_hpt(&kvm->arch.hpt);
4609 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4610 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4614 #ifdef CONFIG_KVM_XICS
4616 * Allocate a per-core structure for managing state about which cores are
4617 * running in the host versus the guest and for exchanging data between
4618 * real mode KVM and CPU running in the host.
4619 * This is only done for the first VM.
4620 * The allocated structure stays even if all VMs have stopped.
4621 * It is only freed when the kvm-hv module is unloaded.
4622 * It's OK for this routine to fail, we just don't support host
4623 * core operations like redirecting H_IPI wakeups.
4625 void kvmppc_alloc_host_rm_ops(void)
4627 struct kvmppc_host_rm_ops *ops;
4628 unsigned long l_ops;
4632 /* Not the first time here ? */
4633 if (kvmppc_host_rm_ops_hv != NULL)
4636 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4640 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4641 ops->rm_core = kzalloc(size, GFP_KERNEL);
4643 if (!ops->rm_core) {
4650 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4651 if (!cpu_online(cpu))
4654 core = cpu >> threads_shift;
4655 ops->rm_core[core].rm_state.in_host = 1;
4658 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4661 * Make the contents of the kvmppc_host_rm_ops structure visible
4662 * to other CPUs before we assign it to the global variable.
4663 * Do an atomic assignment (no locks used here), but if someone
4664 * beats us to it, just free our copy and return.
4667 l_ops = (unsigned long) ops;
4669 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4671 kfree(ops->rm_core);
4676 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4677 "ppc/kvm_book3s:prepare",
4678 kvmppc_set_host_core,
4679 kvmppc_clear_host_core);
4683 void kvmppc_free_host_rm_ops(void)
4685 if (kvmppc_host_rm_ops_hv) {
4686 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4687 kfree(kvmppc_host_rm_ops_hv->rm_core);
4688 kfree(kvmppc_host_rm_ops_hv);
4689 kvmppc_host_rm_ops_hv = NULL;
4694 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4696 unsigned long lpcr, lpid;
4700 /* Allocate the guest's logical partition ID */
4702 lpid = kvmppc_alloc_lpid();
4705 kvm->arch.lpid = lpid;
4707 kvmppc_alloc_host_rm_ops();
4709 kvmhv_vm_nested_init(kvm);
4712 * Since we don't flush the TLB when tearing down a VM,
4713 * and this lpid might have previously been used,
4714 * make sure we flush on each core before running the new VM.
4715 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4716 * does this flush for us.
4718 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4719 cpumask_setall(&kvm->arch.need_tlb_flush);
4721 /* Start out with the default set of hcalls enabled */
4722 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4723 sizeof(kvm->arch.enabled_hcalls));
4725 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4726 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4728 /* Init LPCR for virtual RMA mode */
4729 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4730 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4731 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4732 lpcr &= LPCR_PECE | LPCR_LPES;
4736 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4737 LPCR_VPM0 | LPCR_VPM1;
4738 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4739 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4740 /* On POWER8 turn on online bit to enable PURR/SPURR */
4741 if (cpu_has_feature(CPU_FTR_ARCH_207S))
4744 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4745 * Set HVICE bit to enable hypervisor virtualization interrupts.
4746 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4747 * be unnecessary but better safe than sorry in case we re-enable
4748 * EE in HV mode with this LPCR still set)
4750 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4752 lpcr |= LPCR_HVICE | LPCR_HEIC;
4755 * If xive is enabled, we route 0x500 interrupts directly
4763 * If the host uses radix, the guest starts out as radix.
4765 if (radix_enabled()) {
4766 kvm->arch.radix = 1;
4767 kvm->arch.mmu_ready = 1;
4769 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4770 ret = kvmppc_init_vm_radix(kvm);
4772 kvmppc_free_lpid(kvm->arch.lpid);
4775 kvmppc_setup_partition_table(kvm);
4778 kvm->arch.lpcr = lpcr;
4780 /* Initialization for future HPT resizes */
4781 kvm->arch.resize_hpt = NULL;
4784 * Work out how many sets the TLB has, for the use of
4785 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4787 if (radix_enabled())
4788 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
4789 else if (cpu_has_feature(CPU_FTR_ARCH_300))
4790 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
4791 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4792 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
4794 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
4797 * Track that we now have a HV mode VM active. This blocks secondary
4798 * CPU threads from coming online.
4799 * On POWER9, we only need to do this if the "indep_threads_mode"
4800 * module parameter has been set to N.
4802 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4803 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4804 pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4805 kvm->arch.threads_indep = true;
4807 kvm->arch.threads_indep = indep_threads_mode;
4810 if (!kvm->arch.threads_indep)
4811 kvm_hv_vm_activated();
4814 * Initialize smt_mode depending on processor.
