1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
7 * Paul Mackerras <paulus@au1.ibm.com>
8 * Alexander Graf <agraf@suse.de>
9 * Kevin Wolf <mail@kevin-wolf.de>
11 * Description: KVM functions specific to running on Book 3S
12 * processors in hypervisor mode (specifically POWER7 and later).
14 * This file is derived from arch/powerpc/kvm/book3s.c,
15 * by Alexander Graf <agraf@suse.de>.
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
46 #include <asm/ftrace.h>
48 #include <asm/ppc-opcode.h>
49 #include <asm/asm-prototypes.h>
50 #include <asm/archrandom.h>
51 #include <asm/debug.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <linux/uaccess.h>
57 #include <asm/kvm_ppc.h>
58 #include <asm/kvm_book3s.h>
59 #include <asm/mmu_context.h>
60 #include <asm/lppaca.h>
61 #include <asm/processor.h>
62 #include <asm/cputhreads.h>
64 #include <asm/hvcall.h>
65 #include <asm/switch_to.h>
67 #include <asm/dbell.h>
69 #include <asm/pnv-pci.h>
74 #include <asm/hw_breakpoint.h>
75 #include <asm/kvm_host.h>
76 #include <asm/kvm_book3s_uvmem.h>
77 #include <asm/ultravisor.h>
81 #define CREATE_TRACE_POINTS
84 /* #define EXIT_DEBUG */
85 /* #define EXIT_DEBUG_SIMPLE */
86 /* #define EXIT_DEBUG_INT */
88 /* Used to indicate that a guest page fault needs to be handled */
89 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
90 /* Used to indicate that a guest passthrough interrupt needs to be handled */
91 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
93 /* Used as a "null" value for timebase values */
94 #define TB_NIL (~(u64)0)
96 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
98 static int dynamic_mt_modes = 6;
99 module_param(dynamic_mt_modes, int, 0644);
100 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
101 static int target_smt_mode;
102 module_param(target_smt_mode, int, 0644);
103 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
105 static bool indep_threads_mode = true;
106 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
107 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
109 static bool one_vm_per_core;
110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
113 #ifdef CONFIG_KVM_XICS
114 static struct kernel_param_ops module_param_ops = {
115 .set = param_set_int,
116 .get = param_get_int,
119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
126 /* If set, guests are allowed to create and control nested guests */
127 static bool nested = true;
128 module_param(nested, bool, S_IRUGO | S_IWUSR);
129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
131 static inline bool nesting_enabled(struct kvm *kvm)
133 return kvm->arch.nested_enable && kvm_is_radix(kvm);
136 /* If set, the threads on each CPU core have to be in the same MMU mode */
137 static bool no_mixing_hpt_and_radix;
139 static 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_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
345 vcpu->arch.pvr = pvr;
348 /* Dummy value used in computing PCR value below */
349 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
351 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
353 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
354 struct kvmppc_vcore *vc = vcpu->arch.vcore;
356 /* We can (emulate) our own architecture version and anything older */
357 if (cpu_has_feature(CPU_FTR_ARCH_300))
358 host_pcr_bit = PCR_ARCH_300;
359 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
360 host_pcr_bit = PCR_ARCH_207;
361 else if (cpu_has_feature(CPU_FTR_ARCH_206))
362 host_pcr_bit = PCR_ARCH_206;
364 host_pcr_bit = PCR_ARCH_205;
366 /* Determine lowest PCR bit needed to run guest in given PVR level */
367 guest_pcr_bit = host_pcr_bit;
369 switch (arch_compat) {
371 guest_pcr_bit = PCR_ARCH_205;
375 guest_pcr_bit = PCR_ARCH_206;
378 guest_pcr_bit = PCR_ARCH_207;
381 guest_pcr_bit = PCR_ARCH_300;
388 /* Check requested PCR bits don't exceed our capabilities */
389 if (guest_pcr_bit > host_pcr_bit)
392 spin_lock(&vc->lock);
393 vc->arch_compat = arch_compat;
395 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
396 * Also set all reserved PCR bits
398 vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
399 spin_unlock(&vc->lock);
404 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
408 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
409 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
410 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
411 for (r = 0; r < 16; ++r)
412 pr_err("r%2d = %.16lx r%d = %.16lx\n",
413 r, kvmppc_get_gpr(vcpu, r),
414 r+16, kvmppc_get_gpr(vcpu, r+16));
415 pr_err("ctr = %.16lx lr = %.16lx\n",
416 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
417 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
418 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
419 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
420 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
421 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
422 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
423 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
424 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
425 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
426 pr_err("fault dar = %.16lx dsisr = %.8x\n",
427 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
428 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
429 for (r = 0; r < vcpu->arch.slb_max; ++r)
430 pr_err(" ESID = %.16llx VSID = %.16llx\n",
431 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
432 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
433 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
434 vcpu->arch.last_inst);
437 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
439 return kvm_get_vcpu_by_id(kvm, id);
442 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
444 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
445 vpa->yield_count = cpu_to_be32(1);
448 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
449 unsigned long addr, unsigned long len)
451 /* check address is cacheline aligned */
452 if (addr & (L1_CACHE_BYTES - 1))
454 spin_lock(&vcpu->arch.vpa_update_lock);
455 if (v->next_gpa != addr || v->len != len) {
457 v->len = addr ? len : 0;
458 v->update_pending = 1;
460 spin_unlock(&vcpu->arch.vpa_update_lock);
464 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
473 static int vpa_is_registered(struct kvmppc_vpa *vpap)
475 if (vpap->update_pending)
476 return vpap->next_gpa != 0;
477 return vpap->pinned_addr != NULL;
480 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
482 unsigned long vcpuid, unsigned long vpa)
484 struct kvm *kvm = vcpu->kvm;
485 unsigned long len, nb;
487 struct kvm_vcpu *tvcpu;
490 struct kvmppc_vpa *vpap;
492 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
496 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
497 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
498 subfunc == H_VPA_REG_SLB) {
499 /* Registering new area - address must be cache-line aligned */
500 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
503 /* convert logical addr to kernel addr and read length */
504 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
507 if (subfunc == H_VPA_REG_VPA)
508 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
510 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
511 kvmppc_unpin_guest_page(kvm, va, vpa, false);
514 if (len > nb || len < sizeof(struct reg_vpa))
523 spin_lock(&tvcpu->arch.vpa_update_lock);
526 case H_VPA_REG_VPA: /* register VPA */
528 * The size of our lppaca is 1kB because of the way we align
529 * it for the guest to avoid crossing a 4kB boundary. We only
530 * use 640 bytes of the structure though, so we should accept
531 * clients that set a size of 640.
533 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
534 if (len < sizeof(struct lppaca))
536 vpap = &tvcpu->arch.vpa;
540 case H_VPA_REG_DTL: /* register DTL */
541 if (len < sizeof(struct dtl_entry))
543 len -= len % sizeof(struct dtl_entry);
545 /* Check that they have previously registered a VPA */
547 if (!vpa_is_registered(&tvcpu->arch.vpa))
550 vpap = &tvcpu->arch.dtl;
554 case H_VPA_REG_SLB: /* register SLB shadow buffer */
555 /* Check that they have previously registered a VPA */
557 if (!vpa_is_registered(&tvcpu->arch.vpa))
560 vpap = &tvcpu->arch.slb_shadow;
564 case H_VPA_DEREG_VPA: /* deregister VPA */
565 /* Check they don't still have a DTL or SLB buf registered */
567 if (vpa_is_registered(&tvcpu->arch.dtl) ||
568 vpa_is_registered(&tvcpu->arch.slb_shadow))
571 vpap = &tvcpu->arch.vpa;
575 case H_VPA_DEREG_DTL: /* deregister DTL */
576 vpap = &tvcpu->arch.dtl;
580 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
581 vpap = &tvcpu->arch.slb_shadow;
587 vpap->next_gpa = vpa;
589 vpap->update_pending = 1;
592 spin_unlock(&tvcpu->arch.vpa_update_lock);
597 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
599 struct kvm *kvm = vcpu->kvm;
605 * We need to pin the page pointed to by vpap->next_gpa,
606 * but we can't call kvmppc_pin_guest_page under the lock
607 * as it does get_user_pages() and down_read(). So we
608 * have to drop the lock, pin the page, then get the lock
609 * again and check that a new area didn't get registered
613 gpa = vpap->next_gpa;
614 spin_unlock(&vcpu->arch.vpa_update_lock);
618 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
619 spin_lock(&vcpu->arch.vpa_update_lock);
620 if (gpa == vpap->next_gpa)
622 /* sigh... unpin that one and try again */
624 kvmppc_unpin_guest_page(kvm, va, gpa, false);
627 vpap->update_pending = 0;
628 if (va && nb < vpap->len) {
630 * If it's now too short, it must be that userspace
631 * has changed the mappings underlying guest memory,
632 * so unregister the region.
634 kvmppc_unpin_guest_page(kvm, va, gpa, false);
637 if (vpap->pinned_addr)
638 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
641 vpap->pinned_addr = va;
644 vpap->pinned_end = va + vpap->len;
647 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
649 if (!(vcpu->arch.vpa.update_pending ||
650 vcpu->arch.slb_shadow.update_pending ||
651 vcpu->arch.dtl.update_pending))
654 spin_lock(&vcpu->arch.vpa_update_lock);
655 if (vcpu->arch.vpa.update_pending) {
656 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
657 if (vcpu->arch.vpa.pinned_addr)
658 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
660 if (vcpu->arch.dtl.update_pending) {
661 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
662 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
663 vcpu->arch.dtl_index = 0;
665 if (vcpu->arch.slb_shadow.update_pending)
666 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
667 spin_unlock(&vcpu->arch.vpa_update_lock);
671 * Return the accumulated stolen time for the vcore up until `now'.
672 * The caller should hold the vcore lock.
674 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
679 spin_lock_irqsave(&vc->stoltb_lock, flags);
681 if (vc->vcore_state != VCORE_INACTIVE &&
682 vc->preempt_tb != TB_NIL)
683 p += now - vc->preempt_tb;
684 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
688 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
689 struct kvmppc_vcore *vc)
691 struct dtl_entry *dt;
693 unsigned long stolen;
694 unsigned long core_stolen;
698 dt = vcpu->arch.dtl_ptr;
699 vpa = vcpu->arch.vpa.pinned_addr;
701 core_stolen = vcore_stolen_time(vc, now);
702 stolen = core_stolen - vcpu->arch.stolen_logged;
703 vcpu->arch.stolen_logged = core_stolen;
704 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
705 stolen += vcpu->arch.busy_stolen;
706 vcpu->arch.busy_stolen = 0;
707 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
710 memset(dt, 0, sizeof(struct dtl_entry));
711 dt->dispatch_reason = 7;
712 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
713 dt->timebase = cpu_to_be64(now + vc->tb_offset);
714 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
715 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
716 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
718 if (dt == vcpu->arch.dtl.pinned_end)
719 dt = vcpu->arch.dtl.pinned_addr;
720 vcpu->arch.dtl_ptr = dt;
721 /* order writing *dt vs. writing vpa->dtl_idx */
723 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
724 vcpu->arch.dtl.dirty = true;
727 /* See if there is a doorbell interrupt pending for a vcpu */
728 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
731 struct kvmppc_vcore *vc;
733 if (vcpu->arch.doorbell_request)
736 * Ensure that the read of vcore->dpdes comes after the read
737 * of vcpu->doorbell_request. This barrier matches the
738 * smp_wmb() in kvmppc_guest_entry_inject().
