3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
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,
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 (xive_enabled()) {
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));
941 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
942 kvmppc_get_gpr(vcpu, 5));
943 if (ret == H_TOO_HARD)
947 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
948 kvmppc_get_gpr(vcpu, 5),
949 kvmppc_get_gpr(vcpu, 6));
950 if (ret == H_TOO_HARD)
953 case H_PUT_TCE_INDIRECT:
954 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
955 kvmppc_get_gpr(vcpu, 5),
956 kvmppc_get_gpr(vcpu, 6),
957 kvmppc_get_gpr(vcpu, 7));
958 if (ret == H_TOO_HARD)
962 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
963 kvmppc_get_gpr(vcpu, 5),
964 kvmppc_get_gpr(vcpu, 6),
965 kvmppc_get_gpr(vcpu, 7));
966 if (ret == H_TOO_HARD)
970 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
974 case H_SET_PARTITION_TABLE:
976 if (nesting_enabled(vcpu->kvm))
977 ret = kvmhv_set_partition_table(vcpu);
981 if (!nesting_enabled(vcpu->kvm))
983 ret = kvmhv_enter_nested_guest(vcpu);
984 if (ret == H_INTERRUPT) {
985 kvmppc_set_gpr(vcpu, 3, 0);
986 vcpu->arch.hcall_needed = 0;
990 case H_TLB_INVALIDATE:
992 if (nesting_enabled(vcpu->kvm))
993 ret = kvmhv_do_nested_tlbie(vcpu);
999 kvmppc_set_gpr(vcpu, 3, ret);
1000 vcpu->arch.hcall_needed = 0;
1001 return RESUME_GUEST;
1005 * Handle H_CEDE in the nested virtualization case where we haven't
1006 * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1007 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1008 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1010 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1012 vcpu->arch.shregs.msr |= MSR_EE;
1013 vcpu->arch.ceded = 1;
1015 if (vcpu->arch.prodded) {
1016 vcpu->arch.prodded = 0;
1018 vcpu->arch.ceded = 0;
1022 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1028 case H_REGISTER_VPA:
1030 case H_LOGICAL_CI_LOAD:
1031 case H_LOGICAL_CI_STORE:
1032 #ifdef CONFIG_KVM_XICS
1043 /* See if it's in the real-mode table */
1044 return kvmppc_hcall_impl_hv_realmode(cmd);
1047 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1048 struct kvm_vcpu *vcpu)
1052 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1055 * Fetch failed, so return to guest and
1056 * try executing it again.
1058 return RESUME_GUEST;
1061 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1062 run->exit_reason = KVM_EXIT_DEBUG;
1063 run->debug.arch.address = kvmppc_get_pc(vcpu);
1066 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1067 return RESUME_GUEST;
1071 static void do_nothing(void *x)
1075 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1077 int thr, cpu, pcpu, nthreads;
1079 unsigned long dpdes;
1081 nthreads = vcpu->kvm->arch.emul_smt_mode;
1083 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1084 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1085 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1089 * If the vcpu is currently running on a physical cpu thread,
1090 * interrupt it in order to pull it out of the guest briefly,
1091 * which will update its vcore->dpdes value.
1093 pcpu = READ_ONCE(v->cpu);
1095 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1096 if (kvmppc_doorbell_pending(v))
1103 * On POWER9, emulate doorbell-related instructions in order to
1104 * give the guest the illusion of running on a multi-threaded core.
1105 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1108 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1112 struct kvm *kvm = vcpu->kvm;
1113 struct kvm_vcpu *tvcpu;
1115 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1116 return RESUME_GUEST;
1117 if (get_op(inst) != 31)
1118 return EMULATE_FAIL;
1120 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1121 switch (get_xop(inst)) {
1122 case OP_31_XOP_MSGSNDP:
1123 arg = kvmppc_get_gpr(vcpu, rb);
1124 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1127 if (arg >= kvm->arch.emul_smt_mode)
1129 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1132 if (!tvcpu->arch.doorbell_request) {
1133 tvcpu->arch.doorbell_request = 1;
1134 kvmppc_fast_vcpu_kick_hv(tvcpu);
1137 case OP_31_XOP_MSGCLRP:
1138 arg = kvmppc_get_gpr(vcpu, rb);
1139 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1141 vcpu->arch.vcore->dpdes = 0;
1142 vcpu->arch.doorbell_request = 0;
1144 case OP_31_XOP_MFSPR:
1145 switch (get_sprn(inst)) {
1150 arg = kvmppc_read_dpdes(vcpu);
1153 return EMULATE_FAIL;
1155 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1158 return EMULATE_FAIL;
1160 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1161 return RESUME_GUEST;
1164 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1165 struct task_struct *tsk)
1167 int r = RESUME_HOST;
1169 vcpu->stat.sum_exits++;
1172 * This can happen if an interrupt occurs in the last stages
1173 * of guest entry or the first stages of guest exit (i.e. after
1174 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1175 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1176 * That can happen due to a bug, or due to a machine check
1177 * occurring at just the wrong time.
1179 if (vcpu->arch.shregs.msr & MSR_HV) {
1180 printk(KERN_EMERG "KVM trap in HV mode!\n");
1181 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1182 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1183 vcpu->arch.shregs.msr);
1184 kvmppc_dump_regs(vcpu);
1185 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1186 run->hw.hardware_exit_reason = vcpu->arch.trap;
1189 run->exit_reason = KVM_EXIT_UNKNOWN;
1190 run->ready_for_interrupt_injection = 1;
1191 switch (vcpu->arch.trap) {
1192 /* We're good on these - the host merely wanted to get our attention */
1193 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1194 vcpu->stat.dec_exits++;
1197 case BOOK3S_INTERRUPT_EXTERNAL:
1198 case BOOK3S_INTERRUPT_H_DOORBELL:
1199 case BOOK3S_INTERRUPT_H_VIRT:
1200 vcpu->stat.ext_intr_exits++;
1203 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1204 case BOOK3S_INTERRUPT_HMI:
1205 case BOOK3S_INTERRUPT_PERFMON:
1206 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1209 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1210 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1211 run->exit_reason = KVM_EXIT_NMI;
1212 run->hw.hardware_exit_reason = vcpu->arch.trap;
1213 /* Clear out the old NMI status from run->flags */
1214 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1215 /* Now set the NMI status */
1216 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1217 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1219 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1222 /* Print the MCE event to host console. */
1223 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1225 case BOOK3S_INTERRUPT_PROGRAM:
1229 * Normally program interrupts are delivered directly
1230 * to the guest by the hardware, but we can get here
1231 * as a result of a hypervisor emulation interrupt
1232 * (e40) getting turned into a 700 by BML RTAS.
1234 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1235 kvmppc_core_queue_program(vcpu, flags);
1239 case BOOK3S_INTERRUPT_SYSCALL:
1241 /* hcall - punt to userspace */
1244 /* hypercall with MSR_PR has already been handled in rmode,
1245 * and never reaches here.
1248 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1249 for (i = 0; i < 9; ++i)
1250 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1251 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1252 vcpu->arch.hcall_needed = 1;
1257 * We get these next two if the guest accesses a page which it thinks
1258 * it has mapped but which is not actually present, either because
1259 * it is for an emulated I/O device or because the corresonding
1260 * host page has been paged out. Any other HDSI/HISI interrupts
1261 * have been handled already.
1263 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1264 r = RESUME_PAGE_FAULT;
1266 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1267 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1268 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1269 DSISR_SRR1_MATCH_64S;
1270 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1271 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1272 r = RESUME_PAGE_FAULT;
1275 * This occurs if the guest executes an illegal instruction.
1276 * If the guest debug is disabled, generate a program interrupt
1277 * to the guest. If guest debug is enabled, we need to check
1278 * whether the instruction is a software breakpoint instruction.
1279 * Accordingly return to Guest or Host.
1281 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1282 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1283 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1284 swab32(vcpu->arch.emul_inst) :
1285 vcpu->arch.emul_inst;
1286 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1287 r = kvmppc_emulate_debug_inst(run, vcpu);
1289 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1294 * This occurs if the guest (kernel or userspace), does something that
1295 * is prohibited by HFSCR.
1296 * On POWER9, this could be a doorbell instruction that we need
1298 * Otherwise, we just generate a program interrupt to the guest.
1300 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1302 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1303 cpu_has_feature(CPU_FTR_ARCH_300))
1304 r = kvmppc_emulate_doorbell_instr(vcpu);
1305 if (r == EMULATE_FAIL) {
1306 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1311 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1312 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1314 * This occurs for various TM-related instructions that
1315 * we need to emulate on POWER9 DD2.2. We have already
1316 * handled the cases where the guest was in real-suspend
1317 * mode and was transitioning to transactional state.
1319 r = kvmhv_p9_tm_emulation(vcpu);
1323 case BOOK3S_INTERRUPT_HV_RM_HARD:
1324 r = RESUME_PASSTHROUGH;
1327 kvmppc_dump_regs(vcpu);
1328 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1329 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1330 vcpu->arch.shregs.msr);
1331 run->hw.hardware_exit_reason = vcpu->arch.trap;
1339 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1344 vcpu->stat.sum_exits++;
1347 * This can happen if an interrupt occurs in the last stages
1348 * of guest entry or the first stages of guest exit (i.e. after
1349 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1350 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1351 * That can happen due to a bug, or due to a machine check
1352 * occurring at just the wrong time.
1354 if (vcpu->arch.shregs.msr & MSR_HV) {
1355 pr_emerg("KVM trap in HV mode while nested!\n");
1356 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1357 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1358 vcpu->arch.shregs.msr);
1359 kvmppc_dump_regs(vcpu);
1362 switch (vcpu->arch.trap) {
1363 /* We're good on these - the host merely wanted to get our attention */
1364 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1365 vcpu->stat.dec_exits++;
1368 case BOOK3S_INTERRUPT_EXTERNAL:
1369 vcpu->stat.ext_intr_exits++;
1372 case BOOK3S_INTERRUPT_H_DOORBELL:
1373 case BOOK3S_INTERRUPT_H_VIRT:
1374 vcpu->stat.ext_intr_exits++;
1377 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1378 case BOOK3S_INTERRUPT_HMI:
1379 case BOOK3S_INTERRUPT_PERFMON:
1380 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1383 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1384 /* Pass the machine check to the L1 guest */
1386 /* Print the MCE event to host console. */
1387 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1390 * We get these next two if the guest accesses a page which it thinks
1391 * it has mapped but which is not actually present, either because
1392 * it is for an emulated I/O device or because the corresonding
1393 * host page has been paged out.
1395 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1396 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1397 r = kvmhv_nested_page_fault(vcpu);
1398 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1400 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1401 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1402 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1403 DSISR_SRR1_MATCH_64S;
1404 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1405 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1406 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1407 r = kvmhv_nested_page_fault(vcpu);
1408 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1411 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1412 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1414 * This occurs for various TM-related instructions that
1415 * we need to emulate on POWER9 DD2.2. We have already
1416 * handled the cases where the guest was in real-suspend
1417 * mode and was transitioning to transactional state.
1419 r = kvmhv_p9_tm_emulation(vcpu);
1423 case BOOK3S_INTERRUPT_HV_RM_HARD:
1424 vcpu->arch.trap = 0;
1426 if (!xive_enabled())
1427 kvmppc_xics_rm_complete(vcpu, 0);
1437 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1438 struct kvm_sregs *sregs)
1442 memset(sregs, 0, sizeof(struct kvm_sregs));
1443 sregs->pvr = vcpu->arch.pvr;
1444 for (i = 0; i < vcpu->arch.slb_max; i++) {
1445 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1446 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1452 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1453 struct kvm_sregs *sregs)
1457 /* Only accept the same PVR as the host's, since we can't spoof it */
1458 if (sregs->pvr != vcpu->arch.pvr)
1462 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1463 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1464 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1465 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1469 vcpu->arch.slb_max = j;
1474 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1475 bool preserve_top32)
1477 struct kvm *kvm = vcpu->kvm;
1478 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1481 mutex_lock(&kvm->lock);
1482 spin_lock(&vc->lock);
1484 * If ILE (interrupt little-endian) has changed, update the
1485 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1487 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1488 struct kvm_vcpu *vcpu;
1491 kvm_for_each_vcpu(i, vcpu, kvm) {
1492 if (vcpu->arch.vcore != vc)
1494 if (new_lpcr & LPCR_ILE)
1495 vcpu->arch.intr_msr |= MSR_LE;
1497 vcpu->arch.intr_msr &= ~MSR_LE;
1502 * Userspace can only modify DPFD (default prefetch depth),
1503 * ILE (interrupt little-endian) and TC (translation control).
1504 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1506 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1507 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1510 * On POWER9, allow userspace to enable large decrementer for the
1511 * guest, whether or not the host has it enabled.