4815 * POWER8 and earlier have to use "strict" threading, where
4816 * all vCPUs in a vcore have to run on the same (sub)core,
4817 * whereas on POWER9 the threads can each run a different
4820 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4821 kvm->arch.smt_mode = threads_per_subcore;
4823 kvm->arch.smt_mode = 1;
4824 kvm->arch.emul_smt_mode = 1;
4827 * Create a debugfs directory for the VM
4829 snprintf(buf, sizeof(buf), "vm%d", current->pid);
4830 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4831 kvmppc_mmu_debugfs_init(kvm);
4832 if (radix_enabled())
4833 kvmhv_radix_debugfs_init(kvm);
4838 static void kvmppc_free_vcores(struct kvm *kvm)
4842 for (i = 0; i < KVM_MAX_VCORES; ++i)
4843 kfree(kvm->arch.vcores[i]);
4844 kvm->arch.online_vcores = 0;
4847 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4849 debugfs_remove_recursive(kvm->arch.debugfs_dir);
4851 if (!kvm->arch.threads_indep)
4852 kvm_hv_vm_deactivated();
4854 kvmppc_free_vcores(kvm);
4857 if (kvm_is_radix(kvm))
4858 kvmppc_free_radix(kvm);
4860 kvmppc_free_hpt(&kvm->arch.hpt);
4862 /* Perform global invalidation and return lpid to the pool */
4863 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4864 if (nesting_enabled(kvm))
4865 kvmhv_release_all_nested(kvm);
4866 kvm->arch.process_table = 0;
4867 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4869 kvmppc_free_lpid(kvm->arch.lpid);
4871 kvmppc_free_pimap(kvm);
4874 /* We don't need to emulate any privileged instructions or dcbz */
4875 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4876 unsigned int inst, int *advance)
4878 return EMULATE_FAIL;
4881 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4884 return EMULATE_FAIL;
4887 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4890 return EMULATE_FAIL;
4893 static int kvmppc_core_check_processor_compat_hv(void)
4895 if (cpu_has_feature(CPU_FTR_HVMODE) &&
4896 cpu_has_feature(CPU_FTR_ARCH_206))
4899 /* POWER9 in radix mode is capable of being a nested hypervisor. */
4900 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4906 #ifdef CONFIG_KVM_XICS
4908 void kvmppc_free_pimap(struct kvm *kvm)
4910 kfree(kvm->arch.pimap);
4913 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4915 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4918 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4920 struct irq_desc *desc;
4921 struct kvmppc_irq_map *irq_map;
4922 struct kvmppc_passthru_irqmap *pimap;
4923 struct irq_chip *chip;
4926 if (!kvm_irq_bypass)
4929 desc = irq_to_desc(host_irq);
4933 mutex_lock(&kvm->lock);
4935 pimap = kvm->arch.pimap;
4936 if (pimap == NULL) {
4937 /* First call, allocate structure to hold IRQ map */
4938 pimap = kvmppc_alloc_pimap();
4939 if (pimap == NULL) {
4940 mutex_unlock(&kvm->lock);
4943 kvm->arch.pimap = pimap;
4947 * For now, we only support interrupts for which the EOI operation
4948 * is an OPAL call followed by a write to XIRR, since that's
4949 * what our real-mode EOI code does, or a XIVE interrupt
4951 chip = irq_data_get_irq_chip(&desc->irq_data);
4952 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4953 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4954 host_irq, guest_gsi);
4955 mutex_unlock(&kvm->lock);
4960 * See if we already have an entry for this guest IRQ number.
4961 * If it's mapped to a hardware IRQ number, that's an error,
4962 * otherwise re-use this entry.
4964 for (i = 0; i < pimap->n_mapped; i++) {
4965 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4966 if (pimap->mapped[i].r_hwirq) {
4967 mutex_unlock(&kvm->lock);
4974 if (i == KVMPPC_PIRQ_MAPPED) {
4975 mutex_unlock(&kvm->lock);
4976 return -EAGAIN; /* table is full */
4979 irq_map = &pimap->mapped[i];
4981 irq_map->v_hwirq = guest_gsi;
4982 irq_map->desc = desc;
4985 * Order the above two stores before the next to serialize with
4986 * the KVM real mode handler.