741 vc = vcpu->arch.vcore;
742 thr = vcpu->vcpu_id - vc->first_vcpuid;
743 return !!(vc->dpdes & (1 << thr));
746 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
748 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
750 if ((!vcpu->arch.vcore->arch_compat) &&
751 cpu_has_feature(CPU_FTR_ARCH_207S))
756 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
757 unsigned long resource, unsigned long value1,
758 unsigned long value2)
761 case H_SET_MODE_RESOURCE_SET_CIABR:
762 if (!kvmppc_power8_compatible(vcpu))
767 return H_UNSUPPORTED_FLAG_START;
768 /* Guests can't breakpoint the hypervisor */
769 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
771 vcpu->arch.ciabr = value1;
773 case H_SET_MODE_RESOURCE_SET_DAWR:
774 if (!kvmppc_power8_compatible(vcpu))
776 if (!ppc_breakpoint_available())
779 return H_UNSUPPORTED_FLAG_START;
780 if (value2 & DABRX_HYP)
782 vcpu->arch.dawr = value1;
783 vcpu->arch.dawrx = value2;
785 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
786 /* KVM does not support mflags=2 (AIL=2) */
787 if (mflags != 0 && mflags != 3)
788 return H_UNSUPPORTED_FLAG_START;
795 /* Copy guest memory in place - must reside within a single memslot */
796 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
799 struct kvm_memory_slot *to_memslot = NULL;
800 struct kvm_memory_slot *from_memslot = NULL;
801 unsigned long to_addr, from_addr;
804 /* Get HPA for from address */
805 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
808 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
811 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
812 if (kvm_is_error_hva(from_addr))
814 from_addr |= (from & (PAGE_SIZE - 1));
816 /* Get HPA for to address */
817 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
820 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
823 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
824 if (kvm_is_error_hva(to_addr))
826 to_addr |= (to & (PAGE_SIZE - 1));
829 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
833 mark_page_dirty(kvm, to >> PAGE_SHIFT);
837 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
838 unsigned long dest, unsigned long src)
840 u64 pg_sz = SZ_4K; /* 4K page size */
841 u64 pg_mask = SZ_4K - 1;
844 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
845 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
846 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
849 /* dest (and src if copy_page flag set) must be page aligned */
850 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
853 /* zero and/or copy the page as determined by the flags */
854 if (flags & H_COPY_PAGE) {
855 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
858 } else if (flags & H_ZERO_PAGE) {
859 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
864 /* We can ignore the remaining flags */
869 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
871 struct kvmppc_vcore *vcore = target->arch.vcore;
874 * We expect to have been called by the real mode handler
875 * (kvmppc_rm_h_confer()) which would have directly returned
876 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
877 * have useful work to do and should not confer) so we don't
881 spin_lock(&vcore->lock);
882 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
883 vcore->vcore_state != VCORE_INACTIVE &&
885 target = vcore->runner;
886 spin_unlock(&vcore->lock);
888 return kvm_vcpu_yield_to(target);
891 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
894 struct lppaca *lppaca;
896 spin_lock(&vcpu->arch.vpa_update_lock);
897 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
899 yield_count = be32_to_cpu(lppaca->yield_count);
900 spin_unlock(&vcpu->arch.vpa_update_lock);
904 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
906 unsigned long req = kvmppc_get_gpr(vcpu, 3);
907 unsigned long target, ret = H_SUCCESS;
909 struct kvm_vcpu *tvcpu;
912 if (req <= MAX_HCALL_OPCODE &&
913 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
920 target = kvmppc_get_gpr(vcpu, 4);
921 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
926 tvcpu->arch.prodded = 1;
928 if (tvcpu->arch.ceded)
929 kvmppc_fast_vcpu_kick_hv(tvcpu);
932 target = kvmppc_get_gpr(vcpu, 4);
935 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
940 yield_count = kvmppc_get_gpr(vcpu, 5);
941 if (kvmppc_get_yield_count(tvcpu) != yield_count)
943 kvm_arch_vcpu_yield_to(tvcpu);
946 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
947 kvmppc_get_gpr(vcpu, 5),
948 kvmppc_get_gpr(vcpu, 6));
951 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
954 idx = srcu_read_lock(&vcpu->kvm->srcu);
955 rc = kvmppc_rtas_hcall(vcpu);
956 srcu_read_unlock(&vcpu->kvm->srcu, idx);
963 /* Send the error out to userspace via KVM_RUN */
965 case H_LOGICAL_CI_LOAD:
966 ret = kvmppc_h_logical_ci_load(vcpu);
967 if (ret == H_TOO_HARD)
970 case H_LOGICAL_CI_STORE:
971 ret = kvmppc_h_logical_ci_store(vcpu);
972 if (ret == H_TOO_HARD)
976 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
977 kvmppc_get_gpr(vcpu, 5),
978 kvmppc_get_gpr(vcpu, 6),
979 kvmppc_get_gpr(vcpu, 7));
980 if (ret == H_TOO_HARD)
989 if (kvmppc_xics_enabled(vcpu)) {
990 if (xics_on_xive()) {
991 ret = H_NOT_AVAILABLE;
994 ret = kvmppc_xics_hcall(vcpu, req);
999 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1002 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1003 kvmppc_get_gpr(vcpu, 5));
1005 #ifdef CONFIG_SPAPR_TCE_IOMMU
1007 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1008 kvmppc_get_gpr(vcpu, 5));
1009 if (ret == H_TOO_HARD)
1013 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1014 kvmppc_get_gpr(vcpu, 5),
1015 kvmppc_get_gpr(vcpu, 6));
1016 if (ret == H_TOO_HARD)
1019 case H_PUT_TCE_INDIRECT:
1020 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1021 kvmppc_get_gpr(vcpu, 5),
1022 kvmppc_get_gpr(vcpu, 6),
1023 kvmppc_get_gpr(vcpu, 7));
1024 if (ret == H_TOO_HARD)
1028 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1029 kvmppc_get_gpr(vcpu, 5),
1030 kvmppc_get_gpr(vcpu, 6),
1031 kvmppc_get_gpr(vcpu, 7));
1032 if (ret == H_TOO_HARD)
1037 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1041 case H_SET_PARTITION_TABLE:
1043 if (nesting_enabled(vcpu->kvm))
1044 ret = kvmhv_set_partition_table(vcpu);
1046 case H_ENTER_NESTED:
1048 if (!nesting_enabled(vcpu->kvm))
1050 ret = kvmhv_enter_nested_guest(vcpu);
1051 if (ret == H_INTERRUPT) {
1052 kvmppc_set_gpr(vcpu, 3, 0);
1053 vcpu->arch.hcall_needed = 0;
1055 } else if (ret == H_TOO_HARD) {
1056 kvmppc_set_gpr(vcpu, 3, 0);
1057 vcpu->arch.hcall_needed = 0;
1061 case H_TLB_INVALIDATE:
1063 if (nesting_enabled(vcpu->kvm))
1064 ret = kvmhv_do_nested_tlbie(vcpu);
1066 case H_COPY_TOFROM_GUEST:
1068 if (nesting_enabled(vcpu->kvm))
1069 ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1072 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1073 kvmppc_get_gpr(vcpu, 5),
1074 kvmppc_get_gpr(vcpu, 6));
1077 ret = kvmppc_h_svm_page_in(vcpu->kvm,
1078 kvmppc_get_gpr(vcpu, 4),
1079 kvmppc_get_gpr(vcpu, 5),
1080 kvmppc_get_gpr(vcpu, 6));
1082 case H_SVM_PAGE_OUT:
1083 ret = kvmppc_h_svm_page_out(vcpu->kvm,
1084 kvmppc_get_gpr(vcpu, 4),
1085 kvmppc_get_gpr(vcpu, 5),
1086 kvmppc_get_gpr(vcpu, 6));
1088 case H_SVM_INIT_START:
1089 ret = kvmppc_h_svm_init_start(vcpu->kvm);
1091 case H_SVM_INIT_DONE:
1092 ret = kvmppc_h_svm_init_done(vcpu->kvm);
1098 kvmppc_set_gpr(vcpu, 3, ret);
1099 vcpu->arch.hcall_needed = 0;
1100 return RESUME_GUEST;
1104 * Handle H_CEDE in the nested virtualization case where we haven't
1105 * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1106 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1107 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1109 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1111 vcpu->arch.shregs.msr |= MSR_EE;
1112 vcpu->arch.ceded = 1;
1114 if (vcpu->arch.prodded) {
1115 vcpu->arch.prodded = 0;
1117 vcpu->arch.ceded = 0;
1121 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1127 case H_REGISTER_VPA:
1129 case H_LOGICAL_CI_LOAD:
1130 case H_LOGICAL_CI_STORE:
1131 #ifdef CONFIG_KVM_XICS
1143 /* See if it's in the real-mode table */
1144 return kvmppc_hcall_impl_hv_realmode(cmd);
1147 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1148 struct kvm_vcpu *vcpu)
1152 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1155 * Fetch failed, so return to guest and
1156 * try executing it again.
1158 return RESUME_GUEST;
1161 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1162 run->exit_reason = KVM_EXIT_DEBUG;
1163 run->debug.arch.address = kvmppc_get_pc(vcpu);
1166 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1167 return RESUME_GUEST;
1171 static void do_nothing(void *x)
1175 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1177 int thr, cpu, pcpu, nthreads;
1179 unsigned long dpdes;
1181 nthreads = vcpu->kvm->arch.emul_smt_mode;
1183 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1184 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1185 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1189 * If the vcpu is currently running on a physical cpu thread,
1190 * interrupt it in order to pull it out of the guest briefly,
1191 * which will update its vcore->dpdes value.
1193 pcpu = READ_ONCE(v->cpu);
1195 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1196 if (kvmppc_doorbell_pending(v))
1203 * On POWER9, emulate doorbell-related instructions in order to
1204 * give the guest the illusion of running on a multi-threaded core.
1205 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1208 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1212 struct kvm *kvm = vcpu->kvm;
1213 struct kvm_vcpu *tvcpu;
1215 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1216 return RESUME_GUEST;
1217 if (get_op(inst) != 31)
1218 return EMULATE_FAIL;
1220 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1221 switch (get_xop(inst)) {
1222 case OP_31_XOP_MSGSNDP:
1223 arg = kvmppc_get_gpr(vcpu, rb);
1224 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1227 if (arg >= kvm->arch.emul_smt_mode)
1229 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1232 if (!tvcpu->arch.doorbell_request) {
1233 tvcpu->arch.doorbell_request = 1;
1234 kvmppc_fast_vcpu_kick_hv(tvcpu);
1237 case OP_31_XOP_MSGCLRP:
1238 arg = kvmppc_get_gpr(vcpu, rb);
1239 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1241 vcpu->arch.vcore->dpdes = 0;
1242 vcpu->arch.doorbell_request = 0;
1244 case OP_31_XOP_MFSPR:
1245 switch (get_sprn(inst)) {
1250 arg = kvmppc_read_dpdes(vcpu);
1253 return EMULATE_FAIL;
1255 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1258 return EMULATE_FAIL;
1260 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1261 return RESUME_GUEST;
1264 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1265 struct task_struct *tsk)
1267 int r = RESUME_HOST;
1269 vcpu->stat.sum_exits++;
1272 * This can happen if an interrupt occurs in the last stages
1273 * of guest entry or the first stages of guest exit (i.e. after
1274 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1275 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1276 * That can happen due to a bug, or due to a machine check
1277 * occurring at just the wrong time.
1279 if (vcpu->arch.shregs.msr & MSR_HV) {
1280 printk(KERN_EMERG "KVM trap in HV mode!\n");
1281 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1282 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1283 vcpu->arch.shregs.msr);
1284 kvmppc_dump_regs(vcpu);
1285 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1286 run->hw.hardware_exit_reason = vcpu->arch.trap;
1289 run->exit_reason = KVM_EXIT_UNKNOWN;
1290 run->ready_for_interrupt_injection = 1;
1291 switch (vcpu->arch.trap) {
1292 /* We're good on these - the host merely wanted to get our attention */
1293 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1294 vcpu->stat.dec_exits++;
1297 case BOOK3S_INTERRUPT_EXTERNAL:
1298 case BOOK3S_INTERRUPT_H_DOORBELL:
1299 case BOOK3S_INTERRUPT_H_VIRT:
1300 vcpu->stat.ext_intr_exits++;
1303 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1304 case BOOK3S_INTERRUPT_HMI:
1305 case BOOK3S_INTERRUPT_PERFMON:
1306 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1309 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1310 /* Print the MCE event to host console. */
1311 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1314 * If the guest can do FWNMI, exit to userspace so it can
1315 * deliver a FWNMI to the guest.
1316 * Otherwise we synthesize a machine check for the guest
1317 * so that it knows that the machine check occurred.
1319 if (!vcpu->kvm->arch.fwnmi_enabled) {
1320 ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1321 kvmppc_core_queue_machine_check(vcpu, flags);
1326 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1327 run->exit_reason = KVM_EXIT_NMI;
1328 run->hw.hardware_exit_reason = vcpu->arch.trap;
1329 /* Clear out the old NMI status from run->flags */
1330 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1331 /* Now set the NMI status */
1332 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1333 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1335 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1339 case BOOK3S_INTERRUPT_PROGRAM:
1343 * Normally program interrupts are delivered directly
1344 * to the guest by the hardware, but we can get here
1345 * as a result of a hypervisor emulation interrupt
1346 * (e40) getting turned into a 700 by BML RTAS.
1348 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1349 kvmppc_core_queue_program(vcpu, flags);
1353 case BOOK3S_INTERRUPT_SYSCALL:
1355 /* hcall - punt to userspace */
1358 /* hypercall with MSR_PR has already been handled in rmode,
1359 * and never reaches here.
1362 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1363 for (i = 0; i < 9; ++i)
1364 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1365 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1366 vcpu->arch.hcall_needed = 1;
1371 * We get these next two if the guest accesses a page which it thinks
1372 * it has mapped but which is not actually present, either because
1373 * it is for an emulated I/O device or because the corresonding
1374 * host page has been paged out. Any other HDSI/HISI interrupts
1375 * have been handled already.
1377 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1378 r = RESUME_PAGE_FAULT;
1380 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1381 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1382 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1383 DSISR_SRR1_MATCH_64S;
1384 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1385 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1386 r = RESUME_PAGE_FAULT;
1389 * This occurs if the guest executes an illegal instruction.
1390 * If the guest debug is disabled, generate a program interrupt
1391 * to the guest. If guest debug is enabled, we need to check
1392 * whether the instruction is a software breakpoint instruction.
1393 * Accordingly return to Guest or Host.
1395 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1396 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1397 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1398 swab32(vcpu->arch.emul_inst) :
1399 vcpu->arch.emul_inst;
1400 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1401 r = kvmppc_emulate_debug_inst(run, vcpu);
1403 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1408 * This occurs if the guest (kernel or userspace), does something that
1409 * is prohibited by HFSCR.
1410 * On POWER9, this could be a doorbell instruction that we need
1412 * Otherwise, we just generate a program interrupt to the guest.
1414 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1416 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1417 cpu_has_feature(CPU_FTR_ARCH_300))
1418 r = kvmppc_emulate_doorbell_instr(vcpu);
1419 if (r == EMULATE_FAIL) {
1420 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1425 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1426 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1428 * This occurs for various TM-related instructions that
1429 * we need to emulate on POWER9 DD2.2. We have already
1430 * handled the cases where the guest was in real-suspend
1431 * mode and was transitioning to transactional state.
1433 r = kvmhv_p9_tm_emulation(vcpu);
1437 case BOOK3S_INTERRUPT_HV_RM_HARD:
1438 r = RESUME_PASSTHROUGH;
1441 kvmppc_dump_regs(vcpu);
1442 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1443 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1444 vcpu->arch.shregs.msr);
1445 run->hw.hardware_exit_reason = vcpu->arch.trap;
1453 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1458 vcpu->stat.sum_exits++;
1461 * This can happen if an interrupt occurs in the last stages
1462 * of guest entry or the first stages of guest exit (i.e. after
1463 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1464 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1465 * That can happen due to a bug, or due to a machine check
1466 * occurring at just the wrong time.
1468 if (vcpu->arch.shregs.msr & MSR_HV) {
1469 pr_emerg("KVM trap in HV mode while nested!\n");
1470 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1471 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1472 vcpu->arch.shregs.msr);
1473 kvmppc_dump_regs(vcpu);
1476 switch (vcpu->arch.trap) {
1477 /* We're good on these - the host merely wanted to get our attention */
1478 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1479 vcpu->stat.dec_exits++;
1482 case BOOK3S_INTERRUPT_EXTERNAL:
1483 vcpu->stat.ext_intr_exits++;
1486 case BOOK3S_INTERRUPT_H_DOORBELL:
1487 case BOOK3S_INTERRUPT_H_VIRT:
1488 vcpu->stat.ext_intr_exits++;
1491 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1492 case BOOK3S_INTERRUPT_HMI:
1493 case BOOK3S_INTERRUPT_PERFMON:
1494 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1497 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1498 /* Pass the machine check to the L1 guest */
1500 /* Print the MCE event to host console. */
1501 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1504 * We get these next two if the guest accesses a page which it thinks
1505 * it has mapped but which is not actually present, either because
1506 * it is for an emulated I/O device or because the corresonding
1507 * host page has been paged out.
1509 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1510 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1511 r = kvmhv_nested_page_fault(run, vcpu);
1512 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1514 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1515 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1516 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1517 DSISR_SRR1_MATCH_64S;
1518 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1519 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1520 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1521 r = kvmhv_nested_page_fault(run, vcpu);
1522 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1525 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1526 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1528 * This occurs for various TM-related instructions that
1529 * we need to emulate on POWER9 DD2.2. We have already
1530 * handled the cases where the guest was in real-suspend
1531 * mode and was transitioning to transactional state.
1533 r = kvmhv_p9_tm_emulation(vcpu);
1537 case BOOK3S_INTERRUPT_HV_RM_HARD:
1538 vcpu->arch.trap = 0;
1540 if (!xics_on_xive())
1541 kvmppc_xics_rm_complete(vcpu, 0);
1551 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1552 struct kvm_sregs *sregs)
1556 memset(sregs, 0, sizeof(struct kvm_sregs));
1557 sregs->pvr = vcpu->arch.pvr;
1558 for (i = 0; i < vcpu->arch.slb_max; i++) {
1559 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1560 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1566 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1567 struct kvm_sregs *sregs)
1571 /* Only accept the same PVR as the host's, since we can't spoof it */
1572 if (sregs->pvr != vcpu->arch.pvr)
1576 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1577 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1578 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1579 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1583 vcpu->arch.slb_max = j;
1588 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1589 bool preserve_top32)
1591 struct kvm *kvm = vcpu->kvm;
1592 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1595 spin_lock(&vc->lock);
1597 * If ILE (interrupt little-endian) has changed, update the
1598 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1600 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1601 struct kvm_vcpu *vcpu;
1604 kvm_for_each_vcpu(i, vcpu, kvm) {
1605 if (vcpu->arch.vcore != vc)
1607 if (new_lpcr & LPCR_ILE)
1608 vcpu->arch.intr_msr |= MSR_LE;
1610 vcpu->arch.intr_msr &= ~MSR_LE;
1615 * Userspace can only modify DPFD (default prefetch depth),
1616 * ILE (interrupt little-endian) and TC (translation control).
1617 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1619 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1620 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1623 * On POWER9, allow userspace to enable large decrementer for the
1624 * guest, whether or not the host has it enabled.