1513 if (cpu_has_feature(CPU_FTR_ARCH_300))
1516 /* Broken 32-bit version of LPCR must not clear top bits */
1519 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1520 spin_unlock(&vc->lock);
1521 mutex_unlock(&kvm->lock);
1524 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1525 union kvmppc_one_reg *val)
1531 case KVM_REG_PPC_DEBUG_INST:
1532 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1534 case KVM_REG_PPC_HIOR:
1535 *val = get_reg_val(id, 0);
1537 case KVM_REG_PPC_DABR:
1538 *val = get_reg_val(id, vcpu->arch.dabr);
1540 case KVM_REG_PPC_DABRX:
1541 *val = get_reg_val(id, vcpu->arch.dabrx);
1543 case KVM_REG_PPC_DSCR:
1544 *val = get_reg_val(id, vcpu->arch.dscr);
1546 case KVM_REG_PPC_PURR:
1547 *val = get_reg_val(id, vcpu->arch.purr);
1549 case KVM_REG_PPC_SPURR:
1550 *val = get_reg_val(id, vcpu->arch.spurr);
1552 case KVM_REG_PPC_AMR:
1553 *val = get_reg_val(id, vcpu->arch.amr);
1555 case KVM_REG_PPC_UAMOR:
1556 *val = get_reg_val(id, vcpu->arch.uamor);
1558 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1559 i = id - KVM_REG_PPC_MMCR0;
1560 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1562 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1563 i = id - KVM_REG_PPC_PMC1;
1564 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1566 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1567 i = id - KVM_REG_PPC_SPMC1;
1568 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1570 case KVM_REG_PPC_SIAR:
1571 *val = get_reg_val(id, vcpu->arch.siar);
1573 case KVM_REG_PPC_SDAR:
1574 *val = get_reg_val(id, vcpu->arch.sdar);
1576 case KVM_REG_PPC_SIER:
1577 *val = get_reg_val(id, vcpu->arch.sier);
1579 case KVM_REG_PPC_IAMR:
1580 *val = get_reg_val(id, vcpu->arch.iamr);
1582 case KVM_REG_PPC_PSPB:
1583 *val = get_reg_val(id, vcpu->arch.pspb);
1585 case KVM_REG_PPC_DPDES:
1586 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1588 case KVM_REG_PPC_VTB:
1589 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1591 case KVM_REG_PPC_DAWR:
1592 *val = get_reg_val(id, vcpu->arch.dawr);
1594 case KVM_REG_PPC_DAWRX:
1595 *val = get_reg_val(id, vcpu->arch.dawrx);
1597 case KVM_REG_PPC_CIABR:
1598 *val = get_reg_val(id, vcpu->arch.ciabr);
1600 case KVM_REG_PPC_CSIGR:
1601 *val = get_reg_val(id, vcpu->arch.csigr);
1603 case KVM_REG_PPC_TACR:
1604 *val = get_reg_val(id, vcpu->arch.tacr);
1606 case KVM_REG_PPC_TCSCR:
1607 *val = get_reg_val(id, vcpu->arch.tcscr);
1609 case KVM_REG_PPC_PID:
1610 *val = get_reg_val(id, vcpu->arch.pid);
1612 case KVM_REG_PPC_ACOP:
1613 *val = get_reg_val(id, vcpu->arch.acop);
1615 case KVM_REG_PPC_WORT:
1616 *val = get_reg_val(id, vcpu->arch.wort);
1618 case KVM_REG_PPC_TIDR:
1619 *val = get_reg_val(id, vcpu->arch.tid);
1621 case KVM_REG_PPC_PSSCR:
1622 *val = get_reg_val(id, vcpu->arch.psscr);
1624 case KVM_REG_PPC_VPA_ADDR:
1625 spin_lock(&vcpu->arch.vpa_update_lock);
1626 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1627 spin_unlock(&vcpu->arch.vpa_update_lock);
1629 case KVM_REG_PPC_VPA_SLB:
1630 spin_lock(&vcpu->arch.vpa_update_lock);
1631 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1632 val->vpaval.length = vcpu->arch.slb_shadow.len;
1633 spin_unlock(&vcpu->arch.vpa_update_lock);
1635 case KVM_REG_PPC_VPA_DTL:
1636 spin_lock(&vcpu->arch.vpa_update_lock);
1637 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1638 val->vpaval.length = vcpu->arch.dtl.len;
1639 spin_unlock(&vcpu->arch.vpa_update_lock);
1641 case KVM_REG_PPC_TB_OFFSET:
1642 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1644 case KVM_REG_PPC_LPCR:
1645 case KVM_REG_PPC_LPCR_64:
1646 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1648 case KVM_REG_PPC_PPR:
1649 *val = get_reg_val(id, vcpu->arch.ppr);
1651 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1652 case KVM_REG_PPC_TFHAR:
1653 *val = get_reg_val(id, vcpu->arch.tfhar);
1655 case KVM_REG_PPC_TFIAR:
1656 *val = get_reg_val(id, vcpu->arch.tfiar);
1658 case KVM_REG_PPC_TEXASR:
1659 *val = get_reg_val(id, vcpu->arch.texasr);
1661 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1662 i = id - KVM_REG_PPC_TM_GPR0;
1663 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1665 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1668 i = id - KVM_REG_PPC_TM_VSR0;
1670 for (j = 0; j < TS_FPRWIDTH; j++)
1671 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1673 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1674 val->vval = vcpu->arch.vr_tm.vr[i-32];
1680 case KVM_REG_PPC_TM_CR:
1681 *val = get_reg_val(id, vcpu->arch.cr_tm);
1683 case KVM_REG_PPC_TM_XER:
1684 *val = get_reg_val(id, vcpu->arch.xer_tm);
1686 case KVM_REG_PPC_TM_LR:
1687 *val = get_reg_val(id, vcpu->arch.lr_tm);
1689 case KVM_REG_PPC_TM_CTR:
1690 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1692 case KVM_REG_PPC_TM_FPSCR:
1693 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1695 case KVM_REG_PPC_TM_AMR:
1696 *val = get_reg_val(id, vcpu->arch.amr_tm);
1698 case KVM_REG_PPC_TM_PPR:
1699 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1701 case KVM_REG_PPC_TM_VRSAVE:
1702 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1704 case KVM_REG_PPC_TM_VSCR:
1705 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1706 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1710 case KVM_REG_PPC_TM_DSCR:
1711 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1713 case KVM_REG_PPC_TM_TAR:
1714 *val = get_reg_val(id, vcpu->arch.tar_tm);
1717 case KVM_REG_PPC_ARCH_COMPAT:
1718 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1720 case KVM_REG_PPC_DEC_EXPIRY:
1721 *val = get_reg_val(id, vcpu->arch.dec_expires +
1722 vcpu->arch.vcore->tb_offset);
1724 case KVM_REG_PPC_ONLINE:
1725 *val = get_reg_val(id, vcpu->arch.online);
1727 case KVM_REG_PPC_PTCR:
1728 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1738 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1739 union kvmppc_one_reg *val)
1743 unsigned long addr, len;
1746 case KVM_REG_PPC_HIOR:
1747 /* Only allow this to be set to zero */
1748 if (set_reg_val(id, *val))
1751 case KVM_REG_PPC_DABR:
1752 vcpu->arch.dabr = set_reg_val(id, *val);
1754 case KVM_REG_PPC_DABRX:
1755 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1757 case KVM_REG_PPC_DSCR:
1758 vcpu->arch.dscr = set_reg_val(id, *val);
1760 case KVM_REG_PPC_PURR:
1761 vcpu->arch.purr = set_reg_val(id, *val);
1763 case KVM_REG_PPC_SPURR:
1764 vcpu->arch.spurr = set_reg_val(id, *val);
1766 case KVM_REG_PPC_AMR:
1767 vcpu->arch.amr = set_reg_val(id, *val);
1769 case KVM_REG_PPC_UAMOR:
1770 vcpu->arch.uamor = set_reg_val(id, *val);
1772 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1773 i = id - KVM_REG_PPC_MMCR0;
1774 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1776 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1777 i = id - KVM_REG_PPC_PMC1;
1778 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1780 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1781 i = id - KVM_REG_PPC_SPMC1;
1782 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1784 case KVM_REG_PPC_SIAR:
1785 vcpu->arch.siar = set_reg_val(id, *val);
1787 case KVM_REG_PPC_SDAR:
1788 vcpu->arch.sdar = set_reg_val(id, *val);
1790 case KVM_REG_PPC_SIER:
1791 vcpu->arch.sier = set_reg_val(id, *val);
1793 case KVM_REG_PPC_IAMR:
1794 vcpu->arch.iamr = set_reg_val(id, *val);
1796 case KVM_REG_PPC_PSPB:
1797 vcpu->arch.pspb = set_reg_val(id, *val);
1799 case KVM_REG_PPC_DPDES:
1800 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1802 case KVM_REG_PPC_VTB:
1803 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1805 case KVM_REG_PPC_DAWR:
1806 vcpu->arch.dawr = set_reg_val(id, *val);
1808 case KVM_REG_PPC_DAWRX:
1809 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1811 case KVM_REG_PPC_CIABR:
1812 vcpu->arch.ciabr = set_reg_val(id, *val);
1813 /* Don't allow setting breakpoints in hypervisor code */
1814 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1815 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1817 case KVM_REG_PPC_CSIGR:
1818 vcpu->arch.csigr = set_reg_val(id, *val);
1820 case KVM_REG_PPC_TACR:
1821 vcpu->arch.tacr = set_reg_val(id, *val);
1823 case KVM_REG_PPC_TCSCR:
1824 vcpu->arch.tcscr = set_reg_val(id, *val);
1826 case KVM_REG_PPC_PID:
1827 vcpu->arch.pid = set_reg_val(id, *val);
1829 case KVM_REG_PPC_ACOP:
1830 vcpu->arch.acop = set_reg_val(id, *val);
1832 case KVM_REG_PPC_WORT:
1833 vcpu->arch.wort = set_reg_val(id, *val);
1835 case KVM_REG_PPC_TIDR:
1836 vcpu->arch.tid = set_reg_val(id, *val);
1838 case KVM_REG_PPC_PSSCR:
1839 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1841 case KVM_REG_PPC_VPA_ADDR:
1842 addr = set_reg_val(id, *val);
1844 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1845 vcpu->arch.dtl.next_gpa))
1847 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1849 case KVM_REG_PPC_VPA_SLB:
1850 addr = val->vpaval.addr;
1851 len = val->vpaval.length;
1853 if (addr && !vcpu->arch.vpa.next_gpa)
1855 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1857 case KVM_REG_PPC_VPA_DTL:
1858 addr = val->vpaval.addr;
1859 len = val->vpaval.length;
1861 if (addr && (len < sizeof(struct dtl_entry) ||
1862 !vcpu->arch.vpa.next_gpa))
1864 len -= len % sizeof(struct dtl_entry);
1865 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1867 case KVM_REG_PPC_TB_OFFSET:
1868 /* round up to multiple of 2^24 */
1869 vcpu->arch.vcore->tb_offset =
1870 ALIGN(set_reg_val(id, *val), 1UL << 24);
1872 case KVM_REG_PPC_LPCR:
1873 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1875 case KVM_REG_PPC_LPCR_64:
1876 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1878 case KVM_REG_PPC_PPR:
1879 vcpu->arch.ppr = set_reg_val(id, *val);
1881 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1882 case KVM_REG_PPC_TFHAR:
1883 vcpu->arch.tfhar = set_reg_val(id, *val);
1885 case KVM_REG_PPC_TFIAR:
1886 vcpu->arch.tfiar = set_reg_val(id, *val);
1888 case KVM_REG_PPC_TEXASR:
1889 vcpu->arch.texasr = set_reg_val(id, *val);
1891 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1892 i = id - KVM_REG_PPC_TM_GPR0;
1893 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1895 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1898 i = id - KVM_REG_PPC_TM_VSR0;
1900 for (j = 0; j < TS_FPRWIDTH; j++)
1901 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1903 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1904 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1909 case KVM_REG_PPC_TM_CR:
1910 vcpu->arch.cr_tm = set_reg_val(id, *val);
1912 case KVM_REG_PPC_TM_XER:
1913 vcpu->arch.xer_tm = set_reg_val(id, *val);
1915 case KVM_REG_PPC_TM_LR:
1916 vcpu->arch.lr_tm = set_reg_val(id, *val);
1918 case KVM_REG_PPC_TM_CTR:
1919 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1921 case KVM_REG_PPC_TM_FPSCR:
1922 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1924 case KVM_REG_PPC_TM_AMR:
1925 vcpu->arch.amr_tm = set_reg_val(id, *val);
1927 case KVM_REG_PPC_TM_PPR:
1928 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1930 case KVM_REG_PPC_TM_VRSAVE:
1931 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1933 case KVM_REG_PPC_TM_VSCR:
1934 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1935 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1939 case KVM_REG_PPC_TM_DSCR:
1940 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1942 case KVM_REG_PPC_TM_TAR:
1943 vcpu->arch.tar_tm = set_reg_val(id, *val);
1946 case KVM_REG_PPC_ARCH_COMPAT:
1947 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1949 case KVM_REG_PPC_DEC_EXPIRY:
1950 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1951 vcpu->arch.vcore->tb_offset;
1953 case KVM_REG_PPC_ONLINE:
1954 i = set_reg_val(id, *val);
1955 if (i && !vcpu->arch.online)
1956 atomic_inc(&vcpu->arch.vcore->online_count);
1957 else if (!i && vcpu->arch.online)
1958 atomic_dec(&vcpu->arch.vcore->online_count);
1959 vcpu->arch.online = i;
1961 case KVM_REG_PPC_PTCR:
1962 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
1973 * On POWER9, threads are independent and can be in different partitions.
1974 * Therefore we consider each thread to be a subcore.
1975 * There is a restriction that all threads have to be in the same
1976 * MMU mode (radix or HPT), unfortunately, but since we only support
1977 * HPT guests on a HPT host so far, that isn't an impediment yet.