4989 irq_map->r_hwirq = desc->irq_data.hwirq;
4991 if (i == pimap->n_mapped)
4995 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4997 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4999 irq_map->r_hwirq = 0;
5001 mutex_unlock(&kvm->lock);
5006 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5008 struct irq_desc *desc;
5009 struct kvmppc_passthru_irqmap *pimap;
5012 if (!kvm_irq_bypass)
5015 desc = irq_to_desc(host_irq);
5019 mutex_lock(&kvm->lock);
5020 if (!kvm->arch.pimap)
5023 pimap = kvm->arch.pimap;
5025 for (i = 0; i < pimap->n_mapped; i++) {
5026 if (guest_gsi == pimap->mapped[i].v_hwirq)
5030 if (i == pimap->n_mapped) {
5031 mutex_unlock(&kvm->lock);
5036 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5038 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5040 /* invalidate the entry (what do do on error from the above ?) */
5041 pimap->mapped[i].r_hwirq = 0;
5044 * We don't free this structure even when the count goes to
5045 * zero. The structure is freed when we destroy the VM.
5048 mutex_unlock(&kvm->lock);
5052 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5053 struct irq_bypass_producer *prod)
5056 struct kvm_kernel_irqfd *irqfd =
5057 container_of(cons, struct kvm_kernel_irqfd, consumer);
5059 irqfd->producer = prod;
5061 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5063 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5064 prod->irq, irqfd->gsi, ret);
5069 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5070 struct irq_bypass_producer *prod)
5073 struct kvm_kernel_irqfd *irqfd =
5074 container_of(cons, struct kvm_kernel_irqfd, consumer);
5076 irqfd->producer = NULL;
5079 * When producer of consumer is unregistered, we change back to
5080 * default external interrupt handling mode - KVM real mode
5081 * will switch back to host.
5083 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5085 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5086 prod->irq, irqfd->gsi, ret);
5090 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5091 unsigned int ioctl, unsigned long arg)
5093 struct kvm *kvm __maybe_unused = filp->private_data;
5094 void __user *argp = (void __user *)arg;
5099 case KVM_PPC_ALLOCATE_HTAB: {
5103 if (get_user(htab_order, (u32 __user *)argp))
5105 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5112 case KVM_PPC_GET_HTAB_FD: {
5113 struct kvm_get_htab_fd ghf;
5116 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5118 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5122 case KVM_PPC_RESIZE_HPT_PREPARE: {
5123 struct kvm_ppc_resize_hpt rhpt;
5126 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5129 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5133 case KVM_PPC_RESIZE_HPT_COMMIT: {
5134 struct kvm_ppc_resize_hpt rhpt;
5137 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5140 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5152 * List of hcall numbers to enable by default.
5153 * For compatibility with old userspace, we enable by default
5154 * all hcalls that were implemented before the hcall-enabling
5155 * facility was added. Note this list should not include H_RTAS.
5157 static unsigned int default_hcall_list[] = {
5171 #ifdef CONFIG_KVM_XICS
5182 static void init_default_hcalls(void)
5187 for (i = 0; default_hcall_list[i]; ++i) {
5188 hcall = default_hcall_list[i];
5189 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5190 __set_bit(hcall / 4, default_enabled_hcalls);
5194 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5200 /* If not on a POWER9, reject it */
5201 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5204 /* If any unknown flags set, reject it */
5205 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5208 /* GR (guest radix) bit in process_table field must match */
5209 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5210 if (!!(cfg->process_table & PATB_GR) != radix)
5213 /* Process table size field must be reasonable, i.e. <= 24 */
5214 if ((cfg->process_table & PRTS_MASK) > 24)
5217 /* We can change a guest to/from radix now, if the host is radix */
5218 if (radix && !radix_enabled())
5221 /* If we're a nested hypervisor, we currently only support radix */
5222 if (kvmhv_on_pseries() && !radix)
5225 mutex_lock(&kvm->lock);
5226 if (radix != kvm_is_radix(kvm)) {
5227 if (kvm->arch.mmu_ready) {
5228 kvm->arch.mmu_ready = 0;
5229 /* order mmu_ready vs. vcpus_running */
5231 if (atomic_read(&kvm->arch.vcpus_running)) {
5232 kvm->arch.mmu_ready = 1;
5238 err = kvmppc_switch_mmu_to_radix(kvm);
5240 err = kvmppc_switch_mmu_to_hpt(kvm);
5245 kvm->arch.process_table = cfg->process_table;
5246 kvmppc_setup_partition_table(kvm);
5248 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5249 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5253 mutex_unlock(&kvm->lock);
5257 static int kvmhv_enable_nested(struct kvm *kvm)
5261 if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5264 /* kvm == NULL means the caller is testing if the capability exists */