1626 if (cpu_has_feature(CPU_FTR_ARCH_300))
1629 /* Broken 32-bit version of LPCR must not clear top bits */
1632 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1633 spin_unlock(&vc->lock);
1636 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1637 union kvmppc_one_reg *val)
1643 case KVM_REG_PPC_DEBUG_INST:
1644 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1646 case KVM_REG_PPC_HIOR:
1647 *val = get_reg_val(id, 0);
1649 case KVM_REG_PPC_DABR:
1650 *val = get_reg_val(id, vcpu->arch.dabr);
1652 case KVM_REG_PPC_DABRX:
1653 *val = get_reg_val(id, vcpu->arch.dabrx);
1655 case KVM_REG_PPC_DSCR:
1656 *val = get_reg_val(id, vcpu->arch.dscr);
1658 case KVM_REG_PPC_PURR:
1659 *val = get_reg_val(id, vcpu->arch.purr);
1661 case KVM_REG_PPC_SPURR:
1662 *val = get_reg_val(id, vcpu->arch.spurr);
1664 case KVM_REG_PPC_AMR:
1665 *val = get_reg_val(id, vcpu->arch.amr);
1667 case KVM_REG_PPC_UAMOR:
1668 *val = get_reg_val(id, vcpu->arch.uamor);
1670 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1671 i = id - KVM_REG_PPC_MMCR0;
1672 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1674 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1675 i = id - KVM_REG_PPC_PMC1;
1676 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1678 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1679 i = id - KVM_REG_PPC_SPMC1;
1680 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1682 case KVM_REG_PPC_SIAR:
1683 *val = get_reg_val(id, vcpu->arch.siar);
1685 case KVM_REG_PPC_SDAR:
1686 *val = get_reg_val(id, vcpu->arch.sdar);
1688 case KVM_REG_PPC_SIER:
1689 *val = get_reg_val(id, vcpu->arch.sier);
1691 case KVM_REG_PPC_IAMR:
1692 *val = get_reg_val(id, vcpu->arch.iamr);
1694 case KVM_REG_PPC_PSPB:
1695 *val = get_reg_val(id, vcpu->arch.pspb);
1697 case KVM_REG_PPC_DPDES:
1699 * On POWER9, where we are emulating msgsndp etc.,
1700 * we return 1 bit for each vcpu, which can come from
1701 * either vcore->dpdes or doorbell_request.
1702 * On POWER8, doorbell_request is 0.
1704 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1705 vcpu->arch.doorbell_request);
1707 case KVM_REG_PPC_VTB:
1708 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1710 case KVM_REG_PPC_DAWR:
1711 *val = get_reg_val(id, vcpu->arch.dawr);
1713 case KVM_REG_PPC_DAWRX:
1714 *val = get_reg_val(id, vcpu->arch.dawrx);
1716 case KVM_REG_PPC_CIABR:
1717 *val = get_reg_val(id, vcpu->arch.ciabr);
1719 case KVM_REG_PPC_CSIGR:
1720 *val = get_reg_val(id, vcpu->arch.csigr);
1722 case KVM_REG_PPC_TACR:
1723 *val = get_reg_val(id, vcpu->arch.tacr);
1725 case KVM_REG_PPC_TCSCR:
1726 *val = get_reg_val(id, vcpu->arch.tcscr);
1728 case KVM_REG_PPC_PID:
1729 *val = get_reg_val(id, vcpu->arch.pid);
1731 case KVM_REG_PPC_ACOP:
1732 *val = get_reg_val(id, vcpu->arch.acop);
1734 case KVM_REG_PPC_WORT:
1735 *val = get_reg_val(id, vcpu->arch.wort);
1737 case KVM_REG_PPC_TIDR:
1738 *val = get_reg_val(id, vcpu->arch.tid);
1740 case KVM_REG_PPC_PSSCR:
1741 *val = get_reg_val(id, vcpu->arch.psscr);
1743 case KVM_REG_PPC_VPA_ADDR:
1744 spin_lock(&vcpu->arch.vpa_update_lock);
1745 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1746 spin_unlock(&vcpu->arch.vpa_update_lock);
1748 case KVM_REG_PPC_VPA_SLB:
1749 spin_lock(&vcpu->arch.vpa_update_lock);
1750 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1751 val->vpaval.length = vcpu->arch.slb_shadow.len;
1752 spin_unlock(&vcpu->arch.vpa_update_lock);
1754 case KVM_REG_PPC_VPA_DTL:
1755 spin_lock(&vcpu->arch.vpa_update_lock);
1756 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1757 val->vpaval.length = vcpu->arch.dtl.len;
1758 spin_unlock(&vcpu->arch.vpa_update_lock);
1760 case KVM_REG_PPC_TB_OFFSET:
1761 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1763 case KVM_REG_PPC_LPCR:
1764 case KVM_REG_PPC_LPCR_64:
1765 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1767 case KVM_REG_PPC_PPR:
1768 *val = get_reg_val(id, vcpu->arch.ppr);
1770 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1771 case KVM_REG_PPC_TFHAR:
1772 *val = get_reg_val(id, vcpu->arch.tfhar);
1774 case KVM_REG_PPC_TFIAR:
1775 *val = get_reg_val(id, vcpu->arch.tfiar);
1777 case KVM_REG_PPC_TEXASR:
1778 *val = get_reg_val(id, vcpu->arch.texasr);
1780 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1781 i = id - KVM_REG_PPC_TM_GPR0;
1782 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1784 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1787 i = id - KVM_REG_PPC_TM_VSR0;
1789 for (j = 0; j < TS_FPRWIDTH; j++)
1790 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1792 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1793 val->vval = vcpu->arch.vr_tm.vr[i-32];
1799 case KVM_REG_PPC_TM_CR:
1800 *val = get_reg_val(id, vcpu->arch.cr_tm);
1802 case KVM_REG_PPC_TM_XER:
1803 *val = get_reg_val(id, vcpu->arch.xer_tm);
1805 case KVM_REG_PPC_TM_LR:
1806 *val = get_reg_val(id, vcpu->arch.lr_tm);
1808 case KVM_REG_PPC_TM_CTR:
1809 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1811 case KVM_REG_PPC_TM_FPSCR:
1812 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1814 case KVM_REG_PPC_TM_AMR:
1815 *val = get_reg_val(id, vcpu->arch.amr_tm);
1817 case KVM_REG_PPC_TM_PPR:
1818 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1820 case KVM_REG_PPC_TM_VRSAVE:
1821 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1823 case KVM_REG_PPC_TM_VSCR:
1824 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1825 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1829 case KVM_REG_PPC_TM_DSCR:
1830 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1832 case KVM_REG_PPC_TM_TAR:
1833 *val = get_reg_val(id, vcpu->arch.tar_tm);
1836 case KVM_REG_PPC_ARCH_COMPAT:
1837 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1839 case KVM_REG_PPC_DEC_EXPIRY:
1840 *val = get_reg_val(id, vcpu->arch.dec_expires +
1841 vcpu->arch.vcore->tb_offset);
1843 case KVM_REG_PPC_ONLINE:
1844 *val = get_reg_val(id, vcpu->arch.online);
1846 case KVM_REG_PPC_PTCR:
1847 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1857 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1858 union kvmppc_one_reg *val)
1862 unsigned long addr, len;
1865 case KVM_REG_PPC_HIOR:
1866 /* Only allow this to be set to zero */
1867 if (set_reg_val(id, *val))
1870 case KVM_REG_PPC_DABR:
1871 vcpu->arch.dabr = set_reg_val(id, *val);
1873 case KVM_REG_PPC_DABRX:
1874 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1876 case KVM_REG_PPC_DSCR:
1877 vcpu->arch.dscr = set_reg_val(id, *val);
1879 case KVM_REG_PPC_PURR:
1880 vcpu->arch.purr = set_reg_val(id, *val);
1882 case KVM_REG_PPC_SPURR:
1883 vcpu->arch.spurr = set_reg_val(id, *val);
1885 case KVM_REG_PPC_AMR:
1886 vcpu->arch.amr = set_reg_val(id, *val);
1888 case KVM_REG_PPC_UAMOR:
1889 vcpu->arch.uamor = set_reg_val(id, *val);
1891 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1892 i = id - KVM_REG_PPC_MMCR0;
1893 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1895 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1896 i = id - KVM_REG_PPC_PMC1;
1897 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1899 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1900 i = id - KVM_REG_PPC_SPMC1;
1901 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1903 case KVM_REG_PPC_SIAR:
1904 vcpu->arch.siar = set_reg_val(id, *val);
1906 case KVM_REG_PPC_SDAR:
1907 vcpu->arch.sdar = set_reg_val(id, *val);
1909 case KVM_REG_PPC_SIER:
1910 vcpu->arch.sier = set_reg_val(id, *val);
1912 case KVM_REG_PPC_IAMR:
1913 vcpu->arch.iamr = set_reg_val(id, *val);
1915 case KVM_REG_PPC_PSPB:
1916 vcpu->arch.pspb = set_reg_val(id, *val);
1918 case KVM_REG_PPC_DPDES:
1919 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1921 case KVM_REG_PPC_VTB:
1922 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1924 case KVM_REG_PPC_DAWR:
1925 vcpu->arch.dawr = set_reg_val(id, *val);
1927 case KVM_REG_PPC_DAWRX:
1928 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1930 case KVM_REG_PPC_CIABR:
1931 vcpu->arch.ciabr = set_reg_val(id, *val);
1932 /* Don't allow setting breakpoints in hypervisor code */
1933 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1934 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1936 case KVM_REG_PPC_CSIGR:
1937 vcpu->arch.csigr = set_reg_val(id, *val);
1939 case KVM_REG_PPC_TACR:
1940 vcpu->arch.tacr = set_reg_val(id, *val);
1942 case KVM_REG_PPC_TCSCR:
1943 vcpu->arch.tcscr = set_reg_val(id, *val);
1945 case KVM_REG_PPC_PID:
1946 vcpu->arch.pid = set_reg_val(id, *val);
1948 case KVM_REG_PPC_ACOP:
1949 vcpu->arch.acop = set_reg_val(id, *val);
1951 case KVM_REG_PPC_WORT:
1952 vcpu->arch.wort = set_reg_val(id, *val);
1954 case KVM_REG_PPC_TIDR:
1955 vcpu->arch.tid = set_reg_val(id, *val);
1957 case KVM_REG_PPC_PSSCR:
1958 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1960 case KVM_REG_PPC_VPA_ADDR:
1961 addr = set_reg_val(id, *val);
1963 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1964 vcpu->arch.dtl.next_gpa))
1966 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1968 case KVM_REG_PPC_VPA_SLB:
1969 addr = val->vpaval.addr;
1970 len = val->vpaval.length;
1972 if (addr && !vcpu->arch.vpa.next_gpa)
1974 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1976 case KVM_REG_PPC_VPA_DTL:
1977 addr = val->vpaval.addr;
1978 len = val->vpaval.length;
1980 if (addr && (len < sizeof(struct dtl_entry) ||
1981 !vcpu->arch.vpa.next_gpa))
1983 len -= len % sizeof(struct dtl_entry);
1984 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1986 case KVM_REG_PPC_TB_OFFSET:
1987 /* round up to multiple of 2^24 */
1988 vcpu->arch.vcore->tb_offset =
1989 ALIGN(set_reg_val(id, *val), 1UL << 24);
1991 case KVM_REG_PPC_LPCR:
1992 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1994 case KVM_REG_PPC_LPCR_64:
1995 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1997 case KVM_REG_PPC_PPR:
1998 vcpu->arch.ppr = set_reg_val(id, *val);
2000 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2001 case KVM_REG_PPC_TFHAR:
2002 vcpu->arch.tfhar = set_reg_val(id, *val);
2004 case KVM_REG_PPC_TFIAR:
2005 vcpu->arch.tfiar = set_reg_val(id, *val);
2007 case KVM_REG_PPC_TEXASR:
2008 vcpu->arch.texasr = set_reg_val(id, *val);
2010 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2011 i = id - KVM_REG_PPC_TM_GPR0;
2012 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2014 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2017 i = id - KVM_REG_PPC_TM_VSR0;
2019 for (j = 0; j < TS_FPRWIDTH; j++)
2020 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2022 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2023 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2028 case KVM_REG_PPC_TM_CR:
2029 vcpu->arch.cr_tm = set_reg_val(id, *val);
2031 case KVM_REG_PPC_TM_XER:
2032 vcpu->arch.xer_tm = set_reg_val(id, *val);
2034 case KVM_REG_PPC_TM_LR:
2035 vcpu->arch.lr_tm = set_reg_val(id, *val);
2037 case KVM_REG_PPC_TM_CTR:
2038 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2040 case KVM_REG_PPC_TM_FPSCR:
2041 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2043 case KVM_REG_PPC_TM_AMR:
2044 vcpu->arch.amr_tm = set_reg_val(id, *val);
2046 case KVM_REG_PPC_TM_PPR:
2047 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2049 case KVM_REG_PPC_TM_VRSAVE:
2050 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2052 case KVM_REG_PPC_TM_VSCR:
2053 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2054 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2058 case KVM_REG_PPC_TM_DSCR:
2059 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2061 case KVM_REG_PPC_TM_TAR:
2062 vcpu->arch.tar_tm = set_reg_val(id, *val);
2065 case KVM_REG_PPC_ARCH_COMPAT:
2066 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2068 case KVM_REG_PPC_DEC_EXPIRY:
2069 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2070 vcpu->arch.vcore->tb_offset;
2072 case KVM_REG_PPC_ONLINE:
2073 i = set_reg_val(id, *val);
2074 if (i && !vcpu->arch.online)
2075 atomic_inc(&vcpu->arch.vcore->online_count);
2076 else if (!i && vcpu->arch.online)
2077 atomic_dec(&vcpu->arch.vcore->online_count);
2078 vcpu->arch.online = i;
2080 case KVM_REG_PPC_PTCR:
2081 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2092 * On POWER9, threads are independent and can be in different partitions.
2093 * Therefore we consider each thread to be a subcore.
2094 * There is a restriction that all threads have to be in the same
2095 * MMU mode (radix or HPT), unfortunately, but since we only support
2096 * HPT guests on a HPT host so far, that isn't an impediment yet.