1979 static int threads_per_vcore(struct kvm *kvm)
1981 if (kvm->arch.threads_indep)
1983 return threads_per_subcore;
1986 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
1988 struct kvmppc_vcore *vcore;
1990 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1995 spin_lock_init(&vcore->lock);
1996 spin_lock_init(&vcore->stoltb_lock);
1997 init_swait_queue_head(&vcore->wq);
1998 vcore->preempt_tb = TB_NIL;
1999 vcore->lpcr = kvm->arch.lpcr;
2000 vcore->first_vcpuid = id;
2002 INIT_LIST_HEAD(&vcore->preempt_list);
2007 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2008 static struct debugfs_timings_element {
2012 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2013 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2014 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2015 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2016 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2019 #define N_TIMINGS (ARRAY_SIZE(timings))
2021 struct debugfs_timings_state {
2022 struct kvm_vcpu *vcpu;
2023 unsigned int buflen;
2024 char buf[N_TIMINGS * 100];
2027 static int debugfs_timings_open(struct inode *inode, struct file *file)
2029 struct kvm_vcpu *vcpu = inode->i_private;
2030 struct debugfs_timings_state *p;
2032 p = kzalloc(sizeof(*p), GFP_KERNEL);
2036 kvm_get_kvm(vcpu->kvm);
2038 file->private_data = p;
2040 return nonseekable_open(inode, file);
2043 static int debugfs_timings_release(struct inode *inode, struct file *file)
2045 struct debugfs_timings_state *p = file->private_data;
2047 kvm_put_kvm(p->vcpu->kvm);
2052 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2053 size_t len, loff_t *ppos)
2055 struct debugfs_timings_state *p = file->private_data;
2056 struct kvm_vcpu *vcpu = p->vcpu;
2058 struct kvmhv_tb_accumulator tb;
2067 buf_end = s + sizeof(p->buf);
2068 for (i = 0; i < N_TIMINGS; ++i) {
2069 struct kvmhv_tb_accumulator *acc;
2071 acc = (struct kvmhv_tb_accumulator *)
2072 ((unsigned long)vcpu + timings[i].offset);
2074 for (loops = 0; loops < 1000; ++loops) {
2075 count = acc->seqcount;
2080 if (count == acc->seqcount) {
2088 snprintf(s, buf_end - s, "%s: stuck\n",
2091 snprintf(s, buf_end - s,
2092 "%s: %llu %llu %llu %llu\n",
2093 timings[i].name, count / 2,
2094 tb_to_ns(tb.tb_total),
2095 tb_to_ns(tb.tb_min),
2096 tb_to_ns(tb.tb_max));
2099 p->buflen = s - p->buf;
2103 if (pos >= p->buflen)
2105 if (len > p->buflen - pos)
2106 len = p->buflen - pos;
2107 n = copy_to_user(buf, p->buf + pos, len);
2117 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2118 size_t len, loff_t *ppos)
2123 static const struct file_operations debugfs_timings_ops = {
2124 .owner = THIS_MODULE,
2125 .open = debugfs_timings_open,
2126 .release = debugfs_timings_release,
2127 .read = debugfs_timings_read,
2128 .write = debugfs_timings_write,
2129 .llseek = generic_file_llseek,
2132 /* Create a debugfs directory for the vcpu */
2133 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2136 struct kvm *kvm = vcpu->kvm;
2138 snprintf(buf, sizeof(buf), "vcpu%u", id);
2139 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2141 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2142 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2144 vcpu->arch.debugfs_timings =
2145 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2146 vcpu, &debugfs_timings_ops);
2149 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2150 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2153 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2155 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2158 struct kvm_vcpu *vcpu;
2161 struct kvmppc_vcore *vcore;
2164 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2168 err = kvm_vcpu_init(vcpu, kvm, id);
2172 vcpu->arch.shared = &vcpu->arch.shregs;
2173 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2175 * The shared struct is never shared on HV,
2176 * so we can always use host endianness
2178 #ifdef __BIG_ENDIAN__
2179 vcpu->arch.shared_big_endian = true;
2181 vcpu->arch.shared_big_endian = false;
2184 vcpu->arch.mmcr[0] = MMCR0_FC;
2185 vcpu->arch.ctrl = CTRL_RUNLATCH;
2186 /* default to host PVR, since we can't spoof it */
2187 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2188 spin_lock_init(&vcpu->arch.vpa_update_lock);
2189 spin_lock_init(&vcpu->arch.tbacct_lock);
2190 vcpu->arch.busy_preempt = TB_NIL;
2191 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2194 * Set the default HFSCR for the guest from the host value.
2195 * This value is only used on POWER9.
2196 * On POWER9, we want to virtualize the doorbell facility, so we
2197 * don't set the HFSCR_MSGP bit, and that causes those instructions
2198 * to trap and then we emulate them.
2200 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2201 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2202 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2203 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2204 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2205 vcpu->arch.hfscr |= HFSCR_TM;
2207 if (cpu_has_feature(CPU_FTR_TM_COMP))
2208 vcpu->arch.hfscr |= HFSCR_TM;
2210 kvmppc_mmu_book3s_hv_init(vcpu);
2212 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2214 init_waitqueue_head(&vcpu->arch.cpu_run);
2216 mutex_lock(&kvm->lock);
2219 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2220 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2221 pr_devel("KVM: VCPU ID too high\n");
2222 core = KVM_MAX_VCORES;
2224 BUG_ON(kvm->arch.smt_mode != 1);
2225 core = kvmppc_pack_vcpu_id(kvm, id);
2228 core = id / kvm->arch.smt_mode;
2230 if (core < KVM_MAX_VCORES) {
2231 vcore = kvm->arch.vcores[core];
2232 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2233 pr_devel("KVM: collision on id %u", id);
2235 } else if (!vcore) {
2237 vcore = kvmppc_vcore_create(kvm,
2238 id & ~(kvm->arch.smt_mode - 1));
2239 kvm->arch.vcores[core] = vcore;
2240 kvm->arch.online_vcores++;
2243 mutex_unlock(&kvm->lock);
2248 spin_lock(&vcore->lock);
2249 ++vcore->num_threads;
2250 spin_unlock(&vcore->lock);
2251 vcpu->arch.vcore = vcore;
2252 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2253 vcpu->arch.thread_cpu = -1;
2254 vcpu->arch.prev_cpu = -1;
2256 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2257 kvmppc_sanity_check(vcpu);
2259 debugfs_vcpu_init(vcpu, id);
2264 kmem_cache_free(kvm_vcpu_cache, vcpu);
2266 return ERR_PTR(err);
2269 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2270 unsigned long flags)
2277 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2279 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2281 * On POWER8 (or POWER7), the threading mode is "strict",
2282 * so we pack smt_mode vcpus per vcore.
2284 if (smt_mode > threads_per_subcore)
2288 * On POWER9, the threading mode is "loose",
2289 * so each vcpu gets its own vcore.
2294 mutex_lock(&kvm->lock);
2296 if (!kvm->arch.online_vcores) {
2297 kvm->arch.smt_mode = smt_mode;
2298 kvm->arch.emul_smt_mode = esmt;
2301 mutex_unlock(&kvm->lock);
2306 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2308 if (vpa->pinned_addr)
2309 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2313 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2315 spin_lock(&vcpu->arch.vpa_update_lock);
2316 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2317 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2318 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2319 spin_unlock(&vcpu->arch.vpa_update_lock);
2320 kvm_vcpu_uninit(vcpu);
2321 kmem_cache_free(kvm_vcpu_cache, vcpu);
2324 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2326 /* Indicate we want to get back into the guest */
2330 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2332 unsigned long dec_nsec, now;
2335 if (now > vcpu->arch.dec_expires) {
2336 /* decrementer has already gone negative */
2337 kvmppc_core_queue_dec(vcpu);
2338 kvmppc_core_prepare_to_enter(vcpu);
2341 dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2342 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2343 vcpu->arch.timer_running = 1;
2346 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2348 vcpu->arch.ceded = 0;
2349 if (vcpu->arch.timer_running) {
2350 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2351 vcpu->arch.timer_running = 0;
2355 extern int __kvmppc_vcore_entry(void);
2357 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2358 struct kvm_vcpu *vcpu)
2362 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2364 spin_lock_irq(&vcpu->arch.tbacct_lock);
2366 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2367 vcpu->arch.stolen_logged;
2368 vcpu->arch.busy_preempt = now;
2369 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2370 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2372 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2375 static int kvmppc_grab_hwthread(int cpu)
2377 struct paca_struct *tpaca;
2378 long timeout = 10000;
2380 tpaca = paca_ptrs[cpu];
2382 /* Ensure the thread won't go into the kernel if it wakes */
2383 tpaca->kvm_hstate.kvm_vcpu = NULL;
2384 tpaca->kvm_hstate.kvm_vcore = NULL;
2385 tpaca->kvm_hstate.napping = 0;
2387 tpaca->kvm_hstate.hwthread_req = 1;
2390 * If the thread is already executing in the kernel (e.g. handling
2391 * a stray interrupt), wait for it to get back to nap mode.
2392 * The smp_mb() is to ensure that our setting of hwthread_req
2393 * is visible before we look at hwthread_state, so if this
2394 * races with the code at system_reset_pSeries and the thread
2395 * misses our setting of hwthread_req, we are sure to see its
2396 * setting of hwthread_state, and vice versa.
2399 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2400 if (--timeout <= 0) {
2401 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2409 static void kvmppc_release_hwthread(int cpu)
2411 struct paca_struct *tpaca;
2413 tpaca = paca_ptrs[cpu];
2414 tpaca->kvm_hstate.hwthread_req = 0;
2415 tpaca->kvm_hstate.kvm_vcpu = NULL;
2416 tpaca->kvm_hstate.kvm_vcore = NULL;
2417 tpaca->kvm_hstate.kvm_split_mode = NULL;
2420 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2422 struct kvm_nested_guest *nested = vcpu->arch.nested;
2423 cpumask_t *cpu_in_guest;
2426 cpu = cpu_first_thread_sibling(cpu);
2428 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2429 cpu_in_guest = &nested->cpu_in_guest;
2431 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2432 cpu_in_guest = &kvm->arch.cpu_in_guest;
2435 * Make sure setting of bit in need_tlb_flush precedes
2436 * testing of cpu_in_guest bits. The matching barrier on
2437 * the other side is the first smp_mb() in kvmppc_run_core().
2440 for (i = 0; i < threads_per_core; ++i)
2441 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2442 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2445 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2447 struct kvm_nested_guest *nested = vcpu->arch.nested;
2448 struct kvm *kvm = vcpu->kvm;
2451 if (!cpu_has_feature(CPU_FTR_HVMODE))
2455 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2457 prev_cpu = vcpu->arch.prev_cpu;
2460 * With radix, the guest can do TLB invalidations itself,
2461 * and it could choose to use the local form (tlbiel) if
2462 * it is invalidating a translation that has only ever been
2463 * used on one vcpu. However, that doesn't mean it has
2464 * only ever been used on one physical cpu, since vcpus
2465 * can move around between pcpus. To cope with this, when
2466 * a vcpu moves from one pcpu to another, we need to tell
2467 * any vcpus running on the same core as this vcpu previously
2468 * ran to flush the TLB. The TLB is shared between threads,
2469 * so we use a single bit in .need_tlb_flush for all 4 threads.
2471 if (prev_cpu != pcpu) {
2472 if (prev_cpu >= 0 &&
2473 cpu_first_thread_sibling(prev_cpu) !=
2474 cpu_first_thread_sibling(pcpu))
2475 radix_flush_cpu(kvm, prev_cpu, vcpu);
2477 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2479 vcpu->arch.prev_cpu = pcpu;
2483 static void kvmppc_radix_check_need_tlb_flush(struct kvm *kvm, int pcpu,
2484 struct kvm_nested_guest *nested)
2486 cpumask_t *need_tlb_flush;
2489 if (!cpu_has_feature(CPU_FTR_HVMODE))
2492 if (cpu_has_feature(CPU_FTR_ARCH_300))
2496 lpid = nested->shadow_lpid;
2497 need_tlb_flush = &nested->need_tlb_flush;
2499 lpid = kvm->arch.lpid;
2500 need_tlb_flush = &kvm->arch.need_tlb_flush;
2503 mtspr(SPRN_LPID, lpid);
2507 if (cpumask_test_cpu(pcpu, need_tlb_flush)) {
2508 radix__local_flush_tlb_lpid_guest(lpid);
2509 /* Clear the bit after the TLB flush */
2510 cpumask_clear_cpu(pcpu, need_tlb_flush);
2514 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2517 struct paca_struct *tpaca;
2518 struct kvm *kvm = vc->kvm;
2522 if (vcpu->arch.timer_running) {
2523 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2524 vcpu->arch.timer_running = 0;
2526 cpu += vcpu->arch.ptid;
2527 vcpu->cpu = vc->pcpu;
2528 vcpu->arch.thread_cpu = cpu;
2529 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2531 tpaca = paca_ptrs[cpu];
2532 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2533 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2534 tpaca->kvm_hstate.fake_suspend = 0;
2535 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2537 tpaca->kvm_hstate.kvm_vcore = vc;
2538 if (cpu != smp_processor_id())
2539 kvmppc_ipi_thread(cpu);
2542 static void kvmppc_wait_for_nap(int n_threads)
2544 int cpu = smp_processor_id();
2549 for (loops = 0; loops < 1000000; ++loops) {
2551 * Check if all threads are finished.
2552 * We set the vcore pointer when starting a thread
2553 * and the thread clears it when finished, so we look
2554 * for any threads that still have a non-NULL vcore ptr.
2556 for (i = 1; i < n_threads; ++i)
2557 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2559 if (i == n_threads) {
2566 for (i = 1; i < n_threads; ++i)
2567 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2568 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2572 * Check that we are on thread 0 and that any other threads in
2573 * this core are off-line. Then grab the threads so they can't
2576 static int on_primary_thread(void)
2578 int cpu = smp_processor_id();
2581 /* Are we on a primary subcore? */
2582 if (cpu_thread_in_subcore(cpu))
2586 while (++thr < threads_per_subcore)
2587 if (cpu_online(cpu + thr))
2590 /* Grab all hw threads so they can't go into the kernel */
2591 for (thr = 1; thr < threads_per_subcore; ++thr) {
2592 if (kvmppc_grab_hwthread(cpu + thr)) {
2593 /* Couldn't grab one; let the others go */
2595 kvmppc_release_hwthread(cpu + thr);
2596 } while (--thr > 0);
2604 * A list of virtual cores for each physical CPU.