5266 kvm->arch.nested_enable = true;
5270 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5275 if (kvmhv_vcpu_is_radix(vcpu)) {
5276 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5282 /* For now quadrants are the only way to access nested guest memory */
5283 if (rc && vcpu->arch.nested)
5289 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5294 if (kvmhv_vcpu_is_radix(vcpu)) {
5295 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5301 /* For now quadrants are the only way to access nested guest memory */
5302 if (rc && vcpu->arch.nested)
5308 static struct kvmppc_ops kvm_ops_hv = {
5309 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5310 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5311 .get_one_reg = kvmppc_get_one_reg_hv,
5312 .set_one_reg = kvmppc_set_one_reg_hv,
5313 .vcpu_load = kvmppc_core_vcpu_load_hv,
5314 .vcpu_put = kvmppc_core_vcpu_put_hv,
5315 .set_msr = kvmppc_set_msr_hv,
5316 .vcpu_run = kvmppc_vcpu_run_hv,
5317 .vcpu_create = kvmppc_core_vcpu_create_hv,
5318 .vcpu_free = kvmppc_core_vcpu_free_hv,
5319 .check_requests = kvmppc_core_check_requests_hv,
5320 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
5321 .flush_memslot = kvmppc_core_flush_memslot_hv,
5322 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5323 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
5324 .unmap_hva_range = kvm_unmap_hva_range_hv,
5325 .age_hva = kvm_age_hva_hv,
5326 .test_age_hva = kvm_test_age_hva_hv,
5327 .set_spte_hva = kvm_set_spte_hva_hv,
5328 .mmu_destroy = kvmppc_mmu_destroy_hv,
5329 .free_memslot = kvmppc_core_free_memslot_hv,
5330 .create_memslot = kvmppc_core_create_memslot_hv,
5331 .init_vm = kvmppc_core_init_vm_hv,
5332 .destroy_vm = kvmppc_core_destroy_vm_hv,
5333 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5334 .emulate_op = kvmppc_core_emulate_op_hv,
5335 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5336 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5337 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5338 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
5339 .hcall_implemented = kvmppc_hcall_impl_hv,
5340 #ifdef CONFIG_KVM_XICS
5341 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5342 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5344 .configure_mmu = kvmhv_configure_mmu,
5345 .get_rmmu_info = kvmhv_get_rmmu_info,
5346 .set_smt_mode = kvmhv_set_smt_mode,
5347 .enable_nested = kvmhv_enable_nested,
5348 .load_from_eaddr = kvmhv_load_from_eaddr,
5349 .store_to_eaddr = kvmhv_store_to_eaddr,
5352 static int kvm_init_subcore_bitmap(void)
5355 int nr_cores = cpu_nr_cores();
5356 struct sibling_subcore_state *sibling_subcore_state;
5358 for (i = 0; i < nr_cores; i++) {
5359 int first_cpu = i * threads_per_core;
5360 int node = cpu_to_node(first_cpu);
5362 /* Ignore if it is already allocated. */
5363 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5366 sibling_subcore_state =
5367 kzalloc_node(sizeof(struct sibling_subcore_state),
5369 if (!sibling_subcore_state)
5373 for (j = 0; j < threads_per_core; j++) {
5374 int cpu = first_cpu + j;
5376 paca_ptrs[cpu]->sibling_subcore_state =
5377 sibling_subcore_state;
5383 static int kvmppc_radix_possible(void)
5385 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5388 static int kvmppc_book3s_init_hv(void)
5392 * FIXME!! Do we need to check on all cpus ?
5394 r = kvmppc_core_check_processor_compat_hv();
5398 r = kvmhv_nested_init();
5402 r = kvm_init_subcore_bitmap();
5407 * We need a way of accessing the XICS interrupt controller,
5408 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5409 * indirectly, via OPAL.
5412 if (!xics_on_xive() && !kvmhv_on_pseries() &&
5413 !local_paca->kvm_hstate.xics_phys) {
5414 struct device_node *np;
5416 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5418 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5421 /* presence of intc confirmed - node can be dropped again */
5426 kvm_ops_hv.owner = THIS_MODULE;
5427 kvmppc_hv_ops = &kvm_ops_hv;
5429 init_default_hcalls();
5433 r = kvmppc_mmu_hv_init();
5437 if (kvmppc_radix_possible())
5438 r = kvmppc_radix_init();
5441 * POWER9 chips before version 2.02 can't have some threads in
5442 * HPT mode and some in radix mode on the same core.
5444 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5445 unsigned int pvr = mfspr(SPRN_PVR);
5446 if ((pvr >> 16) == PVR_POWER9 &&
5447 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5448 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5449 no_mixing_hpt_and_radix = true;
5455 static void kvmppc_book3s_exit_hv(void)
5457 kvmppc_free_host_rm_ops();
5458 if (kvmppc_radix_possible())
5459 kvmppc_radix_exit();
5460 kvmppc_hv_ops = NULL;
5461 kvmhv_nested_exit();
5464 module_init(kvmppc_book3s_init_hv);
5465 module_exit(kvmppc_book3s_exit_hv);
5466 MODULE_LICENSE("GPL");
5467 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5468 MODULE_ALIAS("devname:kvm");