2098 static int threads_per_vcore(struct kvm *kvm)
2100 if (kvm->arch.threads_indep)
2102 return threads_per_subcore;
2105 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2107 struct kvmppc_vcore *vcore;
2109 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2114 spin_lock_init(&vcore->lock);
2115 spin_lock_init(&vcore->stoltb_lock);
2116 init_swait_queue_head(&vcore->wq);
2117 vcore->preempt_tb = TB_NIL;
2118 vcore->lpcr = kvm->arch.lpcr;
2119 vcore->first_vcpuid = id;
2121 INIT_LIST_HEAD(&vcore->preempt_list);
2126 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2127 static struct debugfs_timings_element {
2131 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2132 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2133 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2134 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2135 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2138 #define N_TIMINGS (ARRAY_SIZE(timings))
2140 struct debugfs_timings_state {
2141 struct kvm_vcpu *vcpu;
2142 unsigned int buflen;
2143 char buf[N_TIMINGS * 100];
2146 static int debugfs_timings_open(struct inode *inode, struct file *file)
2148 struct kvm_vcpu *vcpu = inode->i_private;
2149 struct debugfs_timings_state *p;
2151 p = kzalloc(sizeof(*p), GFP_KERNEL);
2155 kvm_get_kvm(vcpu->kvm);
2157 file->private_data = p;
2159 return nonseekable_open(inode, file);
2162 static int debugfs_timings_release(struct inode *inode, struct file *file)
2164 struct debugfs_timings_state *p = file->private_data;
2166 kvm_put_kvm(p->vcpu->kvm);
2171 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2172 size_t len, loff_t *ppos)
2174 struct debugfs_timings_state *p = file->private_data;
2175 struct kvm_vcpu *vcpu = p->vcpu;
2177 struct kvmhv_tb_accumulator tb;
2186 buf_end = s + sizeof(p->buf);
2187 for (i = 0; i < N_TIMINGS; ++i) {
2188 struct kvmhv_tb_accumulator *acc;
2190 acc = (struct kvmhv_tb_accumulator *)
2191 ((unsigned long)vcpu + timings[i].offset);
2193 for (loops = 0; loops < 1000; ++loops) {
2194 count = acc->seqcount;
2199 if (count == acc->seqcount) {
2207 snprintf(s, buf_end - s, "%s: stuck\n",
2210 snprintf(s, buf_end - s,
2211 "%s: %llu %llu %llu %llu\n",
2212 timings[i].name, count / 2,
2213 tb_to_ns(tb.tb_total),
2214 tb_to_ns(tb.tb_min),
2215 tb_to_ns(tb.tb_max));
2218 p->buflen = s - p->buf;
2222 if (pos >= p->buflen)
2224 if (len > p->buflen - pos)
2225 len = p->buflen - pos;
2226 n = copy_to_user(buf, p->buf + pos, len);
2236 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2237 size_t len, loff_t *ppos)
2242 static const struct file_operations debugfs_timings_ops = {
2243 .owner = THIS_MODULE,
2244 .open = debugfs_timings_open,
2245 .release = debugfs_timings_release,
2246 .read = debugfs_timings_read,
2247 .write = debugfs_timings_write,
2248 .llseek = generic_file_llseek,
2251 /* Create a debugfs directory for the vcpu */
2252 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2255 struct kvm *kvm = vcpu->kvm;
2257 snprintf(buf, sizeof(buf), "vcpu%u", id);
2258 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2260 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2261 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2263 vcpu->arch.debugfs_timings =
2264 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2265 vcpu, &debugfs_timings_ops);
2268 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2269 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2272 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2274 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2277 struct kvm_vcpu *vcpu;
2280 struct kvmppc_vcore *vcore;
2283 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2287 err = kvm_vcpu_init(vcpu, kvm, id);
2291 vcpu->arch.shared = &vcpu->arch.shregs;
2292 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2294 * The shared struct is never shared on HV,
2295 * so we can always use host endianness
2297 #ifdef __BIG_ENDIAN__
2298 vcpu->arch.shared_big_endian = true;
2300 vcpu->arch.shared_big_endian = false;
2303 vcpu->arch.mmcr[0] = MMCR0_FC;
2304 vcpu->arch.ctrl = CTRL_RUNLATCH;
2305 /* default to host PVR, since we can't spoof it */
2306 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2307 spin_lock_init(&vcpu->arch.vpa_update_lock);
2308 spin_lock_init(&vcpu->arch.tbacct_lock);
2309 vcpu->arch.busy_preempt = TB_NIL;
2310 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2313 * Set the default HFSCR for the guest from the host value.
2314 * This value is only used on POWER9.
2315 * On POWER9, we want to virtualize the doorbell facility, so we
2316 * don't set the HFSCR_MSGP bit, and that causes those instructions
2317 * to trap and then we emulate them.
2319 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2320 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2321 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2322 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2323 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2324 vcpu->arch.hfscr |= HFSCR_TM;
2326 if (cpu_has_feature(CPU_FTR_TM_COMP))
2327 vcpu->arch.hfscr |= HFSCR_TM;
2329 kvmppc_mmu_book3s_hv_init(vcpu);
2331 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2333 init_waitqueue_head(&vcpu->arch.cpu_run);
2335 mutex_lock(&kvm->lock);
2338 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2339 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2340 pr_devel("KVM: VCPU ID too high\n");
2341 core = KVM_MAX_VCORES;
2343 BUG_ON(kvm->arch.smt_mode != 1);
2344 core = kvmppc_pack_vcpu_id(kvm, id);
2347 core = id / kvm->arch.smt_mode;
2349 if (core < KVM_MAX_VCORES) {
2350 vcore = kvm->arch.vcores[core];
2351 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2352 pr_devel("KVM: collision on id %u", id);
2354 } else if (!vcore) {
2356 * Take mmu_setup_lock for mutual exclusion
2357 * with kvmppc_update_lpcr().
2360 vcore = kvmppc_vcore_create(kvm,
2361 id & ~(kvm->arch.smt_mode - 1));
2362 mutex_lock(&kvm->arch.mmu_setup_lock);
2363 kvm->arch.vcores[core] = vcore;
2364 kvm->arch.online_vcores++;
2365 mutex_unlock(&kvm->arch.mmu_setup_lock);
2368 mutex_unlock(&kvm->lock);
2373 spin_lock(&vcore->lock);
2374 ++vcore->num_threads;
2375 spin_unlock(&vcore->lock);
2376 vcpu->arch.vcore = vcore;
2377 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2378 vcpu->arch.thread_cpu = -1;
2379 vcpu->arch.prev_cpu = -1;
2381 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2382 kvmppc_sanity_check(vcpu);
2384 debugfs_vcpu_init(vcpu, id);
2389 kmem_cache_free(kvm_vcpu_cache, vcpu);
2391 return ERR_PTR(err);
2394 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2395 unsigned long flags)
2402 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2404 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2406 * On POWER8 (or POWER7), the threading mode is "strict",
2407 * so we pack smt_mode vcpus per vcore.
2409 if (smt_mode > threads_per_subcore)
2413 * On POWER9, the threading mode is "loose",
2414 * so each vcpu gets its own vcore.
2419 mutex_lock(&kvm->lock);
2421 if (!kvm->arch.online_vcores) {
2422 kvm->arch.smt_mode = smt_mode;
2423 kvm->arch.emul_smt_mode = esmt;
2426 mutex_unlock(&kvm->lock);
2431 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2433 if (vpa->pinned_addr)
2434 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2438 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2440 spin_lock(&vcpu->arch.vpa_update_lock);
2441 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2442 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2443 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2444 spin_unlock(&vcpu->arch.vpa_update_lock);
2445 kvm_vcpu_uninit(vcpu);
2446 kmem_cache_free(kvm_vcpu_cache, vcpu);
2449 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2451 /* Indicate we want to get back into the guest */
2455 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2457 unsigned long dec_nsec, now;
2460 if (now > vcpu->arch.dec_expires) {
2461 /* decrementer has already gone negative */
2462 kvmppc_core_queue_dec(vcpu);
2463 kvmppc_core_prepare_to_enter(vcpu);
2466 dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2467 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2468 vcpu->arch.timer_running = 1;
2471 extern int __kvmppc_vcore_entry(void);
2473 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2474 struct kvm_vcpu *vcpu)
2478 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2480 spin_lock_irq(&vcpu->arch.tbacct_lock);
2482 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2483 vcpu->arch.stolen_logged;
2484 vcpu->arch.busy_preempt = now;
2485 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2486 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2488 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2491 static int kvmppc_grab_hwthread(int cpu)
2493 struct paca_struct *tpaca;
2494 long timeout = 10000;
2496 tpaca = paca_ptrs[cpu];
2498 /* Ensure the thread won't go into the kernel if it wakes */
2499 tpaca->kvm_hstate.kvm_vcpu = NULL;
2500 tpaca->kvm_hstate.kvm_vcore = NULL;
2501 tpaca->kvm_hstate.napping = 0;
2503 tpaca->kvm_hstate.hwthread_req = 1;
2506 * If the thread is already executing in the kernel (e.g. handling
2507 * a stray interrupt), wait for it to get back to nap mode.
2508 * The smp_mb() is to ensure that our setting of hwthread_req
2509 * is visible before we look at hwthread_state, so if this
2510 * races with the code at system_reset_pSeries and the thread
2511 * misses our setting of hwthread_req, we are sure to see its
2512 * setting of hwthread_state, and vice versa.
2515 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2516 if (--timeout <= 0) {
2517 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2525 static void kvmppc_release_hwthread(int cpu)
2527 struct paca_struct *tpaca;
2529 tpaca = paca_ptrs[cpu];
2530 tpaca->kvm_hstate.hwthread_req = 0;
2531 tpaca->kvm_hstate.kvm_vcpu = NULL;
2532 tpaca->kvm_hstate.kvm_vcore = NULL;
2533 tpaca->kvm_hstate.kvm_split_mode = NULL;
2536 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2538 struct kvm_nested_guest *nested = vcpu->arch.nested;
2539 cpumask_t *cpu_in_guest;
2542 cpu = cpu_first_thread_sibling(cpu);
2544 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2545 cpu_in_guest = &nested->cpu_in_guest;
2547 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2548 cpu_in_guest = &kvm->arch.cpu_in_guest;
2551 * Make sure setting of bit in need_tlb_flush precedes
2552 * testing of cpu_in_guest bits. The matching barrier on
2553 * the other side is the first smp_mb() in kvmppc_run_core().
2556 for (i = 0; i < threads_per_core; ++i)
2557 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2558 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2561 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2563 struct kvm_nested_guest *nested = vcpu->arch.nested;
2564 struct kvm *kvm = vcpu->kvm;
2567 if (!cpu_has_feature(CPU_FTR_HVMODE))
2571 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2573 prev_cpu = vcpu->arch.prev_cpu;
2576 * With radix, the guest can do TLB invalidations itself,
2577 * and it could choose to use the local form (tlbiel) if
2578 * it is invalidating a translation that has only ever been
2579 * used on one vcpu. However, that doesn't mean it has
2580 * only ever been used on one physical cpu, since vcpus
2581 * can move around between pcpus. To cope with this, when
2582 * a vcpu moves from one pcpu to another, we need to tell
2583 * any vcpus running on the same core as this vcpu previously
2584 * ran to flush the TLB. The TLB is shared between threads,
2585 * so we use a single bit in .need_tlb_flush for all 4 threads.
2587 if (prev_cpu != pcpu) {
2588 if (prev_cpu >= 0 &&
2589 cpu_first_thread_sibling(prev_cpu) !=
2590 cpu_first_thread_sibling(pcpu))
2591 radix_flush_cpu(kvm, prev_cpu, vcpu);
2593 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2595 vcpu->arch.prev_cpu = pcpu;
2599 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2602 struct paca_struct *tpaca;
2603 struct kvm *kvm = vc->kvm;
2607 if (vcpu->arch.timer_running) {
2608 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2609 vcpu->arch.timer_running = 0;
2611 cpu += vcpu->arch.ptid;
2612 vcpu->cpu = vc->pcpu;
2613 vcpu->arch.thread_cpu = cpu;
2614 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2616 tpaca = paca_ptrs[cpu];
2617 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2618 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2619 tpaca->kvm_hstate.fake_suspend = 0;
2620 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2622 tpaca->kvm_hstate.kvm_vcore = vc;
2623 if (cpu != smp_processor_id())
2624 kvmppc_ipi_thread(cpu);
2627 static void kvmppc_wait_for_nap(int n_threads)
2629 int cpu = smp_processor_id();
2634 for (loops = 0; loops < 1000000; ++loops) {
2636 * Check if all threads are finished.
2637 * We set the vcore pointer when starting a thread
2638 * and the thread clears it when finished, so we look
2639 * for any threads that still have a non-NULL vcore ptr.
2641 for (i = 1; i < n_threads; ++i)
2642 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2644 if (i == n_threads) {
2651 for (i = 1; i < n_threads; ++i)
2652 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2653 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2657 * Check that we are on thread 0 and that any other threads in
2658 * this core are off-line. Then grab the threads so they can't
2661 static int on_primary_thread(void)
2663 int cpu = smp_processor_id();
2666 /* Are we on a primary subcore? */
2667 if (cpu_thread_in_subcore(cpu))
2671 while (++thr < threads_per_subcore)
2672 if (cpu_online(cpu + thr))
2675 /* Grab all hw threads so they can't go into the kernel */
2676 for (thr = 1; thr < threads_per_subcore; ++thr) {
2677 if (kvmppc_grab_hwthread(cpu + thr)) {
2678 /* Couldn't grab one; let the others go */
2680 kvmppc_release_hwthread(cpu + thr);
2681 } while (--thr > 0);
2689 * A list of virtual cores for each physical CPU.
2690 * These are vcores that could run but their runner VCPU tasks are
2691 * (or may be) preempted.
2693 struct preempted_vcore_list {
2694 struct list_head list;
2698 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2700 static void init_vcore_lists(void)
2704 for_each_possible_cpu(cpu) {
2705 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2706 spin_lock_init(&lp->lock);
2707 INIT_LIST_HEAD(&lp->list);
2711 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2713 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2715 vc->vcore_state = VCORE_PREEMPT;
2716 vc->pcpu = smp_processor_id();
2717 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2718 spin_lock(&lp->lock);
2719 list_add_tail(&vc->preempt_list, &lp->list);
2720 spin_unlock(&lp->lock);
2723 /* Start accumulating stolen time */
2724 kvmppc_core_start_stolen(vc);
2727 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2729 struct preempted_vcore_list *lp;
2731 kvmppc_core_end_stolen(vc);
2732 if (!list_empty(&vc->preempt_list)) {
2733 lp = &per_cpu(preempted_vcores, vc->pcpu);
2734 spin_lock(&lp->lock);
2735 list_del_init(&vc->preempt_list);
2736 spin_unlock(&lp->lock);
2738 vc->vcore_state = VCORE_INACTIVE;
2742 * This stores information about the virtual cores currently
2743 * assigned to a physical core.
2747 int max_subcore_threads;
2749 int subcore_threads[MAX_SUBCORES];
2750 struct kvmppc_vcore *vc[MAX_SUBCORES];
2754 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2755 * respectively in 2-way micro-threading (split-core) mode on POWER8.
2757 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2759 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2761 memset(cip, 0, sizeof(*cip));
2762 cip->n_subcores = 1;
2763 cip->max_subcore_threads = vc->num_threads;
2764 cip->total_threads = vc->num_threads;
2765 cip->subcore_threads[0] = vc->num_threads;
2769 static bool subcore_config_ok(int n_subcores, int n_threads)
2772 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2773 * split-core mode, with one thread per subcore.
2775 if (cpu_has_feature(CPU_FTR_ARCH_300))
2776 return n_subcores <= 4 && n_threads == 1;
2778 /* On POWER8, can only dynamically split if unsplit to begin with */
2779 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2781 if (n_subcores > MAX_SUBCORES)
2783 if (n_subcores > 1) {
2784 if (!(dynamic_mt_modes & 2))
2786 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2790 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2793 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2795 vc->entry_exit_map = 0;
2797 vc->napping_threads = 0;
2798 vc->conferring_threads = 0;
2799 vc->tb_offset_applied = 0;
2802 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2804 int n_threads = vc->num_threads;
2807 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2810 /* In one_vm_per_core mode, require all vcores to be from the same vm */
2811 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2814 /* Some POWER9 chips require all threads to be in the same MMU mode */
2815 if (no_mixing_hpt_and_radix &&
2816 kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2819 if (n_threads < cip->max_subcore_threads)
2820 n_threads = cip->max_subcore_threads;
2821 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2823 cip->max_subcore_threads = n_threads;
2825 sub = cip->n_subcores;
2827 cip->total_threads += vc->num_threads;
2828 cip->subcore_threads[sub] = vc->num_threads;
2830 init_vcore_to_run(vc);
2831 list_del_init(&vc->preempt_list);
2837 * Work out whether it is possible to piggyback the execution of
2838 * vcore *pvc onto the execution of the other vcores described in *cip.