2605 * These are vcores that could run but their runner VCPU tasks are
2606 * (or may be) preempted.
2608 struct preempted_vcore_list {
2609 struct list_head list;
2613 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2615 static void init_vcore_lists(void)
2619 for_each_possible_cpu(cpu) {
2620 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2621 spin_lock_init(&lp->lock);
2622 INIT_LIST_HEAD(&lp->list);
2626 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2628 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2630 vc->vcore_state = VCORE_PREEMPT;
2631 vc->pcpu = smp_processor_id();
2632 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2633 spin_lock(&lp->lock);
2634 list_add_tail(&vc->preempt_list, &lp->list);
2635 spin_unlock(&lp->lock);
2638 /* Start accumulating stolen time */
2639 kvmppc_core_start_stolen(vc);
2642 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2644 struct preempted_vcore_list *lp;
2646 kvmppc_core_end_stolen(vc);
2647 if (!list_empty(&vc->preempt_list)) {
2648 lp = &per_cpu(preempted_vcores, vc->pcpu);
2649 spin_lock(&lp->lock);
2650 list_del_init(&vc->preempt_list);
2651 spin_unlock(&lp->lock);
2653 vc->vcore_state = VCORE_INACTIVE;
2657 * This stores information about the virtual cores currently
2658 * assigned to a physical core.
2662 int max_subcore_threads;
2664 int subcore_threads[MAX_SUBCORES];
2665 struct kvmppc_vcore *vc[MAX_SUBCORES];
2669 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2670 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2672 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2674 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2676 memset(cip, 0, sizeof(*cip));
2677 cip->n_subcores = 1;
2678 cip->max_subcore_threads = vc->num_threads;
2679 cip->total_threads = vc->num_threads;
2680 cip->subcore_threads[0] = vc->num_threads;
2684 static bool subcore_config_ok(int n_subcores, int n_threads)
2687 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2688 * split-core mode, with one thread per subcore.
2690 if (cpu_has_feature(CPU_FTR_ARCH_300))
2691 return n_subcores <= 4 && n_threads == 1;
2693 /* On POWER8, can only dynamically split if unsplit to begin with */
2694 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2696 if (n_subcores > MAX_SUBCORES)
2698 if (n_subcores > 1) {
2699 if (!(dynamic_mt_modes & 2))
2701 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2705 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2708 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2710 vc->entry_exit_map = 0;
2712 vc->napping_threads = 0;
2713 vc->conferring_threads = 0;
2714 vc->tb_offset_applied = 0;
2717 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2719 int n_threads = vc->num_threads;
2722 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2725 /* In one_vm_per_core mode, require all vcores to be from the same vm */
2726 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2729 /* Some POWER9 chips require all threads to be in the same MMU mode */
2730 if (no_mixing_hpt_and_radix &&
2731 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2734 if (n_threads < cip->max_subcore_threads)
2735 n_threads = cip->max_subcore_threads;
2736 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2738 cip->max_subcore_threads = n_threads;
2740 sub = cip->n_subcores;
2742 cip->total_threads += vc->num_threads;
2743 cip->subcore_threads[sub] = vc->num_threads;
2745 init_vcore_to_run(vc);
2746 list_del_init(&vc->preempt_list);
2752 * Work out whether it is possible to piggyback the execution of
2753 * vcore *pvc onto the execution of the other vcores described in *cip.
2755 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2758 if (cip->total_threads + pvc->num_threads > target_threads)
2761 return can_dynamic_split(pvc, cip);
2764 static void prepare_threads(struct kvmppc_vcore *vc)
2767 struct kvm_vcpu *vcpu;
2769 for_each_runnable_thread(i, vcpu, vc) {
2770 if (signal_pending(vcpu->arch.run_task))
2771 vcpu->arch.ret = -EINTR;
2772 else if (vcpu->arch.vpa.update_pending ||
2773 vcpu->arch.slb_shadow.update_pending ||
2774 vcpu->arch.dtl.update_pending)
2775 vcpu->arch.ret = RESUME_GUEST;
2778 kvmppc_remove_runnable(vc, vcpu);
2779 wake_up(&vcpu->arch.cpu_run);
2783 static void collect_piggybacks(struct core_info *cip, int target_threads)
2785 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2786 struct kvmppc_vcore *pvc, *vcnext;
2788 spin_lock(&lp->lock);
2789 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2790 if (!spin_trylock(&pvc->lock))
2792 prepare_threads(pvc);
2793 if (!pvc->n_runnable) {
2794 list_del_init(&pvc->preempt_list);
2795 if (pvc->runner == NULL) {
2796 pvc->vcore_state = VCORE_INACTIVE;
2797 kvmppc_core_end_stolen(pvc);
2799 spin_unlock(&pvc->lock);
2802 if (!can_piggyback(pvc, cip, target_threads)) {
2803 spin_unlock(&pvc->lock);
2806 kvmppc_core_end_stolen(pvc);
2807 pvc->vcore_state = VCORE_PIGGYBACK;
2808 if (cip->total_threads >= target_threads)
2811 spin_unlock(&lp->lock);
2814 static bool recheck_signals(struct core_info *cip)
2817 struct kvm_vcpu *vcpu;
2819 for (sub = 0; sub < cip->n_subcores; ++sub)
2820 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2821 if (signal_pending(vcpu->arch.run_task))
2826 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2828 int still_running = 0, i;
2831 struct kvm_vcpu *vcpu;
2833 spin_lock(&vc->lock);
2835 for_each_runnable_thread(i, vcpu, vc) {
2837 * It's safe to unlock the vcore in the loop here, because
2838 * for_each_runnable_thread() is safe against removal of
2839 * the vcpu, and the vcore state is VCORE_EXITING here,
2840 * so any vcpus becoming runnable will have their arch.trap
2841 * set to zero and can't actually run in the guest.
2843 spin_unlock(&vc->lock);
2844 /* cancel pending dec exception if dec is positive */
2845 if (now < vcpu->arch.dec_expires &&
2846 kvmppc_core_pending_dec(vcpu))
2847 kvmppc_core_dequeue_dec(vcpu);
2849 trace_kvm_guest_exit(vcpu);
2852 if (vcpu->arch.trap)
2853 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2854 vcpu->arch.run_task);
2856 vcpu->arch.ret = ret;
2857 vcpu->arch.trap = 0;
2859 spin_lock(&vc->lock);
2860 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2861 if (vcpu->arch.pending_exceptions)
2862 kvmppc_core_prepare_to_enter(vcpu);
2863 if (vcpu->arch.ceded)
2864 kvmppc_set_timer(vcpu);
2868 kvmppc_remove_runnable(vc, vcpu);
2869 wake_up(&vcpu->arch.cpu_run);
2873 if (still_running > 0) {
2874 kvmppc_vcore_preempt(vc);
2875 } else if (vc->runner) {
2876 vc->vcore_state = VCORE_PREEMPT;
2877 kvmppc_core_start_stolen(vc);
2879 vc->vcore_state = VCORE_INACTIVE;
2881 if (vc->n_runnable > 0 && vc->runner == NULL) {
2882 /* make sure there's a candidate runner awake */
2884 vcpu = next_runnable_thread(vc, &i);
2885 wake_up(&vcpu->arch.cpu_run);
2888 spin_unlock(&vc->lock);
2892 * Clear core from the list of active host cores as we are about to
2893 * enter the guest. Only do this if it is the primary thread of the
2894 * core (not if a subcore) that is entering the guest.
2896 static inline int kvmppc_clear_host_core(unsigned int cpu)
2900 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2903 * Memory barrier can be omitted here as we will do a smp_wmb()
2904 * later in kvmppc_start_thread and we need ensure that state is
2905 * visible to other CPUs only after we enter guest.
2907 core = cpu >> threads_shift;
2908 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2913 * Advertise this core as an active host core since we exited the guest
2914 * Only need to do this if it is the primary thread of the core that is
2917 static inline int kvmppc_set_host_core(unsigned int cpu)
2921 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2925 * Memory barrier can be omitted here because we do a spin_unlock
2926 * immediately after this which provides the memory barrier.
2928 core = cpu >> threads_shift;
2929 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2933 static void set_irq_happened(int trap)
2936 case BOOK3S_INTERRUPT_EXTERNAL:
2937 local_paca->irq_happened |= PACA_IRQ_EE;
2939 case BOOK3S_INTERRUPT_H_DOORBELL:
2940 local_paca->irq_happened |= PACA_IRQ_DBELL;
2942 case BOOK3S_INTERRUPT_HMI:
2943 local_paca->irq_happened |= PACA_IRQ_HMI;
2945 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2946 replay_system_reset();
2952 * Run a set of guest threads on a physical core.
2953 * Called with vc->lock held.
2955 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2957 struct kvm_vcpu *vcpu;
2960 struct core_info core_info;
2961 struct kvmppc_vcore *pvc;
2962 struct kvm_split_mode split_info, *sip;
2963 int split, subcore_size, active;
2966 unsigned long cmd_bit, stat_bit;
2969 int controlled_threads;
2975 * Remove from the list any threads that have a signal pending
2976 * or need a VPA update done
2978 prepare_threads(vc);
2980 /* if the runner is no longer runnable, let the caller pick a new one */
2981 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2987 init_vcore_to_run(vc);
2988 vc->preempt_tb = TB_NIL;
2991 * Number of threads that we will be controlling: the same as
2992 * the number of threads per subcore, except on POWER9,
2993 * where it's 1 because the threads are (mostly) independent.
2995 controlled_threads = threads_per_vcore(vc->kvm);
2998 * Make sure we are running on primary threads, and that secondary
2999 * threads are offline. Also check if the number of threads in this
3000 * guest are greater than the current system threads per guest.
3001 * On POWER9, we need to be not in independent-threads mode if
3002 * this is a HPT guest on a radix host machine where the
3003 * CPU threads may not be in different MMU modes.
3005 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3006 !kvm_is_radix(vc->kvm);
3007 if (((controlled_threads > 1) &&
3008 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3009 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3010 for_each_runnable_thread(i, vcpu, vc) {
3011 vcpu->arch.ret = -EBUSY;
3012 kvmppc_remove_runnable(vc, vcpu);
3013 wake_up(&vcpu->arch.cpu_run);
3019 * See if we could run any other vcores on the physical core
3020 * along with this one.
3022 init_core_info(&core_info, vc);
3023 pcpu = smp_processor_id();
3024 target_threads = controlled_threads;
3025 if (target_smt_mode && target_smt_mode < target_threads)
3026 target_threads = target_smt_mode;
3027 if (vc->num_threads < target_threads)
3028 collect_piggybacks(&core_info, target_threads);
3031 * On radix, arrange for TLB flushing if necessary.
3032 * This has to be done before disabling interrupts since
3033 * it uses smp_call_function().
3035 pcpu = smp_processor_id();
3036 if (kvm_is_radix(vc->kvm)) {
3037 for (sub = 0; sub < core_info.n_subcores; ++sub)
3038 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3039 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3043 * Hard-disable interrupts, and check resched flag and signals.
3044 * If we need to reschedule or deliver a signal, clean up
3045 * and return without going into the guest(s).
3046 * If the mmu_ready flag has been cleared, don't go into the
3047 * guest because that means a HPT resize operation is in progress.
3049 local_irq_disable();
3051 if (lazy_irq_pending() || need_resched() ||
3052 recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3054 vc->vcore_state = VCORE_INACTIVE;
3055 /* Unlock all except the primary vcore */
3056 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3057 pvc = core_info.vc[sub];
3058 /* Put back on to the preempted vcores list */
3059 kvmppc_vcore_preempt(pvc);
3060 spin_unlock(&pvc->lock);
3062 for (i = 0; i < controlled_threads; ++i)
3063 kvmppc_release_hwthread(pcpu + i);
3067 kvmppc_clear_host_core(pcpu);
3069 /* Decide on micro-threading (split-core) mode */
3070 subcore_size = threads_per_subcore;
3071 cmd_bit = stat_bit = 0;
3072 split = core_info.n_subcores;
3074 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3075 && !cpu_has_feature(CPU_FTR_ARCH_300);
3077 if (split > 1 || hpt_on_radix) {
3079 memset(&split_info, 0, sizeof(split_info));
3080 for (sub = 0; sub < core_info.n_subcores; ++sub)
3081 split_info.vc[sub] = core_info.vc[sub];
3084 if (split == 2 && (dynamic_mt_modes & 2)) {
3085 cmd_bit = HID0_POWER8_1TO2LPAR;
3086 stat_bit = HID0_POWER8_2LPARMODE;
3089 cmd_bit = HID0_POWER8_1TO4LPAR;
3090 stat_bit = HID0_POWER8_4LPARMODE;
3092 subcore_size = MAX_SMT_THREADS / split;
3093 split_info.rpr = mfspr(SPRN_RPR);
3094 split_info.pmmar = mfspr(SPRN_PMMAR);
3095 split_info.ldbar = mfspr(SPRN_LDBAR);
3096 split_info.subcore_size = subcore_size;
3098 split_info.subcore_size = 1;
3100 /* Use the split_info for LPCR/LPIDR changes */
3101 split_info.lpcr_req = vc->lpcr;
3102 split_info.lpidr_req = vc->kvm->arch.lpid;
3103 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3104 split_info.do_set = 1;
3108 /* order writes to split_info before kvm_split_mode pointer */
3112 for (thr = 0; thr < controlled_threads; ++thr) {
3113 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3115 paca->kvm_hstate.tid = thr;
3116 paca->kvm_hstate.napping = 0;
3117 paca->kvm_hstate.kvm_split_mode = sip;
3120 /* Initiate micro-threading (split-core) on POWER8 if required */
3122 unsigned long hid0 = mfspr(SPRN_HID0);
3124 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3126 mtspr(SPRN_HID0, hid0);
3129 hid0 = mfspr(SPRN_HID0);
3130 if (hid0 & stat_bit)
3137 * On POWER8, set RWMR register.