2840 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2843 if (cip->total_threads + pvc->num_threads > target_threads)
2846 return can_dynamic_split(pvc, cip);
2849 static void prepare_threads(struct kvmppc_vcore *vc)
2852 struct kvm_vcpu *vcpu;
2854 for_each_runnable_thread(i, vcpu, vc) {
2855 if (signal_pending(vcpu->arch.run_task))
2856 vcpu->arch.ret = -EINTR;
2857 else if (vcpu->arch.vpa.update_pending ||
2858 vcpu->arch.slb_shadow.update_pending ||
2859 vcpu->arch.dtl.update_pending)
2860 vcpu->arch.ret = RESUME_GUEST;
2863 kvmppc_remove_runnable(vc, vcpu);
2864 wake_up(&vcpu->arch.cpu_run);
2868 static void collect_piggybacks(struct core_info *cip, int target_threads)
2870 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2871 struct kvmppc_vcore *pvc, *vcnext;
2873 spin_lock(&lp->lock);
2874 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2875 if (!spin_trylock(&pvc->lock))
2877 prepare_threads(pvc);
2878 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2879 list_del_init(&pvc->preempt_list);
2880 if (pvc->runner == NULL) {
2881 pvc->vcore_state = VCORE_INACTIVE;
2882 kvmppc_core_end_stolen(pvc);
2884 spin_unlock(&pvc->lock);
2887 if (!can_piggyback(pvc, cip, target_threads)) {
2888 spin_unlock(&pvc->lock);
2891 kvmppc_core_end_stolen(pvc);
2892 pvc->vcore_state = VCORE_PIGGYBACK;
2893 if (cip->total_threads >= target_threads)
2896 spin_unlock(&lp->lock);
2899 static bool recheck_signals_and_mmu(struct core_info *cip)
2902 struct kvm_vcpu *vcpu;
2903 struct kvmppc_vcore *vc;
2905 for (sub = 0; sub < cip->n_subcores; ++sub) {
2907 if (!vc->kvm->arch.mmu_ready)
2909 for_each_runnable_thread(i, vcpu, vc)
2910 if (signal_pending(vcpu->arch.run_task))
2916 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2918 int still_running = 0, i;
2921 struct kvm_vcpu *vcpu;
2923 spin_lock(&vc->lock);
2925 for_each_runnable_thread(i, vcpu, vc) {
2927 * It's safe to unlock the vcore in the loop here, because
2928 * for_each_runnable_thread() is safe against removal of
2929 * the vcpu, and the vcore state is VCORE_EXITING here,
2930 * so any vcpus becoming runnable will have their arch.trap
2931 * set to zero and can't actually run in the guest.
2933 spin_unlock(&vc->lock);
2934 /* cancel pending dec exception if dec is positive */
2935 if (now < vcpu->arch.dec_expires &&
2936 kvmppc_core_pending_dec(vcpu))
2937 kvmppc_core_dequeue_dec(vcpu);
2939 trace_kvm_guest_exit(vcpu);
2942 if (vcpu->arch.trap)
2943 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2944 vcpu->arch.run_task);
2946 vcpu->arch.ret = ret;
2947 vcpu->arch.trap = 0;
2949 spin_lock(&vc->lock);
2950 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2951 if (vcpu->arch.pending_exceptions)
2952 kvmppc_core_prepare_to_enter(vcpu);
2953 if (vcpu->arch.ceded)
2954 kvmppc_set_timer(vcpu);
2958 kvmppc_remove_runnable(vc, vcpu);
2959 wake_up(&vcpu->arch.cpu_run);
2963 if (still_running > 0) {
2964 kvmppc_vcore_preempt(vc);
2965 } else if (vc->runner) {
2966 vc->vcore_state = VCORE_PREEMPT;
2967 kvmppc_core_start_stolen(vc);
2969 vc->vcore_state = VCORE_INACTIVE;
2971 if (vc->n_runnable > 0 && vc->runner == NULL) {
2972 /* make sure there's a candidate runner awake */
2974 vcpu = next_runnable_thread(vc, &i);
2975 wake_up(&vcpu->arch.cpu_run);
2978 spin_unlock(&vc->lock);
2982 * Clear core from the list of active host cores as we are about to
2983 * enter the guest. Only do this if it is the primary thread of the
2984 * core (not if a subcore) that is entering the guest.
2986 static inline int kvmppc_clear_host_core(unsigned int cpu)
2990 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2993 * Memory barrier can be omitted here as we will do a smp_wmb()
2994 * later in kvmppc_start_thread and we need ensure that state is
2995 * visible to other CPUs only after we enter guest.
2997 core = cpu >> threads_shift;
2998 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3003 * Advertise this core as an active host core since we exited the guest
3004 * Only need to do this if it is the primary thread of the core that is
3007 static inline int kvmppc_set_host_core(unsigned int cpu)
3011 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3015 * Memory barrier can be omitted here because we do a spin_unlock
3016 * immediately after this which provides the memory barrier.
3018 core = cpu >> threads_shift;
3019 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3023 static void set_irq_happened(int trap)
3026 case BOOK3S_INTERRUPT_EXTERNAL:
3027 local_paca->irq_happened |= PACA_IRQ_EE;
3029 case BOOK3S_INTERRUPT_H_DOORBELL:
3030 local_paca->irq_happened |= PACA_IRQ_DBELL;
3032 case BOOK3S_INTERRUPT_HMI:
3033 local_paca->irq_happened |= PACA_IRQ_HMI;
3035 case BOOK3S_INTERRUPT_SYSTEM_RESET:
3036 replay_system_reset();
3042 * Run a set of guest threads on a physical core.
3043 * Called with vc->lock held.
3045 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3047 struct kvm_vcpu *vcpu;
3050 struct core_info core_info;
3051 struct kvmppc_vcore *pvc;
3052 struct kvm_split_mode split_info, *sip;
3053 int split, subcore_size, active;
3056 unsigned long cmd_bit, stat_bit;
3059 int controlled_threads;
3065 * Remove from the list any threads that have a signal pending
3066 * or need a VPA update done
3068 prepare_threads(vc);
3070 /* if the runner is no longer runnable, let the caller pick a new one */
3071 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3077 init_vcore_to_run(vc);
3078 vc->preempt_tb = TB_NIL;
3081 * Number of threads that we will be controlling: the same as
3082 * the number of threads per subcore, except on POWER9,
3083 * where it's 1 because the threads are (mostly) independent.
3085 controlled_threads = threads_per_vcore(vc->kvm);
3088 * Make sure we are running on primary threads, and that secondary
3089 * threads are offline. Also check if the number of threads in this
3090 * guest are greater than the current system threads per guest.
3091 * On POWER9, we need to be not in independent-threads mode if
3092 * this is a HPT guest on a radix host machine where the
3093 * CPU threads may not be in different MMU modes.
3095 hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3096 !kvm_is_radix(vc->kvm);
3097 if (((controlled_threads > 1) &&
3098 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3099 (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3100 for_each_runnable_thread(i, vcpu, vc) {
3101 vcpu->arch.ret = -EBUSY;
3102 kvmppc_remove_runnable(vc, vcpu);
3103 wake_up(&vcpu->arch.cpu_run);
3109 * See if we could run any other vcores on the physical core
3110 * along with this one.
3112 init_core_info(&core_info, vc);
3113 pcpu = smp_processor_id();
3114 target_threads = controlled_threads;
3115 if (target_smt_mode && target_smt_mode < target_threads)
3116 target_threads = target_smt_mode;
3117 if (vc->num_threads < target_threads)
3118 collect_piggybacks(&core_info, target_threads);
3121 * On radix, arrange for TLB flushing if necessary.
3122 * This has to be done before disabling interrupts since
3123 * it uses smp_call_function().
3125 pcpu = smp_processor_id();
3126 if (kvm_is_radix(vc->kvm)) {
3127 for (sub = 0; sub < core_info.n_subcores; ++sub)
3128 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3129 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3133 * Hard-disable interrupts, and check resched flag and signals.
3134 * If we need to reschedule or deliver a signal, clean up
3135 * and return without going into the guest(s).
3136 * If the mmu_ready flag has been cleared, don't go into the
3137 * guest because that means a HPT resize operation is in progress.
3139 local_irq_disable();
3141 if (lazy_irq_pending() || need_resched() ||
3142 recheck_signals_and_mmu(&core_info)) {
3144 vc->vcore_state = VCORE_INACTIVE;
3145 /* Unlock all except the primary vcore */
3146 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3147 pvc = core_info.vc[sub];
3148 /* Put back on to the preempted vcores list */
3149 kvmppc_vcore_preempt(pvc);
3150 spin_unlock(&pvc->lock);
3152 for (i = 0; i < controlled_threads; ++i)
3153 kvmppc_release_hwthread(pcpu + i);
3157 kvmppc_clear_host_core(pcpu);
3159 /* Decide on micro-threading (split-core) mode */
3160 subcore_size = threads_per_subcore;
3161 cmd_bit = stat_bit = 0;
3162 split = core_info.n_subcores;
3164 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3165 && !cpu_has_feature(CPU_FTR_ARCH_300);
3167 if (split > 1 || hpt_on_radix) {
3169 memset(&split_info, 0, sizeof(split_info));
3170 for (sub = 0; sub < core_info.n_subcores; ++sub)
3171 split_info.vc[sub] = core_info.vc[sub];
3174 if (split == 2 && (dynamic_mt_modes & 2)) {
3175 cmd_bit = HID0_POWER8_1TO2LPAR;
3176 stat_bit = HID0_POWER8_2LPARMODE;
3179 cmd_bit = HID0_POWER8_1TO4LPAR;
3180 stat_bit = HID0_POWER8_4LPARMODE;
3182 subcore_size = MAX_SMT_THREADS / split;
3183 split_info.rpr = mfspr(SPRN_RPR);
3184 split_info.pmmar = mfspr(SPRN_PMMAR);
3185 split_info.ldbar = mfspr(SPRN_LDBAR);
3186 split_info.subcore_size = subcore_size;
3188 split_info.subcore_size = 1;
3190 /* Use the split_info for LPCR/LPIDR changes */
3191 split_info.lpcr_req = vc->lpcr;
3192 split_info.lpidr_req = vc->kvm->arch.lpid;
3193 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3194 split_info.do_set = 1;
3198 /* order writes to split_info before kvm_split_mode pointer */
3202 for (thr = 0; thr < controlled_threads; ++thr) {
3203 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3205 paca->kvm_hstate.tid = thr;
3206 paca->kvm_hstate.napping = 0;
3207 paca->kvm_hstate.kvm_split_mode = sip;
3210 /* Initiate micro-threading (split-core) on POWER8 if required */
3212 unsigned long hid0 = mfspr(SPRN_HID0);
3214 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3216 mtspr(SPRN_HID0, hid0);
3219 hid0 = mfspr(SPRN_HID0);
3220 if (hid0 & stat_bit)
3227 * On POWER8, set RWMR register.
3228 * Since it only affects PURR and SPURR, it doesn't affect
3229 * the host, so we don't save/restore the host value.
3232 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3233 int n_online = atomic_read(&vc->online_count);
3236 * Use the 8-thread value if we're doing split-core
3237 * or if the vcore's online count looks bogus.
3239 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3240 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3241 rwmr_val = p8_rwmr_values[n_online];
3242 mtspr(SPRN_RWMR, rwmr_val);
3245 /* Start all the threads */
3247 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3248 thr = is_power8 ? subcore_thread_map[sub] : sub;
3251 pvc = core_info.vc[sub];
3252 pvc->pcpu = pcpu + thr;
3253 for_each_runnable_thread(i, vcpu, pvc) {
3254 kvmppc_start_thread(vcpu, pvc);
3255 kvmppc_create_dtl_entry(vcpu, pvc);
3256 trace_kvm_guest_enter(vcpu);
3257 if (!vcpu->arch.ptid)
3259 active |= 1 << (thr + vcpu->arch.ptid);
3262 * We need to start the first thread of each subcore
3263 * even if it doesn't have a vcpu.
3266 kvmppc_start_thread(NULL, pvc);
3270 * Ensure that split_info.do_nap is set after setting
3271 * the vcore pointer in the PACA of the secondaries.
3276 * When doing micro-threading, poke the inactive threads as well.
3277 * This gets them to the nap instruction after kvm_do_nap,
3278 * which reduces the time taken to unsplit later.
3279 * For POWER9 HPT guest on radix host, we need all the secondary
3280 * threads woken up so they can do the LPCR/LPIDR change.
3282 if (cmd_bit || hpt_on_radix) {
3283 split_info.do_nap = 1; /* ask secondaries to nap when done */
3284 for (thr = 1; thr < threads_per_subcore; ++thr)
3285 if (!(active & (1 << thr)))
3286 kvmppc_ipi_thread(pcpu + thr);
3289 vc->vcore_state = VCORE_RUNNING;
3292 trace_kvmppc_run_core(vc, 0);
3294 for (sub = 0; sub < core_info.n_subcores; ++sub)
3295 spin_unlock(&core_info.vc[sub]->lock);
3297 guest_enter_irqoff();
3299 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3301 this_cpu_disable_ftrace();
3304 * Interrupts will be enabled once we get into the guest,
3305 * so tell lockdep that we're about to enable interrupts.
3307 trace_hardirqs_on();
3309 trap = __kvmppc_vcore_entry();
3311 trace_hardirqs_off();
3313 this_cpu_enable_ftrace();
3315 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3317 set_irq_happened(trap);
3319 spin_lock(&vc->lock);
3320 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3321 vc->vcore_state = VCORE_EXITING;
3323 /* wait for secondary threads to finish writing their state to memory */
3324 kvmppc_wait_for_nap(controlled_threads);
3326 /* Return to whole-core mode if we split the core earlier */
3328 unsigned long hid0 = mfspr(SPRN_HID0);
3329 unsigned long loops = 0;
3331 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3332 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3334 mtspr(SPRN_HID0, hid0);
3337 hid0 = mfspr(SPRN_HID0);
3338 if (!(hid0 & stat_bit))
3343 } else if (hpt_on_radix) {
3344 /* Wait for all threads to have seen final sync */
3345 for (thr = 1; thr < controlled_threads; ++thr) {
3346 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3348 while (paca->kvm_hstate.kvm_split_mode) {
3355 split_info.do_nap = 0;
3357 kvmppc_set_host_core(pcpu);
3362 /* Let secondaries go back to the offline loop */
3363 for (i = 0; i < controlled_threads; ++i) {
3364 kvmppc_release_hwthread(pcpu + i);
3365 if (sip && sip->napped[i])
3366 kvmppc_ipi_thread(pcpu + i);
3367 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3370 spin_unlock(&vc->lock);
3372 /* make sure updates to secondary vcpu structs are visible now */
3377 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3378 pvc = core_info.vc[sub];
3379 post_guest_process(pvc, pvc == vc);
3382 spin_lock(&vc->lock);
3385 vc->vcore_state = VCORE_INACTIVE;
3386 trace_kvmppc_run_core(vc, 1);
3390 * Load up hypervisor-mode registers on P9.