3138 * Since it only affects PURR and SPURR, it doesn't affect
3139 * the host, so we don't save/restore the host value.
3142 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3143 int n_online = atomic_read(&vc->online_count);
3146 * Use the 8-thread value if we're doing split-core
3147 * or if the vcore's online count looks bogus.
3149 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3150 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3151 rwmr_val = p8_rwmr_values[n_online];
3152 mtspr(SPRN_RWMR, rwmr_val);
3155 /* Start all the threads */
3157 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3158 thr = is_power8 ? subcore_thread_map[sub] : sub;
3161 pvc = core_info.vc[sub];
3162 pvc->pcpu = pcpu + thr;
3163 for_each_runnable_thread(i, vcpu, pvc) {
3164 kvmppc_start_thread(vcpu, pvc);
3165 kvmppc_create_dtl_entry(vcpu, pvc);
3166 trace_kvm_guest_enter(vcpu);
3167 if (!vcpu->arch.ptid)
3169 active |= 1 << (thr + vcpu->arch.ptid);
3172 * We need to start the first thread of each subcore
3173 * even if it doesn't have a vcpu.
3176 kvmppc_start_thread(NULL, pvc);
3180 * Ensure that split_info.do_nap is set after setting
3181 * the vcore pointer in the PACA of the secondaries.
3186 * When doing micro-threading, poke the inactive threads as well.
3187 * This gets them to the nap instruction after kvm_do_nap,
3188 * which reduces the time taken to unsplit later.
3189 * For POWER9 HPT guest on radix host, we need all the secondary
3190 * threads woken up so they can do the LPCR/LPIDR change.
3192 if (cmd_bit || hpt_on_radix) {
3193 split_info.do_nap = 1; /* ask secondaries to nap when done */
3194 for (thr = 1; thr < threads_per_subcore; ++thr)
3195 if (!(active & (1 << thr)))
3196 kvmppc_ipi_thread(pcpu + thr);
3199 vc->vcore_state = VCORE_RUNNING;
3202 trace_kvmppc_run_core(vc, 0);
3204 for (sub = 0; sub < core_info.n_subcores; ++sub)
3205 spin_unlock(&core_info.vc[sub]->lock);
3207 if (kvm_is_radix(vc->kvm)) {
3209 * Do we need to flush the process scoped TLB for the LPAR?
3211 * On POWER9, individual threads can come in here, but the
3212 * TLB is shared between the 4 threads in a core, hence
3213 * invalidating on one thread invalidates for all.
3214 * Thus we make all 4 threads use the same bit here.
3216 * Hash must be flushed in realmode in order to use tlbiel.
3218 kvmppc_radix_check_need_tlb_flush(vc->kvm, pcpu, NULL);
3222 * Interrupts will be enabled once we get into the guest,
3223 * so tell lockdep that we're about to enable interrupts.
3225 trace_hardirqs_on();
3227 guest_enter_irqoff();
3229 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3231 this_cpu_disable_ftrace();
3233 trap = __kvmppc_vcore_entry();
3235 this_cpu_enable_ftrace();
3237 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3239 trace_hardirqs_off();
3240 set_irq_happened(trap);
3242 spin_lock(&vc->lock);
3243 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3244 vc->vcore_state = VCORE_EXITING;
3246 /* wait for secondary threads to finish writing their state to memory */
3247 kvmppc_wait_for_nap(controlled_threads);
3249 /* Return to whole-core mode if we split the core earlier */
3251 unsigned long hid0 = mfspr(SPRN_HID0);
3252 unsigned long loops = 0;
3254 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3255 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3257 mtspr(SPRN_HID0, hid0);
3260 hid0 = mfspr(SPRN_HID0);
3261 if (!(hid0 & stat_bit))
3266 } else if (hpt_on_radix) {
3267 /* Wait for all threads to have seen final sync */
3268 for (thr = 1; thr < controlled_threads; ++thr) {
3269 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3271 while (paca->kvm_hstate.kvm_split_mode) {
3278 split_info.do_nap = 0;
3280 kvmppc_set_host_core(pcpu);
3285 /* Let secondaries go back to the offline loop */
3286 for (i = 0; i < controlled_threads; ++i) {
3287 kvmppc_release_hwthread(pcpu + i);
3288 if (sip && sip->napped[i])
3289 kvmppc_ipi_thread(pcpu + i);
3290 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3293 spin_unlock(&vc->lock);
3295 /* make sure updates to secondary vcpu structs are visible now */
3300 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3301 pvc = core_info.vc[sub];
3302 post_guest_process(pvc, pvc == vc);
3305 spin_lock(&vc->lock);
3308 vc->vcore_state = VCORE_INACTIVE;
3309 trace_kvmppc_run_core(vc, 1);
3313 * Load up hypervisor-mode registers on P9.
3315 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3318 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3320 u64 tb, purr, spurr;
3322 unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3323 unsigned long host_ciabr = mfspr(SPRN_CIABR);
3324 unsigned long host_dawr = mfspr(SPRN_DAWR);
3325 unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3326 unsigned long host_psscr = mfspr(SPRN_PSSCR);
3327 unsigned long host_pidr = mfspr(SPRN_PID);
3329 hdec = time_limit - mftb();
3331 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3332 mtspr(SPRN_HDEC, hdec);
3334 if (vc->tb_offset) {
3335 u64 new_tb = mftb() + vc->tb_offset;
3336 mtspr(SPRN_TBU40, new_tb);
3338 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3339 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3340 vc->tb_offset_applied = vc->tb_offset;
3344 mtspr(SPRN_PCR, vc->pcr);
3345 mtspr(SPRN_DPDES, vc->dpdes);
3346 mtspr(SPRN_VTB, vc->vtb);
3348 local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3349 local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3350 mtspr(SPRN_PURR, vcpu->arch.purr);
3351 mtspr(SPRN_SPURR, vcpu->arch.spurr);
3353 if (cpu_has_feature(CPU_FTR_DAWR)) {
3354 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3355 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3357 mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3358 mtspr(SPRN_IC, vcpu->arch.ic);
3359 mtspr(SPRN_PID, vcpu->arch.pid);
3361 mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3362 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3364 mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3366 mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3367 mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3368 mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3369 mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3371 mtspr(SPRN_AMOR, ~0UL);
3373 mtspr(SPRN_LPCR, lpcr);
3376 kvmppc_xive_push_vcpu(vcpu);
3378 mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3379 mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3381 trap = __kvmhv_vcpu_entry_p9(vcpu);
3383 /* Advance host PURR/SPURR by the amount used by guest */
3384 purr = mfspr(SPRN_PURR);
3385 spurr = mfspr(SPRN_SPURR);
3386 mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3387 purr - vcpu->arch.purr);
3388 mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3389 spurr - vcpu->arch.spurr);
3390 vcpu->arch.purr = purr;
3391 vcpu->arch.spurr = spurr;
3393 vcpu->arch.ic = mfspr(SPRN_IC);
3394 vcpu->arch.pid = mfspr(SPRN_PID);
3395 vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3397 vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3398 vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3399 vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3400 vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3402 mtspr(SPRN_PSSCR, host_psscr);
3403 mtspr(SPRN_HFSCR, host_hfscr);
3404 mtspr(SPRN_CIABR, host_ciabr);
3405 mtspr(SPRN_DAWR, host_dawr);
3406 mtspr(SPRN_DAWRX, host_dawrx);
3407 mtspr(SPRN_PID, host_pidr);
3410 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3411 * case we interrupted the guest between a tlbie and a ptesync.
3413 asm volatile("eieio; tlbsync; ptesync");
3415 mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid); /* restore host LPID */
3418 vc->dpdes = mfspr(SPRN_DPDES);
3419 vc->vtb = mfspr(SPRN_VTB);
3420 mtspr(SPRN_DPDES, 0);
3424 if (vc->tb_offset_applied) {
3425 u64 new_tb = mftb() - vc->tb_offset_applied;
3426 mtspr(SPRN_TBU40, new_tb);
3428 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3429 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3430 vc->tb_offset_applied = 0;
3433 mtspr(SPRN_HDEC, 0x7fffffff);
3434 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3440 * Virtual-mode guest entry for POWER9 and later when the host and
3441 * guest are both using the radix MMU. The LPIDR has already been set.
3443 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3446 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3447 unsigned long host_dscr = mfspr(SPRN_DSCR);
3448 unsigned long host_tidr = mfspr(SPRN_TIDR);
3449 unsigned long host_iamr = mfspr(SPRN_IAMR);
3454 dec = mfspr(SPRN_DEC);
3457 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3458 local_paca->kvm_hstate.dec_expires = dec + tb;
3459 if (local_paca->kvm_hstate.dec_expires < time_limit)
3460 time_limit = local_paca->kvm_hstate.dec_expires;
3462 vcpu->arch.ceded = 0;
3464 kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */
3466 kvmppc_subcore_enter_guest();
3468 vc->entry_exit_map = 1;
3471 if (vcpu->arch.vpa.pinned_addr) {
3472 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3473 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3474 lp->yield_count = cpu_to_be32(yield_count);
3475 vcpu->arch.vpa.dirty = 1;
3478 if (cpu_has_feature(CPU_FTR_TM) ||
3479 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3480 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3482 kvmhv_load_guest_pmu(vcpu);
3484 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3485 load_fp_state(&vcpu->arch.fp);
3486 #ifdef CONFIG_ALTIVEC
3487 load_vr_state(&vcpu->arch.vr);
3490 mtspr(SPRN_DSCR, vcpu->arch.dscr);
3491 mtspr(SPRN_IAMR, vcpu->arch.iamr);
3492 mtspr(SPRN_PSPB, vcpu->arch.pspb);
3493 mtspr(SPRN_FSCR, vcpu->arch.fscr);
3494 mtspr(SPRN_TAR, vcpu->arch.tar);
3495 mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3496 mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3497 mtspr(SPRN_BESCR, vcpu->arch.bescr);
3498 mtspr(SPRN_WORT, vcpu->arch.wort);
3499 mtspr(SPRN_TIDR, vcpu->arch.tid);
3500 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3501 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3502 mtspr(SPRN_AMR, vcpu->arch.amr);
3503 mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3505 if (!(vcpu->arch.ctrl & 1))
3506 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3508 mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3510 if (kvmhv_on_pseries()) {
3511 /* call our hypervisor to load up HV regs and go */
3512 struct hv_guest_state hvregs;
3514 kvmhv_save_hv_regs(vcpu, &hvregs);
3516 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3517 hvregs.version = HV_GUEST_STATE_VERSION;
3518 if (vcpu->arch.nested) {
3519 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3520 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3522 hvregs.lpid = vcpu->kvm->arch.lpid;
3523 hvregs.vcpu_token = vcpu->vcpu_id;
3525 hvregs.hdec_expiry = time_limit;
3526 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3527 __pa(&vcpu->arch.regs));
3528 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3529 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3530 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3531 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3533 /* H_CEDE has to be handled now, not later */
3534 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3535 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3536 kvmppc_nested_cede(vcpu);
3540 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3543 vcpu->arch.slb_max = 0;
3544 dec = mfspr(SPRN_DEC);
3546 vcpu->arch.dec_expires = dec + tb;
3548 vcpu->arch.thread_cpu = -1;
3549 vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3551 vcpu->arch.iamr = mfspr(SPRN_IAMR);
3552 vcpu->arch.pspb = mfspr(SPRN_PSPB);
3553 vcpu->arch.fscr = mfspr(SPRN_FSCR);
3554 vcpu->arch.tar = mfspr(SPRN_TAR);
3555 vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3556 vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3557 vcpu->arch.bescr = mfspr(SPRN_BESCR);
3558 vcpu->arch.wort = mfspr(SPRN_WORT);
3559 vcpu->arch.tid = mfspr(SPRN_TIDR);
3560 vcpu->arch.amr = mfspr(SPRN_AMR);
3561 vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3562 vcpu->arch.dscr = mfspr(SPRN_DSCR);
3564 mtspr(SPRN_PSPB, 0);
3565 mtspr(SPRN_WORT, 0);
3567 mtspr(SPRN_UAMOR, 0);
3568 mtspr(SPRN_DSCR, host_dscr);
3569 mtspr(SPRN_TIDR, host_tidr);
3570 mtspr(SPRN_IAMR, host_iamr);
3571 mtspr(SPRN_PSPB, 0);
3573 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3574 store_fp_state(&vcpu->arch.fp);
3575 #ifdef CONFIG_ALTIVEC
3576 store_vr_state(&vcpu->arch.vr);
3579 if (cpu_has_feature(CPU_FTR_TM) ||
3580 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3581 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3584 if (vcpu->arch.vpa.pinned_addr) {
3585 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3586 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3587 lp->yield_count = cpu_to_be32(yield_count);
3588 vcpu->arch.vpa.dirty = 1;
3589 save_pmu = lp->pmcregs_in_use;
3592 kvmhv_save_guest_pmu(vcpu, save_pmu);
3594 vc->entry_exit_map = 0x101;
3597 mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3599 kvmhv_load_host_pmu();
3601 kvmppc_subcore_exit_guest();
3607 * Wait for some other vcpu thread to execute us, and
3608 * wake us up when we need to handle something in the host.