3392 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3395 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3397 u64 tb, purr, spurr;
3399 unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3400 unsigned long host_ciabr = mfspr(SPRN_CIABR);
3401 unsigned long host_dawr = mfspr(SPRN_DAWR);
3402 unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3403 unsigned long host_psscr = mfspr(SPRN_PSSCR);
3404 unsigned long host_pidr = mfspr(SPRN_PID);
3406 hdec = time_limit - mftb();
3408 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3409 mtspr(SPRN_HDEC, hdec);
3411 if (vc->tb_offset) {
3412 u64 new_tb = mftb() + vc->tb_offset;
3413 mtspr(SPRN_TBU40, new_tb);
3415 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3416 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3417 vc->tb_offset_applied = vc->tb_offset;
3421 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3422 mtspr(SPRN_DPDES, vc->dpdes);
3423 mtspr(SPRN_VTB, vc->vtb);
3425 local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3426 local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3427 mtspr(SPRN_PURR, vcpu->arch.purr);
3428 mtspr(SPRN_SPURR, vcpu->arch.spurr);
3430 if (dawr_enabled()) {
3431 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3432 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3434 mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3435 mtspr(SPRN_IC, vcpu->arch.ic);
3436 mtspr(SPRN_PID, vcpu->arch.pid);
3438 mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3439 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3441 mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3443 mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3444 mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3445 mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3446 mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3448 mtspr(SPRN_AMOR, ~0UL);
3450 mtspr(SPRN_LPCR, lpcr);
3453 kvmppc_xive_push_vcpu(vcpu);
3455 mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3456 mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3458 trap = __kvmhv_vcpu_entry_p9(vcpu);
3460 /* Advance host PURR/SPURR by the amount used by guest */
3461 purr = mfspr(SPRN_PURR);
3462 spurr = mfspr(SPRN_SPURR);
3463 mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3464 purr - vcpu->arch.purr);
3465 mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3466 spurr - vcpu->arch.spurr);
3467 vcpu->arch.purr = purr;
3468 vcpu->arch.spurr = spurr;
3470 vcpu->arch.ic = mfspr(SPRN_IC);
3471 vcpu->arch.pid = mfspr(SPRN_PID);
3472 vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3474 vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3475 vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3476 vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3477 vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3479 /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3480 mtspr(SPRN_PSSCR, host_psscr |
3481 (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3482 mtspr(SPRN_HFSCR, host_hfscr);
3483 mtspr(SPRN_CIABR, host_ciabr);
3484 mtspr(SPRN_DAWR, host_dawr);
3485 mtspr(SPRN_DAWRX, host_dawrx);
3486 mtspr(SPRN_PID, host_pidr);
3489 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3490 * case we interrupted the guest between a tlbie and a ptesync.
3492 asm volatile("eieio; tlbsync; ptesync");
3494 mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid); /* restore host LPID */
3497 vc->dpdes = mfspr(SPRN_DPDES);
3498 vc->vtb = mfspr(SPRN_VTB);
3499 mtspr(SPRN_DPDES, 0);
3501 mtspr(SPRN_PCR, PCR_MASK);
3503 if (vc->tb_offset_applied) {
3504 u64 new_tb = mftb() - vc->tb_offset_applied;
3505 mtspr(SPRN_TBU40, new_tb);
3507 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3508 mtspr(SPRN_TBU40, new_tb + 0x1000000);
3509 vc->tb_offset_applied = 0;
3512 mtspr(SPRN_HDEC, 0x7fffffff);
3513 mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3519 * Virtual-mode guest entry for POWER9 and later when the host and
3520 * guest are both using the radix MMU. The LPIDR has already been set.
3522 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3525 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3526 unsigned long host_dscr = mfspr(SPRN_DSCR);
3527 unsigned long host_tidr = mfspr(SPRN_TIDR);
3528 unsigned long host_iamr = mfspr(SPRN_IAMR);
3529 unsigned long host_amr = mfspr(SPRN_AMR);
3534 dec = mfspr(SPRN_DEC);
3537 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3538 local_paca->kvm_hstate.dec_expires = dec + tb;
3539 if (local_paca->kvm_hstate.dec_expires < time_limit)
3540 time_limit = local_paca->kvm_hstate.dec_expires;
3542 vcpu->arch.ceded = 0;
3544 kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */
3546 kvmppc_subcore_enter_guest();
3548 vc->entry_exit_map = 1;
3551 if (vcpu->arch.vpa.pinned_addr) {
3552 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3553 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3554 lp->yield_count = cpu_to_be32(yield_count);
3555 vcpu->arch.vpa.dirty = 1;
3558 if (cpu_has_feature(CPU_FTR_TM) ||
3559 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3560 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3562 kvmhv_load_guest_pmu(vcpu);
3564 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3565 load_fp_state(&vcpu->arch.fp);
3566 #ifdef CONFIG_ALTIVEC
3567 load_vr_state(&vcpu->arch.vr);
3569 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3571 mtspr(SPRN_DSCR, vcpu->arch.dscr);
3572 mtspr(SPRN_IAMR, vcpu->arch.iamr);
3573 mtspr(SPRN_PSPB, vcpu->arch.pspb);
3574 mtspr(SPRN_FSCR, vcpu->arch.fscr);
3575 mtspr(SPRN_TAR, vcpu->arch.tar);
3576 mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3577 mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3578 mtspr(SPRN_BESCR, vcpu->arch.bescr);
3579 mtspr(SPRN_WORT, vcpu->arch.wort);
3580 mtspr(SPRN_TIDR, vcpu->arch.tid);
3581 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3582 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3583 mtspr(SPRN_AMR, vcpu->arch.amr);
3584 mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3586 if (!(vcpu->arch.ctrl & 1))
3587 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3589 mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3591 if (kvmhv_on_pseries()) {
3593 * We need to save and restore the guest visible part of the
3594 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3595 * doesn't do this for us. Note only required if pseries since
3596 * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3598 unsigned long host_psscr;
3599 /* call our hypervisor to load up HV regs and go */
3600 struct hv_guest_state hvregs;
3602 host_psscr = mfspr(SPRN_PSSCR_PR);
3603 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3604 kvmhv_save_hv_regs(vcpu, &hvregs);
3606 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3607 hvregs.version = HV_GUEST_STATE_VERSION;
3608 if (vcpu->arch.nested) {
3609 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3610 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3612 hvregs.lpid = vcpu->kvm->arch.lpid;
3613 hvregs.vcpu_token = vcpu->vcpu_id;
3615 hvregs.hdec_expiry = time_limit;
3616 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3617 __pa(&vcpu->arch.regs));
3618 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3619 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3620 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3621 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3622 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3623 mtspr(SPRN_PSSCR_PR, host_psscr);
3625 /* H_CEDE has to be handled now, not later */
3626 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3627 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3628 kvmppc_nested_cede(vcpu);
3632 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3635 vcpu->arch.slb_max = 0;
3636 dec = mfspr(SPRN_DEC);
3637 if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3640 vcpu->arch.dec_expires = dec + tb;
3642 vcpu->arch.thread_cpu = -1;
3643 vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3645 vcpu->arch.iamr = mfspr(SPRN_IAMR);
3646 vcpu->arch.pspb = mfspr(SPRN_PSPB);
3647 vcpu->arch.fscr = mfspr(SPRN_FSCR);
3648 vcpu->arch.tar = mfspr(SPRN_TAR);
3649 vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3650 vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3651 vcpu->arch.bescr = mfspr(SPRN_BESCR);
3652 vcpu->arch.wort = mfspr(SPRN_WORT);
3653 vcpu->arch.tid = mfspr(SPRN_TIDR);
3654 vcpu->arch.amr = mfspr(SPRN_AMR);
3655 vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3656 vcpu->arch.dscr = mfspr(SPRN_DSCR);
3658 mtspr(SPRN_PSPB, 0);
3659 mtspr(SPRN_WORT, 0);
3660 mtspr(SPRN_UAMOR, 0);
3661 mtspr(SPRN_DSCR, host_dscr);
3662 mtspr(SPRN_TIDR, host_tidr);
3663 mtspr(SPRN_IAMR, host_iamr);
3664 mtspr(SPRN_PSPB, 0);
3666 if (host_amr != vcpu->arch.amr)
3667 mtspr(SPRN_AMR, host_amr);
3669 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3670 store_fp_state(&vcpu->arch.fp);
3671 #ifdef CONFIG_ALTIVEC
3672 store_vr_state(&vcpu->arch.vr);
3674 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3676 if (cpu_has_feature(CPU_FTR_TM) ||
3677 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3678 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3681 if (vcpu->arch.vpa.pinned_addr) {
3682 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3683 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3684 lp->yield_count = cpu_to_be32(yield_count);
3685 vcpu->arch.vpa.dirty = 1;
3686 save_pmu = lp->pmcregs_in_use;
3688 /* Must save pmu if this guest is capable of running nested guests */
3689 save_pmu |= nesting_enabled(vcpu->kvm);
3691 kvmhv_save_guest_pmu(vcpu, save_pmu);
3693 vc->entry_exit_map = 0x101;
3696 mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3697 mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3699 kvmhv_load_host_pmu();
3701 kvmppc_subcore_exit_guest();
3707 * Wait for some other vcpu thread to execute us, and
3708 * wake us up when we need to handle something in the host.
3710 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3711 struct kvm_vcpu *vcpu, int wait_state)
3715 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3716 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3717 spin_unlock(&vc->lock);
3719 spin_lock(&vc->lock);
3721 finish_wait(&vcpu->arch.cpu_run, &wait);
3724 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3726 if (!halt_poll_ns_grow)
3729 vc->halt_poll_ns *= halt_poll_ns_grow;
3730 if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3731 vc->halt_poll_ns = halt_poll_ns_grow_start;
3734 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3736 if (halt_poll_ns_shrink == 0)
3737 vc->halt_poll_ns = 0;
3739 vc->halt_poll_ns /= halt_poll_ns_shrink;
3742 #ifdef CONFIG_KVM_XICS
3743 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3745 if (!xics_on_xive())
3747 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3748 vcpu->arch.xive_saved_state.cppr;
3751 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3755 #endif /* CONFIG_KVM_XICS */
3757 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3759 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3760 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3767 * Check to see if any of the runnable vcpus on the vcore have pending
3768 * exceptions or are no longer ceded
3770 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3772 struct kvm_vcpu *vcpu;
3775 for_each_runnable_thread(i, vcpu, vc) {
3776 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3784 * All the vcpus in this vcore are idle, so wait for a decrementer
3785 * or external interrupt to one of the vcpus. vc->lock is held.
3787 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3789 ktime_t cur, start_poll, start_wait;
3792 DECLARE_SWAITQUEUE(wait);
3794 /* Poll for pending exceptions and ceded state */
3795 cur = start_poll = ktime_get();
3796 if (vc->halt_poll_ns) {
3797 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3798 ++vc->runner->stat.halt_attempted_poll;
3800 vc->vcore_state = VCORE_POLLING;
3801 spin_unlock(&vc->lock);
3804 if (kvmppc_vcore_check_block(vc)) {
3809 } while (single_task_running() && ktime_before(cur, stop));
3811 spin_lock(&vc->lock);
3812 vc->vcore_state = VCORE_INACTIVE;
3815 ++vc->runner->stat.halt_successful_poll;
3820 prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3822 if (kvmppc_vcore_check_block(vc)) {
3823 finish_swait(&vc->wq, &wait);
3825 /* If we polled, count this as a successful poll */
3826 if (vc->halt_poll_ns)
3827 ++vc->runner->stat.halt_successful_poll;
3831 start_wait = ktime_get();
3833 vc->vcore_state = VCORE_SLEEPING;
3834 trace_kvmppc_vcore_blocked(vc, 0);
3835 spin_unlock(&vc->lock);
3837 finish_swait(&vc->wq, &wait);
3838 spin_lock(&vc->lock);
3839 vc->vcore_state = VCORE_INACTIVE;
3840 trace_kvmppc_vcore_blocked(vc, 1);
3841 ++vc->runner->stat.halt_successful_wait;
3846 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3848 /* Attribute wait time */
3850 vc->runner->stat.halt_wait_ns +=
3851 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3852 /* Attribute failed poll time */
3853 if (vc->halt_poll_ns)
3854 vc->runner->stat.halt_poll_fail_ns +=
3855 ktime_to_ns(start_wait) -
3856 ktime_to_ns(start_poll);
3858 /* Attribute successful poll time */
3859 if (vc->halt_poll_ns)
3860 vc->runner->stat.halt_poll_success_ns +=
3862 ktime_to_ns(start_poll);
3865 /* Adjust poll time */
3867 if (block_ns <= vc->halt_poll_ns)
3869 /* We slept and blocked for longer than the max halt time */
3870 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3871 shrink_halt_poll_ns(vc);
3872 /* We slept and our poll time is too small */
3873 else if (vc->halt_poll_ns < halt_poll_ns &&
3874 block_ns < halt_poll_ns)
3875 grow_halt_poll_ns(vc);
3876 if (vc->halt_poll_ns > halt_poll_ns)
3877 vc->halt_poll_ns = halt_poll_ns;
3879 vc->halt_poll_ns = 0;
3881 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3885 * This never fails for a radix guest, as none of the operations it does
3886 * for a radix guest can fail or have a way to report failure.
3887 * kvmhv_run_single_vcpu() relies on this fact.
3889 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3892 struct kvm *kvm = vcpu->kvm;
3894 mutex_lock(&kvm->arch.mmu_setup_lock);
3895 if (!kvm->arch.mmu_ready) {
3896 if (!kvm_is_radix(kvm))
3897 r = kvmppc_hv_setup_htab_rma(vcpu);
3899 if (cpu_has_feature(CPU_FTR_ARCH_300))
3900 kvmppc_setup_partition_table(kvm);
3901 kvm->arch.mmu_ready = 1;
3904 mutex_unlock(&kvm->arch.mmu_setup_lock);
3908 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3911 struct kvmppc_vcore *vc;
3914 trace_kvmppc_run_vcpu_enter(vcpu);
3916 kvm_run->exit_reason = 0;
3917 vcpu->arch.ret = RESUME_GUEST;
3918 vcpu->arch.trap = 0;
3919 kvmppc_update_vpas(vcpu);
3922 * Synchronize with other threads in this virtual core
3924 vc = vcpu->arch.vcore;
3925 spin_lock(&vc->lock);
3926 vcpu->arch.ceded = 0;
3927 vcpu->arch.run_task = current;
3928 vcpu->arch.kvm_run = kvm_run;
3929 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3930 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3931 vcpu->arch.busy_preempt = TB_NIL;
3932 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3936 * This happens the first time this is called for a vcpu.
3937 * If the vcore is already running, we may be able to start
3938 * this thread straight away and have it join in.
3940 if (!signal_pending(current)) {
3941 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3942 vc->vcore_state == VCORE_RUNNING) &&
3943 !VCORE_IS_EXITING(vc)) {
3944 kvmppc_create_dtl_entry(vcpu, vc);
3945 kvmppc_start_thread(vcpu, vc);
3946 trace_kvm_guest_enter(vcpu);
3947 } else if (vc->vcore_state == VCORE_SLEEPING) {
3948 swake_up_one(&vc->wq);
3953 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3954 !signal_pending(current)) {
3955 /* See if the MMU is ready to go */
3956 if (!vcpu->kvm->arch.mmu_ready) {
3957 spin_unlock(&vc->lock);
3958 r = kvmhv_setup_mmu(vcpu);
3959 spin_lock(&vc->lock);
3961 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3962 kvm_run->fail_entry.
3963 hardware_entry_failure_reason = 0;
3969 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3970 kvmppc_vcore_end_preempt(vc);
3972 if (vc->vcore_state != VCORE_INACTIVE) {
3973 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3976 for_each_runnable_thread(i, v, vc) {
3977 kvmppc_core_prepare_to_enter(v);
3978 if (signal_pending(v->arch.run_task)) {
3979 kvmppc_remove_runnable(vc, v);
3980 v->stat.signal_exits++;
3981 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3982 v->arch.ret = -EINTR;
3983 wake_up(&v->arch.cpu_run);
3986 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3989 for_each_runnable_thread(i, v, vc) {
3990 if (!kvmppc_vcpu_woken(v))
3991 n_ceded += v->arch.ceded;
3996 if (n_ceded == vc->n_runnable) {
3997 kvmppc_vcore_blocked(vc);
3998 } else if (need_resched()) {
3999 kvmppc_vcore_preempt(vc);
4000 /* Let something else run */
4001 cond_resched_lock(&vc->lock);
4002 if (vc->vcore_state == VCORE_PREEMPT)
4003 kvmppc_vcore_end_preempt(vc);
4005 kvmppc_run_core(vc);
4010 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4011 (vc->vcore_state == VCORE_RUNNING ||
4012 vc->vcore_state == VCORE_EXITING ||
4013 vc->vcore_state == VCORE_PIGGYBACK))
4014 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4016 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4017 kvmppc_vcore_end_preempt(vc);
4019 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4020 kvmppc_remove_runnable(vc, vcpu);
4021 vcpu->stat.signal_exits++;
4022 kvm_run->exit_reason = KVM_EXIT_INTR;
4023 vcpu->arch.ret = -EINTR;
4026 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4027 /* Wake up some vcpu to run the core */
4029 v = next_runnable_thread(vc, &i);
4030 wake_up(&v->arch.cpu_run);
4033 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4034 spin_unlock(&vc->lock);
4035 return vcpu->arch.ret;
4038 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4039 struct kvm_vcpu *vcpu, u64 time_limit,
4044 struct kvmppc_vcore *vc;
4045 struct kvm *kvm = vcpu->kvm;
4046 struct kvm_nested_guest *nested = vcpu->arch.nested;
4048 trace_kvmppc_run_vcpu_enter(vcpu);
4050 kvm_run->exit_reason = 0;
4051 vcpu->arch.ret = RESUME_GUEST;
4052 vcpu->arch.trap = 0;
4054 vc = vcpu->arch.vcore;
4055 vcpu->arch.ceded = 0;
4056 vcpu->arch.run_task = current;
4057 vcpu->arch.kvm_run = kvm_run;
4058 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4059 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4060 vcpu->arch.busy_preempt = TB_NIL;
4061 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4062 vc->runnable_threads[0] = vcpu;
4066 /* See if the MMU is ready to go */
4067 if (!kvm->arch.mmu_ready)
4068 kvmhv_setup_mmu(vcpu);
4073 kvmppc_update_vpas(vcpu);
4075 init_vcore_to_run(vc);
4076 vc->preempt_tb = TB_NIL;
4079 pcpu = smp_processor_id();
4081 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4083 local_irq_disable();
4085 if (signal_pending(current))
4087 if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4091 kvmppc_core_prepare_to_enter(vcpu);
4092 if (vcpu->arch.doorbell_request) {
4095 vcpu->arch.doorbell_request = 0;
4097 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4098 &vcpu->arch.pending_exceptions))
4100 } else if (vcpu->arch.pending_exceptions ||
4101 vcpu->arch.doorbell_request ||
4102 xive_interrupt_pending(vcpu)) {
4103 vcpu->arch.ret = RESUME_HOST;
4107 kvmppc_clear_host_core(pcpu);
4109 local_paca->kvm_hstate.tid = 0;
4110 local_paca->kvm_hstate.napping = 0;
4111 local_paca->kvm_hstate.kvm_split_mode = NULL;
4112 kvmppc_start_thread(vcpu, vc);
4113 kvmppc_create_dtl_entry(vcpu, vc);
4114 trace_kvm_guest_enter(vcpu);
4116 vc->vcore_state = VCORE_RUNNING;
4117 trace_kvmppc_run_core(vc, 0);
4119 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4120 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4121 mtspr(SPRN_LPID, lpid);
4123 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4126 guest_enter_irqoff();
4128 srcu_idx = srcu_read_lock(&kvm->srcu);
4130 this_cpu_disable_ftrace();
4132 /* Tell lockdep that we're about to enable interrupts */
4133 trace_hardirqs_on();
4135 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4136 vcpu->arch.trap = trap;
4138 trace_hardirqs_off();
4140 this_cpu_enable_ftrace();
4142 srcu_read_unlock(&kvm->srcu, srcu_idx);
4144 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4145 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4149 set_irq_happened(trap);
4151 kvmppc_set_host_core(pcpu);
4156 cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4161 * cancel pending decrementer exception if DEC is now positive, or if
4162 * entering a nested guest in which case the decrementer is now owned
4163 * by L2 and the L1 decrementer is provided in hdec_expires
4165 if (kvmppc_core_pending_dec(vcpu) &&
4166 ((get_tb() < vcpu->arch.dec_expires) ||
4167 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4168 kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4169 kvmppc_core_dequeue_dec(vcpu);
4171 trace_kvm_guest_exit(vcpu);
4175 r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4177 r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4181 if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4182 !kvmppc_vcpu_woken(vcpu)) {
4183 kvmppc_set_timer(vcpu);
4184 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4185 if (signal_pending(current)) {
4186 vcpu->stat.signal_exits++;
4187 kvm_run->exit_reason = KVM_EXIT_INTR;
4188 vcpu->arch.ret = -EINTR;
4191 spin_lock(&vc->lock);
4192 kvmppc_vcore_blocked(vc);
4193 spin_unlock(&vc->lock);
4196 vcpu->arch.ceded = 0;
4198 vc->vcore_state = VCORE_INACTIVE;
4199 trace_kvmppc_run_core(vc, 1);
4202 kvmppc_remove_runnable(vc, vcpu);
4203 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4205 return vcpu->arch.ret;
4208 vcpu->stat.signal_exits++;
4209 kvm_run->exit_reason = KVM_EXIT_INTR;
4210 vcpu->arch.ret = -EINTR;
4217 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4221 unsigned long ebb_regs[3] = {}; /* shut up GCC */
4222 unsigned long user_tar = 0;
4223 unsigned int user_vrsave;
4226 if (!vcpu->arch.sane) {
4227 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4232 * Don't allow entry with a suspended transaction, because
4233 * the guest entry/exit code will lose it.
4234 * If the guest has TM enabled, save away their TM-related SPRs
4235 * (they will get restored by the TM unavailable interrupt).
4237 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4238 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4239 (current->thread.regs->msr & MSR_TM)) {
4240 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4241 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4242 run->fail_entry.hardware_entry_failure_reason = 0;
4245 /* Enable TM so we can read the TM SPRs */
4246 mtmsr(mfmsr() | MSR_TM);
4247 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4248 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4249 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4250 current->thread.regs->msr &= ~MSR_TM;
4255 * Force online to 1 for the sake of old userspace which doesn't
4258 if (!vcpu->arch.online) {
4259 atomic_inc(&vcpu->arch.vcore->online_count);
4260 vcpu->arch.online = 1;
4263 kvmppc_core_prepare_to_enter(vcpu);
4265 /* No need to go into the guest when all we'll do is come back out */
4266 if (signal_pending(current)) {
4267 run->exit_reason = KVM_EXIT_INTR;
4272 atomic_inc(&kvm->arch.vcpus_running);
4273 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4276 flush_all_to_thread(current);
4278 /* Save userspace EBB and other register values */
4279 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4280 ebb_regs[0] = mfspr(SPRN_EBBHR);
4281 ebb_regs[1] = mfspr(SPRN_EBBRR);
4282 ebb_regs[2] = mfspr(SPRN_BESCR);
4283 user_tar = mfspr(SPRN_TAR);
4285 user_vrsave = mfspr(SPRN_VRSAVE);
4287 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4288 vcpu->arch.pgdir = current->mm->pgd;
4289 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4293 * The early POWER9 chips that can't mix radix and HPT threads
4294 * on the same core also need the workaround for the problem
4295 * where the TLB would prefetch entries in the guest exit path
4296 * for radix guests using the guest PIDR value and LPID 0.
4297 * The workaround is in the old path (kvmppc_run_vcpu())
4298 * but not the new path (kvmhv_run_single_vcpu()).
4300 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4301 !no_mixing_hpt_and_radix)
4302 r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4303 vcpu->arch.vcore->lpcr);
4305 r = kvmppc_run_vcpu(run, vcpu);
4307 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4308 !(vcpu->arch.shregs.msr & MSR_PR)) {
4309 trace_kvm_hcall_enter(vcpu);
4310 r = kvmppc_pseries_do_hcall(vcpu);
4311 trace_kvm_hcall_exit(vcpu, r);
4312 kvmppc_core_prepare_to_enter(vcpu);
4313 } else if (r == RESUME_PAGE_FAULT) {
4314 srcu_idx = srcu_read_lock(&kvm->srcu);
4315 r = kvmppc_book3s_hv_page_fault(run, vcpu,
4316 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4317 srcu_read_unlock(&kvm->srcu, srcu_idx);
4318 } else if (r == RESUME_PASSTHROUGH) {
4319 if (WARN_ON(xics_on_xive()))
4322 r = kvmppc_xics_rm_complete(vcpu, 0);
4324 } while (is_kvmppc_resume_guest(r));
4326 /* Restore userspace EBB and other register values */
4327 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4328 mtspr(SPRN_EBBHR, ebb_regs[0]);
4329 mtspr(SPRN_EBBRR, ebb_regs[1]);
4330 mtspr(SPRN_BESCR, ebb_regs[2]);
4331 mtspr(SPRN_TAR, user_tar);
4332 mtspr(SPRN_FSCR, current->thread.fscr);
4334 mtspr(SPRN_VRSAVE, user_vrsave);
4336 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4337 atomic_dec(&kvm->arch.vcpus_running);
4341 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4342 int shift, int sllp)
4344 (*sps)->page_shift = shift;
4345 (*sps)->slb_enc = sllp;
4346 (*sps)->enc[0].page_shift = shift;
4347 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4349 * Add 16MB MPSS support (may get filtered out by userspace)
4352 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4354 (*sps)->enc[1].page_shift = 24;
4355 (*sps)->enc[1].pte_enc = penc;
4361 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4362 struct kvm_ppc_smmu_info *info)
4364 struct kvm_ppc_one_seg_page_size *sps;
4367 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4368 * POWER7 doesn't support keys for instruction accesses,
4369 * POWER8 and POWER9 do.
4371 info->data_keys = 32;
4372 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4374 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4375 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4376 info->slb_size = 32;
4378 /* We only support these sizes for now, and no muti-size segments */
4379 sps = &info->sps[0];
4380 kvmppc_add_seg_page_size(&sps, 12, 0);
4381 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4382 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4384 /* If running as a nested hypervisor, we don't support HPT guests */
4385 if (kvmhv_on_pseries())
4386 info->flags |= KVM_PPC_NO_HASH;
4392 * Get (and clear) the dirty memory log for a memory slot.
4394 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4395 struct kvm_dirty_log *log)
4397 struct kvm_memslots *slots;
4398 struct kvm_memory_slot *memslot;
4401 unsigned long *buf, *p;
4402 struct kvm_vcpu *vcpu;
4404 mutex_lock(&kvm->slots_lock);
4407 if (log->slot >= KVM_USER_MEM_SLOTS)
4410 slots = kvm_memslots(kvm);
4411 memslot = id_to_memslot(slots, log->slot);
4413 if (!memslot->dirty_bitmap)
4417 * Use second half of bitmap area because both HPT and radix
4418 * accumulate bits in the first half.
4420 n = kvm_dirty_bitmap_bytes(memslot);
4421 buf = memslot->dirty_bitmap + n / sizeof(long);
4424 if (kvm_is_radix(kvm))
4425 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4427 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4432 * We accumulate dirty bits in the first half of the
4433 * memslot's dirty_bitmap area, for when pages are paged
4434 * out or modified by the host directly. Pick up these
4435 * bits and add them to the map.
4437 p = memslot->dirty_bitmap;
4438 for (i = 0; i < n / sizeof(long); ++i)
4439 buf[i] |= xchg(&p[i], 0);
4441 /* Harvest dirty bits from VPA and DTL updates */
4442 /* Note: we never modify the SLB shadow buffer areas */
4443 kvm_for_each_vcpu(i, vcpu, kvm) {
4444 spin_lock(&vcpu->arch.vpa_update_lock);
4445 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4446 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4447 spin_unlock(&vcpu->arch.vpa_update_lock);
4451 if (copy_to_user(log->dirty_bitmap, buf, n))
4456 mutex_unlock(&kvm->slots_lock);
4460 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4461 struct kvm_memory_slot *dont)
4463 if (!dont || free->arch.rmap != dont->arch.rmap) {
4464 vfree(free->arch.rmap);
4465 free->arch.rmap = NULL;
4469 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4470 unsigned long npages)
4472 slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4473 if (!slot->arch.rmap)
4479 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4480 struct kvm_memory_slot *memslot,
4481 const struct kvm_userspace_memory_region *mem)
4486 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4487 const struct kvm_userspace_memory_region *mem,
4488 const struct kvm_memory_slot *old,
4489 const struct kvm_memory_slot *new,
4490 enum kvm_mr_change change)
4492 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4495 * If we are making a new memslot, it might make
4496 * some address that was previously cached as emulated
4497 * MMIO be no longer emulated MMIO, so invalidate
4498 * all the caches of emulated MMIO translations.
4501 atomic64_inc(&kvm->arch.mmio_update);
4504 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4505 * have already called kvm_arch_flush_shadow_memslot() to
4506 * flush shadow mappings. For KVM_MR_CREATE we have no
4507 * previous mappings. So the only case to handle is
4508 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4510 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4511 * to get rid of any THP PTEs in the partition-scoped page tables
4512 * so we can track dirtiness at the page level; we flush when
4513 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4516 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4517 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4518 kvmppc_radix_flush_memslot(kvm, old);
4520 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4522 if (!kvm->arch.secure_guest)
4527 if (kvmppc_uvmem_slot_init(kvm, new))
4529 uv_register_mem_slot(kvm->arch.lpid,
4530 new->base_gfn << PAGE_SHIFT,
4531 new->npages * PAGE_SIZE,
4535 uv_unregister_mem_slot(kvm->arch.lpid, old->id);
4536 kvmppc_uvmem_slot_free(kvm, old);
4539 /* TODO: Handle KVM_MR_MOVE */
4545 * Update LPCR values in kvm->arch and in vcores.
4546 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4547 * of kvm->arch.lpcr update).
4549 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4554 if ((kvm->arch.lpcr & mask) == lpcr)
4557 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4559 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4560 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4563 spin_lock(&vc->lock);
4564 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4565 spin_unlock(&vc->lock);
4566 if (++cores_done >= kvm->arch.online_vcores)
4571 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4576 void kvmppc_setup_partition_table(struct kvm *kvm)
4578 unsigned long dw0, dw1;
4580 if (!kvm_is_radix(kvm)) {
4581 /* PS field - page size for VRMA */
4582 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4583 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4584 /* HTABSIZE and HTABORG fields */
4585 dw0 |= kvm->arch.sdr1;
4587 /* Second dword as set by userspace */
4588 dw1 = kvm->arch.process_table;
4590 dw0 = PATB_HR | radix__get_tree_size() |
4591 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4592 dw1 = PATB_GR | kvm->arch.process_table;
4594 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4598 * Set up HPT (hashed page table) and RMA (real-mode area).