3610 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3611 struct kvm_vcpu *vcpu, int wait_state)
3615 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3616 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3617 spin_unlock(&vc->lock);
3619 spin_lock(&vc->lock);
3621 finish_wait(&vcpu->arch.cpu_run, &wait);
3624 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3627 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3628 vc->halt_poll_ns = 10000;
3630 vc->halt_poll_ns *= halt_poll_ns_grow;
3633 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3635 if (halt_poll_ns_shrink == 0)
3636 vc->halt_poll_ns = 0;
3638 vc->halt_poll_ns /= halt_poll_ns_shrink;
3641 #ifdef CONFIG_KVM_XICS
3642 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3644 if (!xive_enabled())
3646 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3647 vcpu->arch.xive_saved_state.cppr;
3650 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3654 #endif /* CONFIG_KVM_XICS */
3656 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3658 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3659 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3666 * Check to see if any of the runnable vcpus on the vcore have pending
3667 * exceptions or are no longer ceded
3669 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3671 struct kvm_vcpu *vcpu;
3674 for_each_runnable_thread(i, vcpu, vc) {
3675 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3683 * All the vcpus in this vcore are idle, so wait for a decrementer
3684 * or external interrupt to one of the vcpus. vc->lock is held.
3686 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3688 ktime_t cur, start_poll, start_wait;
3691 DECLARE_SWAITQUEUE(wait);
3693 /* Poll for pending exceptions and ceded state */
3694 cur = start_poll = ktime_get();
3695 if (vc->halt_poll_ns) {
3696 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3697 ++vc->runner->stat.halt_attempted_poll;
3699 vc->vcore_state = VCORE_POLLING;
3700 spin_unlock(&vc->lock);
3703 if (kvmppc_vcore_check_block(vc)) {
3708 } while (single_task_running() && ktime_before(cur, stop));
3710 spin_lock(&vc->lock);
3711 vc->vcore_state = VCORE_INACTIVE;
3714 ++vc->runner->stat.halt_successful_poll;
3719 prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3721 if (kvmppc_vcore_check_block(vc)) {
3722 finish_swait(&vc->wq, &wait);
3724 /* If we polled, count this as a successful poll */
3725 if (vc->halt_poll_ns)
3726 ++vc->runner->stat.halt_successful_poll;
3730 start_wait = ktime_get();
3732 vc->vcore_state = VCORE_SLEEPING;
3733 trace_kvmppc_vcore_blocked(vc, 0);
3734 spin_unlock(&vc->lock);
3736 finish_swait(&vc->wq, &wait);
3737 spin_lock(&vc->lock);
3738 vc->vcore_state = VCORE_INACTIVE;
3739 trace_kvmppc_vcore_blocked(vc, 1);
3740 ++vc->runner->stat.halt_successful_wait;
3745 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3747 /* Attribute wait time */
3749 vc->runner->stat.halt_wait_ns +=
3750 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3751 /* Attribute failed poll time */
3752 if (vc->halt_poll_ns)
3753 vc->runner->stat.halt_poll_fail_ns +=
3754 ktime_to_ns(start_wait) -
3755 ktime_to_ns(start_poll);
3757 /* Attribute successful poll time */
3758 if (vc->halt_poll_ns)
3759 vc->runner->stat.halt_poll_success_ns +=
3761 ktime_to_ns(start_poll);
3764 /* Adjust poll time */
3766 if (block_ns <= vc->halt_poll_ns)
3768 /* We slept and blocked for longer than the max halt time */
3769 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3770 shrink_halt_poll_ns(vc);
3771 /* We slept and our poll time is too small */
3772 else if (vc->halt_poll_ns < halt_poll_ns &&
3773 block_ns < halt_poll_ns)
3774 grow_halt_poll_ns(vc);
3775 if (vc->halt_poll_ns > halt_poll_ns)
3776 vc->halt_poll_ns = halt_poll_ns;
3778 vc->halt_poll_ns = 0;
3780 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3784 * This never fails for a radix guest, as none of the operations it does
3785 * for a radix guest can fail or have a way to report failure.
3786 * kvmhv_run_single_vcpu() relies on this fact.
3788 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3791 struct kvm *kvm = vcpu->kvm;
3793 mutex_lock(&kvm->lock);
3794 if (!kvm->arch.mmu_ready) {
3795 if (!kvm_is_radix(kvm))
3796 r = kvmppc_hv_setup_htab_rma(vcpu);
3798 if (cpu_has_feature(CPU_FTR_ARCH_300))
3799 kvmppc_setup_partition_table(kvm);
3800 kvm->arch.mmu_ready = 1;
3803 mutex_unlock(&kvm->lock);
3807 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3810 struct kvmppc_vcore *vc;
3813 trace_kvmppc_run_vcpu_enter(vcpu);
3815 kvm_run->exit_reason = 0;
3816 vcpu->arch.ret = RESUME_GUEST;
3817 vcpu->arch.trap = 0;
3818 kvmppc_update_vpas(vcpu);
3821 * Synchronize with other threads in this virtual core
3823 vc = vcpu->arch.vcore;
3824 spin_lock(&vc->lock);
3825 vcpu->arch.ceded = 0;
3826 vcpu->arch.run_task = current;
3827 vcpu->arch.kvm_run = kvm_run;
3828 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3829 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3830 vcpu->arch.busy_preempt = TB_NIL;
3831 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3835 * This happens the first time this is called for a vcpu.
3836 * If the vcore is already running, we may be able to start
3837 * this thread straight away and have it join in.
3839 if (!signal_pending(current)) {
3840 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3841 vc->vcore_state == VCORE_RUNNING) &&
3842 !VCORE_IS_EXITING(vc)) {
3843 kvmppc_create_dtl_entry(vcpu, vc);
3844 kvmppc_start_thread(vcpu, vc);
3845 trace_kvm_guest_enter(vcpu);
3846 } else if (vc->vcore_state == VCORE_SLEEPING) {
3847 swake_up_one(&vc->wq);
3852 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3853 !signal_pending(current)) {
3854 /* See if the MMU is ready to go */
3855 if (!vcpu->kvm->arch.mmu_ready) {
3856 spin_unlock(&vc->lock);
3857 r = kvmhv_setup_mmu(vcpu);
3858 spin_lock(&vc->lock);
3860 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3861 kvm_run->fail_entry.
3862 hardware_entry_failure_reason = 0;
3868 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3869 kvmppc_vcore_end_preempt(vc);
3871 if (vc->vcore_state != VCORE_INACTIVE) {
3872 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3875 for_each_runnable_thread(i, v, vc) {
3876 kvmppc_core_prepare_to_enter(v);
3877 if (signal_pending(v->arch.run_task)) {
3878 kvmppc_remove_runnable(vc, v);
3879 v->stat.signal_exits++;
3880 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3881 v->arch.ret = -EINTR;
3882 wake_up(&v->arch.cpu_run);
3885 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3888 for_each_runnable_thread(i, v, vc) {
3889 if (!kvmppc_vcpu_woken(v))
3890 n_ceded += v->arch.ceded;
3895 if (n_ceded == vc->n_runnable) {
3896 kvmppc_vcore_blocked(vc);
3897 } else if (need_resched()) {
3898 kvmppc_vcore_preempt(vc);
3899 /* Let something else run */
3900 cond_resched_lock(&vc->lock);
3901 if (vc->vcore_state == VCORE_PREEMPT)
3902 kvmppc_vcore_end_preempt(vc);
3904 kvmppc_run_core(vc);
3909 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3910 (vc->vcore_state == VCORE_RUNNING ||
3911 vc->vcore_state == VCORE_EXITING ||
3912 vc->vcore_state == VCORE_PIGGYBACK))
3913 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3915 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3916 kvmppc_vcore_end_preempt(vc);
3918 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3919 kvmppc_remove_runnable(vc, vcpu);
3920 vcpu->stat.signal_exits++;
3921 kvm_run->exit_reason = KVM_EXIT_INTR;
3922 vcpu->arch.ret = -EINTR;
3925 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3926 /* Wake up some vcpu to run the core */
3928 v = next_runnable_thread(vc, &i);
3929 wake_up(&v->arch.cpu_run);
3932 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3933 spin_unlock(&vc->lock);
3934 return vcpu->arch.ret;
3937 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
3938 struct kvm_vcpu *vcpu, u64 time_limit,
3943 struct kvmppc_vcore *vc;
3944 struct kvm *kvm = vcpu->kvm;
3945 struct kvm_nested_guest *nested = vcpu->arch.nested;
3947 trace_kvmppc_run_vcpu_enter(vcpu);
3949 kvm_run->exit_reason = 0;
3950 vcpu->arch.ret = RESUME_GUEST;
3951 vcpu->arch.trap = 0;
3953 vc = vcpu->arch.vcore;
3954 vcpu->arch.ceded = 0;
3955 vcpu->arch.run_task = current;
3956 vcpu->arch.kvm_run = kvm_run;
3957 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3958 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3959 vcpu->arch.busy_preempt = TB_NIL;
3960 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
3961 vc->runnable_threads[0] = vcpu;
3965 /* See if the MMU is ready to go */
3966 if (!kvm->arch.mmu_ready)
3967 kvmhv_setup_mmu(vcpu);
3972 kvmppc_update_vpas(vcpu);
3974 init_vcore_to_run(vc);
3975 vc->preempt_tb = TB_NIL;
3978 pcpu = smp_processor_id();
3980 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3982 local_irq_disable();
3984 if (signal_pending(current))
3986 if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
3990 kvmppc_core_prepare_to_enter(vcpu);
3991 if (vcpu->arch.doorbell_request) {
3994 vcpu->arch.doorbell_request = 0;
3996 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
3997 &vcpu->arch.pending_exceptions))
3999 } else if (vcpu->arch.pending_exceptions ||
4000 vcpu->arch.doorbell_request ||
4001 xive_interrupt_pending(vcpu)) {
4002 vcpu->arch.ret = RESUME_HOST;
4006 kvmppc_clear_host_core(pcpu);
4008 local_paca->kvm_hstate.tid = 0;
4009 local_paca->kvm_hstate.napping = 0;
4010 local_paca->kvm_hstate.kvm_split_mode = NULL;
4011 kvmppc_start_thread(vcpu, vc);
4012 kvmppc_create_dtl_entry(vcpu, vc);
4013 trace_kvm_guest_enter(vcpu);
4015 vc->vcore_state = VCORE_RUNNING;
4016 trace_kvmppc_run_core(vc, 0);
4018 if (cpu_has_feature(CPU_FTR_HVMODE))
4019 kvmppc_radix_check_need_tlb_flush(kvm, pcpu, nested);
4021 trace_hardirqs_on();
4022 guest_enter_irqoff();
4024 srcu_idx = srcu_read_lock(&kvm->srcu);
4026 this_cpu_disable_ftrace();
4028 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4029 vcpu->arch.trap = trap;
4031 this_cpu_enable_ftrace();
4033 srcu_read_unlock(&kvm->srcu, srcu_idx);
4035 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4036 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4040 trace_hardirqs_off();
4041 set_irq_happened(trap);
4043 kvmppc_set_host_core(pcpu);
4048 cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4052 /* cancel pending decrementer exception if DEC is now positive */
4053 if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4054 kvmppc_core_dequeue_dec(vcpu);
4056 trace_kvm_guest_exit(vcpu);
4060 r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4062 r = kvmppc_handle_nested_exit(vcpu);
4066 if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4067 !kvmppc_vcpu_woken(vcpu)) {
4068 kvmppc_set_timer(vcpu);
4069 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4070 if (signal_pending(current)) {
4071 vcpu->stat.signal_exits++;
4072 kvm_run->exit_reason = KVM_EXIT_INTR;
4073 vcpu->arch.ret = -EINTR;
4076 spin_lock(&vc->lock);
4077 kvmppc_vcore_blocked(vc);
4078 spin_unlock(&vc->lock);
4081 vcpu->arch.ceded = 0;
4083 vc->vcore_state = VCORE_INACTIVE;
4084 trace_kvmppc_run_core(vc, 1);
4087 kvmppc_remove_runnable(vc, vcpu);
4088 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4090 return vcpu->arch.ret;
4093 vcpu->stat.signal_exits++;
4094 kvm_run->exit_reason = KVM_EXIT_INTR;
4095 vcpu->arch.ret = -EINTR;
4102 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4106 unsigned long ebb_regs[3] = {}; /* shut up GCC */
4107 unsigned long user_tar = 0;
4108 unsigned int user_vrsave;
4111 if (!vcpu->arch.sane) {
4112 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4117 * Don't allow entry with a suspended transaction, because
4118 * the guest entry/exit code will lose it.
4119 * If the guest has TM enabled, save away their TM-related SPRs
4120 * (they will get restored by the TM unavailable interrupt).
4122 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4123 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4124 (current->thread.regs->msr & MSR_TM)) {
4125 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4126 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4127 run->fail_entry.hardware_entry_failure_reason = 0;
4130 /* Enable TM so we can read the TM SPRs */
4131 mtmsr(mfmsr() | MSR_TM);
4132 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4133 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4134 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4135 current->thread.regs->msr &= ~MSR_TM;
4140 * Force online to 1 for the sake of old userspace which doesn't
4143 if (!vcpu->arch.online) {
4144 atomic_inc(&vcpu->arch.vcore->online_count);
4145 vcpu->arch.online = 1;
4148 kvmppc_core_prepare_to_enter(vcpu);
4150 /* No need to go into the guest when all we'll do is come back out */
4151 if (signal_pending(current)) {
4152 run->exit_reason = KVM_EXIT_INTR;
4157 atomic_inc(&kvm->arch.vcpus_running);
4158 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4161 flush_all_to_thread(current);
4163 /* Save userspace EBB and other register values */
4164 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4165 ebb_regs[0] = mfspr(SPRN_EBBHR);
4166 ebb_regs[1] = mfspr(SPRN_EBBRR);
4167 ebb_regs[2] = mfspr(SPRN_BESCR);
4168 user_tar = mfspr(SPRN_TAR);
4170 user_vrsave = mfspr(SPRN_VRSAVE);
4172 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4173 vcpu->arch.pgdir = current->mm->pgd;
4174 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4178 * The early POWER9 chips that can't mix radix and HPT threads
4179 * on the same core also need the workaround for the problem
4180 * where the TLB would prefetch entries in the guest exit path
4181 * for radix guests using the guest PIDR value and LPID 0.
4182 * The workaround is in the old path (kvmppc_run_vcpu())
4183 * but not the new path (kvmhv_run_single_vcpu()).
4185 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4186 !no_mixing_hpt_and_radix)
4187 r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4188 vcpu->arch.vcore->lpcr);
4190 r = kvmppc_run_vcpu(run, vcpu);
4192 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4193 !(vcpu->arch.shregs.msr & MSR_PR)) {
4194 trace_kvm_hcall_enter(vcpu);
4195 r = kvmppc_pseries_do_hcall(vcpu);
4196 trace_kvm_hcall_exit(vcpu, r);
4197 kvmppc_core_prepare_to_enter(vcpu);
4198 } else if (r == RESUME_PAGE_FAULT) {
4199 srcu_idx = srcu_read_lock(&kvm->srcu);
4200 r = kvmppc_book3s_hv_page_fault(run, vcpu,
4201 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4202 srcu_read_unlock(&kvm->srcu, srcu_idx);
4203 } else if (r == RESUME_PASSTHROUGH) {
4204 if (WARN_ON(xive_enabled()))
4207 r = kvmppc_xics_rm_complete(vcpu, 0);
4209 } while (is_kvmppc_resume_guest(r));
4211 /* Restore userspace EBB and other register values */
4212 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4213 mtspr(SPRN_EBBHR, ebb_regs[0]);
4214 mtspr(SPRN_EBBRR, ebb_regs[1]);
4215 mtspr(SPRN_BESCR, ebb_regs[2]);
4216 mtspr(SPRN_TAR, user_tar);
4217 mtspr(SPRN_FSCR, current->thread.fscr);
4219 mtspr(SPRN_VRSAVE, user_vrsave);
4221 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4222 atomic_dec(&kvm->arch.vcpus_running);
4226 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4227 int shift, int sllp)
4229 (*sps)->page_shift = shift;
4230 (*sps)->slb_enc = sllp;
4231 (*sps)->enc[0].page_shift = shift;
4232 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4234 * Add 16MB MPSS support (may get filtered out by userspace)
4237 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4239 (*sps)->enc[1].page_shift = 24;
4240 (*sps)->enc[1].pte_enc = penc;
4246 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4247 struct kvm_ppc_smmu_info *info)
4249 struct kvm_ppc_one_seg_page_size *sps;
4252 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4253 * POWER7 doesn't support keys for instruction accesses,
4254 * POWER8 and POWER9 do.
4256 info->data_keys = 32;
4257 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4259 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4260 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4261 info->slb_size = 32;
4263 /* We only support these sizes for now, and no muti-size segments */
4264 sps = &info->sps[0];
4265 kvmppc_add_seg_page_size(&sps, 12, 0);
4266 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4267 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4269 /* If running as a nested hypervisor, we don't support HPT guests */
4270 if (kvmhv_on_pseries())
4271 info->flags |= KVM_PPC_NO_HASH;
4277 * Get (and clear) the dirty memory log for a memory slot.
4279 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4280 struct kvm_dirty_log *log)
4282 struct kvm_memslots *slots;
4283 struct kvm_memory_slot *memslot;
4286 unsigned long *buf, *p;
4287 struct kvm_vcpu *vcpu;
4289 mutex_lock(&kvm->slots_lock);
4292 if (log->slot >= KVM_USER_MEM_SLOTS)
4295 slots = kvm_memslots(kvm);
4296 memslot = id_to_memslot(slots, log->slot);
4298 if (!memslot->dirty_bitmap)
4302 * Use second half of bitmap area because both HPT and radix
4303 * accumulate bits in the first half.
4305 n = kvm_dirty_bitmap_bytes(memslot);
4306 buf = memslot->dirty_bitmap + n / sizeof(long);
4309 if (kvm_is_radix(kvm))
4310 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4312 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4317 * We accumulate dirty bits in the first half of the
4318 * memslot's dirty_bitmap area, for when pages are paged
4319 * out or modified by the host directly. Pick up these
4320 * bits and add them to the map.
4322 p = memslot->dirty_bitmap;
4323 for (i = 0; i < n / sizeof(long); ++i)
4324 buf[i] |= xchg(&p[i], 0);
4326 /* Harvest dirty bits from VPA and DTL updates */
4327 /* Note: we never modify the SLB shadow buffer areas */
4328 kvm_for_each_vcpu(i, vcpu, kvm) {
4329 spin_lock(&vcpu->arch.vpa_update_lock);
4330 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4331 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4332 spin_unlock(&vcpu->arch.vpa_update_lock);
4336 if (copy_to_user(log->dirty_bitmap, buf, n))
4341 mutex_unlock(&kvm->slots_lock);
4345 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4346 struct kvm_memory_slot *dont)
4348 if (!dont || free->arch.rmap != dont->arch.rmap) {
4349 vfree(free->arch.rmap);
4350 free->arch.rmap = NULL;
4354 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4355 unsigned long npages)
4357 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4358 if (!slot->arch.rmap)
4364 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4365 struct kvm_memory_slot *memslot,
4366 const struct kvm_userspace_memory_region *mem)
4371 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4372 const struct kvm_userspace_memory_region *mem,
4373 const struct kvm_memory_slot *old,
4374 const struct kvm_memory_slot *new)
4376 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4379 * If we are making a new memslot, it might make
4380 * some address that was previously cached as emulated
4381 * MMIO be no longer emulated MMIO, so invalidate
4382 * all the caches of emulated MMIO translations.
4385 atomic64_inc(&kvm->arch.mmio_update);
4389 * Update LPCR values in kvm->arch and in vcores.
4390 * Caller must hold kvm->lock.
4392 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4397 if ((kvm->arch.lpcr & mask) == lpcr)
4400 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4402 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4403 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4406 spin_lock(&vc->lock);
4407 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4408 spin_unlock(&vc->lock);
4409 if (++cores_done >= kvm->arch.online_vcores)
4414 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4419 void kvmppc_setup_partition_table(struct kvm *kvm)
4421 unsigned long dw0, dw1;
4423 if (!kvm_is_radix(kvm)) {
4424 /* PS field - page size for VRMA */
4425 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4426 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4427 /* HTABSIZE and HTABORG fields */
4428 dw0 |= kvm->arch.sdr1;
4430 /* Second dword as set by userspace */
4431 dw1 = kvm->arch.process_table;
4433 dw0 = PATB_HR | radix__get_tree_size() |
4434 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4435 dw1 = PATB_GR | kvm->arch.process_table;
4437 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4441 * Set up HPT (hashed page table) and RMA (real-mode area).
4442 * Must be called with kvm->lock held.
4444 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4447 struct kvm *kvm = vcpu->kvm;
4449 struct kvm_memory_slot *memslot;
4450 struct vm_area_struct *vma;
4451 unsigned long lpcr = 0, senc;
4452 unsigned long psize, porder;
4455 /* Allocate hashed page table (if not done already) and reset it */
4456 if (!kvm->arch.hpt.virt) {
4457 int order = KVM_DEFAULT_HPT_ORDER;
4458 struct kvm_hpt_info info;
4460 err = kvmppc_allocate_hpt(&info, order);
4461 /* If we get here, it means userspace didn't specify a
4462 * size explicitly. So, try successively smaller
4463 * sizes if the default failed. */
4464 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4465 err = kvmppc_allocate_hpt(&info, order);
4468 pr_err("KVM: Couldn't alloc HPT\n");
4472 kvmppc_set_hpt(kvm, &info);
4475 /* Look up the memslot for guest physical address 0 */
4476 srcu_idx = srcu_read_lock(&kvm->srcu);
4477 memslot = gfn_to_memslot(kvm, 0);
4479 /* We must have some memory at 0 by now */
4481 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4484 /* Look up the VMA for the start of this memory slot */
4485 hva = memslot->userspace_addr;
4486 down_read(¤t->mm->mmap_sem);
4487 vma = find_vma(current->mm, hva);
4488 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4491 psize = vma_kernel_pagesize(vma);
4493 up_read(¤t->mm->mmap_sem);
4495 /* We can handle 4k, 64k or 16M pages in the VRMA */
4496 if (psize >= 0x1000000)
4498 else if (psize >= 0x10000)
4502 porder = __ilog2(psize);
4504 senc = slb_pgsize_encoding(psize);
4505 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4506 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4507 /* Create HPTEs in the hash page table for the VRMA */
4508 kvmppc_map_vrma(vcpu, memslot, porder);
4510 /* Update VRMASD field in the LPCR */
4511 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4512 /* the -4 is to account for senc values starting at 0x10 */
4513 lpcr = senc << (LPCR_VRMASD_SH - 4);
4514 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4517 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4521 srcu_read_unlock(&kvm->srcu, srcu_idx);
4526 up_read(¤t->mm->mmap_sem);
4530 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4531 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4533 if (nesting_enabled(kvm))
4534 kvmhv_release_all_nested(kvm);
4535 kvmppc_free_radix(kvm);
4536 kvmppc_update_lpcr(kvm, LPCR_VPM1,
4537 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4538 kvmppc_rmap_reset(kvm);
4539 kvm->arch.radix = 0;
4540 kvm->arch.process_table = 0;
4544 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4545 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4549 err = kvmppc_init_vm_radix(kvm);
4553 kvmppc_free_hpt(&kvm->arch.hpt);
4554 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4555 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4556 kvmppc_rmap_reset(kvm);
4557 kvm->arch.radix = 1;
4561 #ifdef CONFIG_KVM_XICS
4563 * Allocate a per-core structure for managing state about which cores are
4564 * running in the host versus the guest and for exchanging data between
4565 * real mode KVM and CPU running in the host.
4566 * This is only done for the first VM.
4567 * The allocated structure stays even if all VMs have stopped.
4568 * It is only freed when the kvm-hv module is unloaded.
4569 * It's OK for this routine to fail, we just don't support host
4570 * core operations like redirecting H_IPI wakeups.
4572 void kvmppc_alloc_host_rm_ops(void)
4574 struct kvmppc_host_rm_ops *ops;
4575 unsigned long l_ops;
4579 /* Not the first time here ? */
4580 if (kvmppc_host_rm_ops_hv != NULL)
4583 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4587 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4588 ops->rm_core = kzalloc(size, GFP_KERNEL);
4590 if (!ops->rm_core) {
4597 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4598 if (!cpu_online(cpu))
4601 core = cpu >> threads_shift;
4602 ops->rm_core[core].rm_state.in_host = 1;
4605 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4608 * Make the contents of the kvmppc_host_rm_ops structure visible
4609 * to other CPUs before we assign it to the global variable.
4610 * Do an atomic assignment (no locks used here), but if someone
4611 * beats us to it, just free our copy and return.
4614 l_ops = (unsigned long) ops;
4616 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4618 kfree(ops->rm_core);
4623 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4624 "ppc/kvm_book3s:prepare",
4625 kvmppc_set_host_core,
4626 kvmppc_clear_host_core);
4630 void kvmppc_free_host_rm_ops(void)
4632 if (kvmppc_host_rm_ops_hv) {
4633 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4634 kfree(kvmppc_host_rm_ops_hv->rm_core);
4635 kfree(kvmppc_host_rm_ops_hv);
4636 kvmppc_host_rm_ops_hv = NULL;
4641 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4643 unsigned long lpcr, lpid;
4647 /* Allocate the guest's logical partition ID */
4649 lpid = kvmppc_alloc_lpid();
4652 kvm->arch.lpid = lpid;
4654 kvmppc_alloc_host_rm_ops();
4656 kvmhv_vm_nested_init(kvm);
4659 * Since we don't flush the TLB when tearing down a VM,
4660 * and this lpid might have previously been used,
4661 * make sure we flush on each core before running the new VM.
4662 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4663 * does this flush for us.
4665 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4666 cpumask_setall(&kvm->arch.need_tlb_flush);
4668 /* Start out with the default set of hcalls enabled */
4669 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4670 sizeof(kvm->arch.enabled_hcalls));
4672 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4673 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4675 /* Init LPCR for virtual RMA mode */
4676 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4677 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4678 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4679 lpcr &= LPCR_PECE | LPCR_LPES;
4683 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4684 LPCR_VPM0 | LPCR_VPM1;
4685 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4686 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4687 /* On POWER8 turn on online bit to enable PURR/SPURR */
4688 if (cpu_has_feature(CPU_FTR_ARCH_207S))
4691 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4692 * Set HVICE bit to enable hypervisor virtualization interrupts.
4693 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4694 * be unnecessary but better safe than sorry in case we re-enable
4695 * EE in HV mode with this LPCR still set)
4697 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4699 lpcr |= LPCR_HVICE | LPCR_HEIC;
4702 * If xive is enabled, we route 0x500 interrupts directly
4710 * If the host uses radix, the guest starts out as radix.
4712 if (radix_enabled()) {
4713 kvm->arch.radix = 1;
4714 kvm->arch.mmu_ready = 1;
4716 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4717 ret = kvmppc_init_vm_radix(kvm);
4719 kvmppc_free_lpid(kvm->arch.lpid);
4722 kvmppc_setup_partition_table(kvm);
4725 kvm->arch.lpcr = lpcr;
4727 /* Initialization for future HPT resizes */
4728 kvm->arch.resize_hpt = NULL;
4731 * Work out how many sets the TLB has, for the use of
4732 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4734 if (radix_enabled())
4735 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
4736 else if (cpu_has_feature(CPU_FTR_ARCH_300))
4737 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
4738 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4739 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
4741 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
4744 * Track that we now have a HV mode VM active. This blocks secondary
4745 * CPU threads from coming online.
4746 * On POWER9, we only need to do this if the "indep_threads_mode"
4747 * module parameter has been set to N.
4749 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4750 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4751 pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4752 kvm->arch.threads_indep = true;
4754 kvm->arch.threads_indep = indep_threads_mode;
4757 if (!kvm->arch.threads_indep)
4758 kvm_hv_vm_activated();
4761 * Initialize smt_mode depending on processor.
4762 * POWER8 and earlier have to use "strict" threading, where
4763 * all vCPUs in a vcore have to run on the same (sub)core,
4764 * whereas on POWER9 the threads can each run a different
4767 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4768 kvm->arch.smt_mode = threads_per_subcore;
4770 kvm->arch.smt_mode = 1;
4771 kvm->arch.emul_smt_mode = 1;
4774 * Create a debugfs directory for the VM
4776 snprintf(buf, sizeof(buf), "vm%d", current->pid);
4777 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4778 kvmppc_mmu_debugfs_init(kvm);
4779 if (radix_enabled())
4780 kvmhv_radix_debugfs_init(kvm);
4785 static void kvmppc_free_vcores(struct kvm *kvm)
4789 for (i = 0; i < KVM_MAX_VCORES; ++i)
4790 kfree(kvm->arch.vcores[i]);
4791 kvm->arch.online_vcores = 0;
4794 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4796 debugfs_remove_recursive(kvm->arch.debugfs_dir);
4798 if (!kvm->arch.threads_indep)
4799 kvm_hv_vm_deactivated();
4801 kvmppc_free_vcores(kvm);
4804 if (kvm_is_radix(kvm))
4805 kvmppc_free_radix(kvm);
4807 kvmppc_free_hpt(&kvm->arch.hpt);
4809 /* Perform global invalidation and return lpid to the pool */
4810 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4811 if (nesting_enabled(kvm))
4812 kvmhv_release_all_nested(kvm);
4813 kvm->arch.process_table = 0;
4814 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4816 kvmppc_free_lpid(kvm->arch.lpid);
4818 kvmppc_free_pimap(kvm);
4821 /* We don't need to emulate any privileged instructions or dcbz */
4822 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4823 unsigned int inst, int *advance)
4825 return EMULATE_FAIL;
4828 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4831 return EMULATE_FAIL;
4834 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4837 return EMULATE_FAIL;
4840 static int kvmppc_core_check_processor_compat_hv(void)
4842 if (cpu_has_feature(CPU_FTR_HVMODE) &&
4843 cpu_has_feature(CPU_FTR_ARCH_206))
4846 /* POWER9 in radix mode is capable of being a nested hypervisor. */
4847 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4853 #ifdef CONFIG_KVM_XICS
4855 void kvmppc_free_pimap(struct kvm *kvm)
4857 kfree(kvm->arch.pimap);
4860 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4862 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4865 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4867 struct irq_desc *desc;
4868 struct kvmppc_irq_map *irq_map;
4869 struct kvmppc_passthru_irqmap *pimap;
4870 struct irq_chip *chip;
4873 if (!kvm_irq_bypass)
4876 desc = irq_to_desc(host_irq);
4880 mutex_lock(&kvm->lock);
4882 pimap = kvm->arch.pimap;
4883 if (pimap == NULL) {
4884 /* First call, allocate structure to hold IRQ map */
4885 pimap = kvmppc_alloc_pimap();
4886 if (pimap == NULL) {
4887 mutex_unlock(&kvm->lock);
4890 kvm->arch.pimap = pimap;
4894 * For now, we only support interrupts for which the EOI operation
4895 * is an OPAL call followed by a write to XIRR, since that's
4896 * what our real-mode EOI code does, or a XIVE interrupt
4898 chip = irq_data_get_irq_chip(&desc->irq_data);
4899 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4900 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4901 host_irq, guest_gsi);
4902 mutex_unlock(&kvm->lock);
4907 * See if we already have an entry for this guest IRQ number.
4908 * If it's mapped to a hardware IRQ number, that's an error,
4909 * otherwise re-use this entry.
4911 for (i = 0; i < pimap->n_mapped; i++) {
4912 if (guest_gsi == pimap->mapped[i].v_hwirq) {
4913 if (pimap->mapped[i].r_hwirq) {
4914 mutex_unlock(&kvm->lock);
4921 if (i == KVMPPC_PIRQ_MAPPED) {
4922 mutex_unlock(&kvm->lock);
4923 return -EAGAIN; /* table is full */
4926 irq_map = &pimap->mapped[i];
4928 irq_map->v_hwirq = guest_gsi;
4929 irq_map->desc = desc;
4932 * Order the above two stores before the next to serialize with
4933 * the KVM real mode handler.
4936 irq_map->r_hwirq = desc->irq_data.hwirq;
4938 if (i == pimap->n_mapped)
4942 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4944 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4946 irq_map->r_hwirq = 0;
4948 mutex_unlock(&kvm->lock);
4953 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4955 struct irq_desc *desc;
4956 struct kvmppc_passthru_irqmap *pimap;
4959 if (!kvm_irq_bypass)
4962 desc = irq_to_desc(host_irq);
4966 mutex_lock(&kvm->lock);
4967 if (!kvm->arch.pimap)
4970 pimap = kvm->arch.pimap;
4972 for (i = 0; i < pimap->n_mapped; i++) {
4973 if (guest_gsi == pimap->mapped[i].v_hwirq)
4977 if (i == pimap->n_mapped) {
4978 mutex_unlock(&kvm->lock);
4983 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4985 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4987 /* invalidate the entry (what do do on error from the above ?) */
4988 pimap->mapped[i].r_hwirq = 0;
4991 * We don't free this structure even when the count goes to
4992 * zero. The structure is freed when we destroy the VM.
4995 mutex_unlock(&kvm->lock);
4999 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5000 struct irq_bypass_producer *prod)
5003 struct kvm_kernel_irqfd *irqfd =
5004 container_of(cons, struct kvm_kernel_irqfd, consumer);
5006 irqfd->producer = prod;
5008 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5010 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5011 prod->irq, irqfd->gsi, ret);
5016 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5017 struct irq_bypass_producer *prod)
5020 struct kvm_kernel_irqfd *irqfd =
5021 container_of(cons, struct kvm_kernel_irqfd, consumer);
5023 irqfd->producer = NULL;
5026 * When producer of consumer is unregistered, we change back to
5027 * default external interrupt handling mode - KVM real mode
5028 * will switch back to host.
5030 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5032 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5033 prod->irq, irqfd->gsi, ret);
5037 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5038 unsigned int ioctl, unsigned long arg)
5040 struct kvm *kvm __maybe_unused = filp->private_data;
5041 void __user *argp = (void __user *)arg;
5046 case KVM_PPC_ALLOCATE_HTAB: {
5050 if (get_user(htab_order, (u32 __user *)argp))
5052 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5059 case KVM_PPC_GET_HTAB_FD: {
5060 struct kvm_get_htab_fd ghf;
5063 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5065 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5069 case KVM_PPC_RESIZE_HPT_PREPARE: {
5070 struct kvm_ppc_resize_hpt rhpt;
5073 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5076 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5080 case KVM_PPC_RESIZE_HPT_COMMIT: {
5081 struct kvm_ppc_resize_hpt rhpt;
5084 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5087 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5099 * List of hcall numbers to enable by default.
5100 * For compatibility with old userspace, we enable by default
5101 * all hcalls that were implemented before the hcall-enabling
5102 * facility was added. Note this list should not include H_RTAS.
5104 static unsigned int default_hcall_list[] = {
5118 #ifdef CONFIG_KVM_XICS
5129 static void init_default_hcalls(void)
5134 for (i = 0; default_hcall_list[i]; ++i) {
5135 hcall = default_hcall_list[i];
5136 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5137 __set_bit(hcall / 4, default_enabled_hcalls);
5141 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5147 /* If not on a POWER9, reject it */
5148 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5151 /* If any unknown flags set, reject it */
5152 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5155 /* GR (guest radix) bit in process_table field must match */
5156 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5157 if (!!(cfg->process_table & PATB_GR) != radix)
5160 /* Process table size field must be reasonable, i.e. <= 24 */
5161 if ((cfg->process_table & PRTS_MASK) > 24)
5164 /* We can change a guest to/from radix now, if the host is radix */
5165 if (radix && !radix_enabled())
5168 /* If we're a nested hypervisor, we currently only support radix */
5169 if (kvmhv_on_pseries() && !radix)
5172 mutex_lock(&kvm->lock);
5173 if (radix != kvm_is_radix(kvm)) {
5174 if (kvm->arch.mmu_ready) {
5175 kvm->arch.mmu_ready = 0;
5176 /* order mmu_ready vs. vcpus_running */
5178 if (atomic_read(&kvm->arch.vcpus_running)) {
5179 kvm->arch.mmu_ready = 1;
5185 err = kvmppc_switch_mmu_to_radix(kvm);
5187 err = kvmppc_switch_mmu_to_hpt(kvm);
5192 kvm->arch.process_table = cfg->process_table;
5193 kvmppc_setup_partition_table(kvm);
5195 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5196 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5200 mutex_unlock(&kvm->lock);
5204 static int kvmhv_enable_nested(struct kvm *kvm)
5208 if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5211 /* kvm == NULL means the caller is testing if the capability exists */
5213 kvm->arch.nested_enable = true;
5217 static struct kvmppc_ops kvm_ops_hv = {
5218 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5219 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5220 .get_one_reg = kvmppc_get_one_reg_hv,
5221 .set_one_reg = kvmppc_set_one_reg_hv,
5222 .vcpu_load = kvmppc_core_vcpu_load_hv,
5223 .vcpu_put = kvmppc_core_vcpu_put_hv,
5224 .set_msr = kvmppc_set_msr_hv,
5225 .vcpu_run = kvmppc_vcpu_run_hv,
5226 .vcpu_create = kvmppc_core_vcpu_create_hv,
5227 .vcpu_free = kvmppc_core_vcpu_free_hv,
5228 .check_requests = kvmppc_core_check_requests_hv,
5229 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
5230 .flush_memslot = kvmppc_core_flush_memslot_hv,
5231 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5232 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
5233 .unmap_hva_range = kvm_unmap_hva_range_hv,
5234 .age_hva = kvm_age_hva_hv,
5235 .test_age_hva = kvm_test_age_hva_hv,
5236 .set_spte_hva = kvm_set_spte_hva_hv,
5237 .mmu_destroy = kvmppc_mmu_destroy_hv,
5238 .free_memslot = kvmppc_core_free_memslot_hv,
5239 .create_memslot = kvmppc_core_create_memslot_hv,
5240 .init_vm = kvmppc_core_init_vm_hv,
5241 .destroy_vm = kvmppc_core_destroy_vm_hv,
5242 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5243 .emulate_op = kvmppc_core_emulate_op_hv,
5244 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5245 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5246 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5247 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
5248 .hcall_implemented = kvmppc_hcall_impl_hv,
5249 #ifdef CONFIG_KVM_XICS
5250 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5251 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5253 .configure_mmu = kvmhv_configure_mmu,
5254 .get_rmmu_info = kvmhv_get_rmmu_info,
5255 .set_smt_mode = kvmhv_set_smt_mode,
5256 .enable_nested = kvmhv_enable_nested,
5259 static int kvm_init_subcore_bitmap(void)
5262 int nr_cores = cpu_nr_cores();
5263 struct sibling_subcore_state *sibling_subcore_state;
5265 for (i = 0; i < nr_cores; i++) {
5266 int first_cpu = i * threads_per_core;
5267 int node = cpu_to_node(first_cpu);
5269 /* Ignore if it is already allocated. */
5270 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5273 sibling_subcore_state =
5274 kmalloc_node(sizeof(struct sibling_subcore_state),
5276 if (!sibling_subcore_state)
5279 memset(sibling_subcore_state, 0,
5280 sizeof(struct sibling_subcore_state));
5282 for (j = 0; j < threads_per_core; j++) {
5283 int cpu = first_cpu + j;
5285 paca_ptrs[cpu]->sibling_subcore_state =
5286 sibling_subcore_state;
5292 static int kvmppc_radix_possible(void)
5294 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5297 static int kvmppc_book3s_init_hv(void)
5301 * FIXME!! Do we need to check on all cpus ?
5303 r = kvmppc_core_check_processor_compat_hv();
5307 r = kvmhv_nested_init();
5311 r = kvm_init_subcore_bitmap();
5316 * We need a way of accessing the XICS interrupt controller,
5317 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5318 * indirectly, via OPAL.
5321 if (!xive_enabled() && !kvmhv_on_pseries() &&
5322 !local_paca->kvm_hstate.xics_phys) {
5323 struct device_node *np;
5325 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5327 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5330 /* presence of intc confirmed - node can be dropped again */
5335 kvm_ops_hv.owner = THIS_MODULE;
5336 kvmppc_hv_ops = &kvm_ops_hv;
5338 init_default_hcalls();
5342 r = kvmppc_mmu_hv_init();
5346 if (kvmppc_radix_possible())
5347 r = kvmppc_radix_init();
5350 * POWER9 chips before version 2.02 can't have some threads in
5351 * HPT mode and some in radix mode on the same core.
5353 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5354 unsigned int pvr = mfspr(SPRN_PVR);
5355 if ((pvr >> 16) == PVR_POWER9 &&
5356 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5357 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5358 no_mixing_hpt_and_radix = true;
5364 static void kvmppc_book3s_exit_hv(void)
5366 kvmppc_free_host_rm_ops();
5367 if (kvmppc_radix_possible())
5368 kvmppc_radix_exit();
5369 kvmppc_hv_ops = NULL;
5370 kvmhv_nested_exit();
5373 module_init(kvmppc_book3s_init_hv);
5374 module_exit(kvmppc_book3s_exit_hv);
5375 MODULE_LICENSE("GPL");
5376 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5377 MODULE_ALIAS("devname:kvm");
projects / linux.git / blob