4599 * Must be called with kvm->arch.mmu_setup_lock held.
4601 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4604 struct kvm *kvm = vcpu->kvm;
4606 struct kvm_memory_slot *memslot;
4607 struct vm_area_struct *vma;
4608 unsigned long lpcr = 0, senc;
4609 unsigned long psize, porder;
4612 /* Allocate hashed page table (if not done already) and reset it */
4613 if (!kvm->arch.hpt.virt) {
4614 int order = KVM_DEFAULT_HPT_ORDER;
4615 struct kvm_hpt_info info;
4617 err = kvmppc_allocate_hpt(&info, order);
4618 /* If we get here, it means userspace didn't specify a
4619 * size explicitly. So, try successively smaller
4620 * sizes if the default failed. */
4621 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4622 err = kvmppc_allocate_hpt(&info, order);
4625 pr_err("KVM: Couldn't alloc HPT\n");
4629 kvmppc_set_hpt(kvm, &info);
4632 /* Look up the memslot for guest physical address 0 */
4633 srcu_idx = srcu_read_lock(&kvm->srcu);
4634 memslot = gfn_to_memslot(kvm, 0);
4636 /* We must have some memory at 0 by now */
4638 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4641 /* Look up the VMA for the start of this memory slot */
4642 hva = memslot->userspace_addr;
4643 down_read(¤t->mm->mmap_sem);
4644 vma = find_vma(current->mm, hva);
4645 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4648 psize = vma_kernel_pagesize(vma);
4650 up_read(¤t->mm->mmap_sem);
4652 /* We can handle 4k, 64k or 16M pages in the VRMA */
4653 if (psize >= 0x1000000)
4655 else if (psize >= 0x10000)
4659 porder = __ilog2(psize);
4661 senc = slb_pgsize_encoding(psize);
4662 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4663 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4664 /* Create HPTEs in the hash page table for the VRMA */
4665 kvmppc_map_vrma(vcpu, memslot, porder);
4667 /* Update VRMASD field in the LPCR */
4668 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4669 /* the -4 is to account for senc values starting at 0x10 */
4670 lpcr = senc << (LPCR_VRMASD_SH - 4);
4671 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4674 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4678 srcu_read_unlock(&kvm->srcu, srcu_idx);
4683 up_read(¤t->mm->mmap_sem);
4688 * Must be called with kvm->arch.mmu_setup_lock held and
4689 * mmu_ready = 0 and no vcpus running.
4691 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4693 if (nesting_enabled(kvm))
4694 kvmhv_release_all_nested(kvm);
4695 kvmppc_rmap_reset(kvm);
4696 kvm->arch.process_table = 0;
4697 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4698 spin_lock(&kvm->mmu_lock);
4699 kvm->arch.radix = 0;
4700 spin_unlock(&kvm->mmu_lock);
4701 kvmppc_free_radix(kvm);
4702 kvmppc_update_lpcr(kvm, LPCR_VPM1,
4703 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4708 * Must be called with kvm->arch.mmu_setup_lock held and
4709 * mmu_ready = 0 and no vcpus running.
4711 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4715 err = kvmppc_init_vm_radix(kvm);
4718 kvmppc_rmap_reset(kvm);
4719 /* Mutual exclusion with kvm_unmap_hva_range etc. */
4720 spin_lock(&kvm->mmu_lock);
4721 kvm->arch.radix = 1;
4722 spin_unlock(&kvm->mmu_lock);
4723 kvmppc_free_hpt(&kvm->arch.hpt);
4724 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4725 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4729 #ifdef CONFIG_KVM_XICS
4731 * Allocate a per-core structure for managing state about which cores are
4732 * running in the host versus the guest and for exchanging data between
4733 * real mode KVM and CPU running in the host.
4734 * This is only done for the first VM.
4735 * The allocated structure stays even if all VMs have stopped.
4736 * It is only freed when the kvm-hv module is unloaded.
4737 * It's OK for this routine to fail, we just don't support host
4738 * core operations like redirecting H_IPI wakeups.
4740 void kvmppc_alloc_host_rm_ops(void)
4742 struct kvmppc_host_rm_ops *ops;
4743 unsigned long l_ops;
4747 /* Not the first time here ? */
4748 if (kvmppc_host_rm_ops_hv != NULL)
4751 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4755 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4756 ops->rm_core = kzalloc(size, GFP_KERNEL);
4758 if (!ops->rm_core) {
4765 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4766 if (!cpu_online(cpu))
4769 core = cpu >> threads_shift;
4770 ops->rm_core[core].rm_state.in_host = 1;
4773 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4776 * Make the contents of the kvmppc_host_rm_ops structure visible
4777 * to other CPUs before we assign it to the global variable.
4778 * Do an atomic assignment (no locks used here), but if someone
4779 * beats us to it, just free our copy and return.
4782 l_ops = (unsigned long) ops;
4784 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4786 kfree(ops->rm_core);
4791 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4792 "ppc/kvm_book3s:prepare",
4793 kvmppc_set_host_core,
4794 kvmppc_clear_host_core);
4798 void kvmppc_free_host_rm_ops(void)
4800 if (kvmppc_host_rm_ops_hv) {
4801 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4802 kfree(kvmppc_host_rm_ops_hv->rm_core);
4803 kfree(kvmppc_host_rm_ops_hv);
4804 kvmppc_host_rm_ops_hv = NULL;
4809 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4811 unsigned long lpcr, lpid;
4815 mutex_init(&kvm->arch.uvmem_lock);
4816 INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4817 mutex_init(&kvm->arch.mmu_setup_lock);
4819 /* Allocate the guest's logical partition ID */
4821 lpid = kvmppc_alloc_lpid();
4824 kvm->arch.lpid = lpid;
4826 kvmppc_alloc_host_rm_ops();
4828 kvmhv_vm_nested_init(kvm);
4831 * Since we don't flush the TLB when tearing down a VM,
4832 * and this lpid might have previously been used,
4833 * make sure we flush on each core before running the new VM.
4834 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4835 * does this flush for us.
4837 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4838 cpumask_setall(&kvm->arch.need_tlb_flush);
4840 /* Start out with the default set of hcalls enabled */
4841 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4842 sizeof(kvm->arch.enabled_hcalls));
4844 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4845 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4847 /* Init LPCR for virtual RMA mode */
4848 if (cpu_has_feature(CPU_FTR_HVMODE)) {
4849 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4850 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4851 lpcr &= LPCR_PECE | LPCR_LPES;
4855 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4856 LPCR_VPM0 | LPCR_VPM1;
4857 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4858 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4859 /* On POWER8 turn on online bit to enable PURR/SPURR */
4860 if (cpu_has_feature(CPU_FTR_ARCH_207S))
4863 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4864 * Set HVICE bit to enable hypervisor virtualization interrupts.
4865 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4866 * be unnecessary but better safe than sorry in case we re-enable
4867 * EE in HV mode with this LPCR still set)
4869 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4871 lpcr |= LPCR_HVICE | LPCR_HEIC;
4874 * If xive is enabled, we route 0x500 interrupts directly
4882 * If the host uses radix, the guest starts out as radix.
4884 if (radix_enabled()) {
4885 kvm->arch.radix = 1;
4886 kvm->arch.mmu_ready = 1;
4888 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4889 ret = kvmppc_init_vm_radix(kvm);
4891 kvmppc_free_lpid(kvm->arch.lpid);
4894 kvmppc_setup_partition_table(kvm);
4897 kvm->arch.lpcr = lpcr;
4899 /* Initialization for future HPT resizes */
4900 kvm->arch.resize_hpt = NULL;
4903 * Work out how many sets the TLB has, for the use of
4904 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4906 if (radix_enabled())
4907 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
4908 else if (cpu_has_feature(CPU_FTR_ARCH_300))
4909 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
4910 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4911 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
4913 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
4916 * Track that we now have a HV mode VM active. This blocks secondary
4917 * CPU threads from coming online.
4918 * On POWER9, we only need to do this if the "indep_threads_mode"
4919 * module parameter has been set to N.
4921 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4922 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4923 pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4924 kvm->arch.threads_indep = true;
4926 kvm->arch.threads_indep = indep_threads_mode;
4929 if (!kvm->arch.threads_indep)
4930 kvm_hv_vm_activated();
4933 * Initialize smt_mode depending on processor.
4934 * POWER8 and earlier have to use "strict" threading, where
4935 * all vCPUs in a vcore have to run on the same (sub)core,
4936 * whereas on POWER9 the threads can each run a different
4939 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4940 kvm->arch.smt_mode = threads_per_subcore;
4942 kvm->arch.smt_mode = 1;
4943 kvm->arch.emul_smt_mode = 1;
4946 * Create a debugfs directory for the VM
4948 snprintf(buf, sizeof(buf), "vm%d", current->pid);
4949 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4950 kvmppc_mmu_debugfs_init(kvm);
4951 if (radix_enabled())
4952 kvmhv_radix_debugfs_init(kvm);
4957 static void kvmppc_free_vcores(struct kvm *kvm)
4961 for (i = 0; i < KVM_MAX_VCORES; ++i)
4962 kfree(kvm->arch.vcores[i]);
4963 kvm->arch.online_vcores = 0;
4966 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4968 debugfs_remove_recursive(kvm->arch.debugfs_dir);
4970 if (!kvm->arch.threads_indep)
4971 kvm_hv_vm_deactivated();
4973 kvmppc_free_vcores(kvm);
4976 if (kvm_is_radix(kvm))
4977 kvmppc_free_radix(kvm);
4979 kvmppc_free_hpt(&kvm->arch.hpt);
4981 /* Perform global invalidation and return lpid to the pool */
4982 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4983 if (nesting_enabled(kvm))
4984 kvmhv_release_all_nested(kvm);
4985 kvm->arch.process_table = 0;
4986 if (kvm->arch.secure_guest)
4987 uv_svm_terminate(kvm->arch.lpid);
4988 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4991 kvmppc_free_lpid(kvm->arch.lpid);
4993 kvmppc_free_pimap(kvm);
4996 /* We don't need to emulate any privileged instructions or dcbz */
4997 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4998 unsigned int inst, int *advance)
5000 return EMULATE_FAIL;
5003 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5006 return EMULATE_FAIL;
5009 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5012 return EMULATE_FAIL;
5015 static int kvmppc_core_check_processor_compat_hv(void)
5017 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5018 cpu_has_feature(CPU_FTR_ARCH_206))
5021 /* POWER9 in radix mode is capable of being a nested hypervisor. */
5022 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5028 #ifdef CONFIG_KVM_XICS
5030 void kvmppc_free_pimap(struct kvm *kvm)
5032 kfree(kvm->arch.pimap);
5035 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5037 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5040 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5042 struct irq_desc *desc;
5043 struct kvmppc_irq_map *irq_map;
5044 struct kvmppc_passthru_irqmap *pimap;
5045 struct irq_chip *chip;
5048 if (!kvm_irq_bypass)
5051 desc = irq_to_desc(host_irq);
5055 mutex_lock(&kvm->lock);
5057 pimap = kvm->arch.pimap;
5058 if (pimap == NULL) {
5059 /* First call, allocate structure to hold IRQ map */
5060 pimap = kvmppc_alloc_pimap();
5061 if (pimap == NULL) {
5062 mutex_unlock(&kvm->lock);
5065 kvm->arch.pimap = pimap;
5069 * For now, we only support interrupts for which the EOI operation
5070 * is an OPAL call followed by a write to XIRR, since that's
5071 * what our real-mode EOI code does, or a XIVE interrupt
5073 chip = irq_data_get_irq_chip(&desc->irq_data);
5074 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5075 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5076 host_irq, guest_gsi);
5077 mutex_unlock(&kvm->lock);
5082 * See if we already have an entry for this guest IRQ number.
5083 * If it's mapped to a hardware IRQ number, that's an error,
5084 * otherwise re-use this entry.
5086 for (i = 0; i < pimap->n_mapped; i++) {
5087 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5088 if (pimap->mapped[i].r_hwirq) {
5089 mutex_unlock(&kvm->lock);
5096 if (i == KVMPPC_PIRQ_MAPPED) {
5097 mutex_unlock(&kvm->lock);
5098 return -EAGAIN; /* table is full */
5101 irq_map = &pimap->mapped[i];
5103 irq_map->v_hwirq = guest_gsi;
5104 irq_map->desc = desc;
5107 * Order the above two stores before the next to serialize with
5108 * the KVM real mode handler.
5111 irq_map->r_hwirq = desc->irq_data.hwirq;
5113 if (i == pimap->n_mapped)
5117 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5119 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5121 irq_map->r_hwirq = 0;
5123 mutex_unlock(&kvm->lock);
5128 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5130 struct irq_desc *desc;
5131 struct kvmppc_passthru_irqmap *pimap;
5134 if (!kvm_irq_bypass)
5137 desc = irq_to_desc(host_irq);
5141 mutex_lock(&kvm->lock);
5142 if (!kvm->arch.pimap)
5145 pimap = kvm->arch.pimap;
5147 for (i = 0; i < pimap->n_mapped; i++) {
5148 if (guest_gsi == pimap->mapped[i].v_hwirq)
5152 if (i == pimap->n_mapped) {
5153 mutex_unlock(&kvm->lock);
5158 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5160 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5162 /* invalidate the entry (what do do on error from the above ?) */
5163 pimap->mapped[i].r_hwirq = 0;
5166 * We don't free this structure even when the count goes to
5167 * zero. The structure is freed when we destroy the VM.
5170 mutex_unlock(&kvm->lock);
5174 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5175 struct irq_bypass_producer *prod)
5178 struct kvm_kernel_irqfd *irqfd =
5179 container_of(cons, struct kvm_kernel_irqfd, consumer);
5181 irqfd->producer = prod;
5183 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5185 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5186 prod->irq, irqfd->gsi, ret);
5191 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5192 struct irq_bypass_producer *prod)
5195 struct kvm_kernel_irqfd *irqfd =
5196 container_of(cons, struct kvm_kernel_irqfd, consumer);
5198 irqfd->producer = NULL;
5201 * When producer of consumer is unregistered, we change back to
5202 * default external interrupt handling mode - KVM real mode
5203 * will switch back to host.
5205 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5207 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5208 prod->irq, irqfd->gsi, ret);
5212 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5213 unsigned int ioctl, unsigned long arg)
5215 struct kvm *kvm __maybe_unused = filp->private_data;
5216 void __user *argp = (void __user *)arg;
5221 case KVM_PPC_ALLOCATE_HTAB: {
5225 if (get_user(htab_order, (u32 __user *)argp))
5227 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5234 case KVM_PPC_GET_HTAB_FD: {
5235 struct kvm_get_htab_fd ghf;
5238 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5240 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5244 case KVM_PPC_RESIZE_HPT_PREPARE: {
5245 struct kvm_ppc_resize_hpt rhpt;
5248 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5251 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5255 case KVM_PPC_RESIZE_HPT_COMMIT: {
5256 struct kvm_ppc_resize_hpt rhpt;
5259 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5262 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5274 * List of hcall numbers to enable by default.
5275 * For compatibility with old userspace, we enable by default
5276 * all hcalls that were implemented before the hcall-enabling
5277 * facility was added. Note this list should not include H_RTAS.
5279 static unsigned int default_hcall_list[] = {
5293 #ifdef CONFIG_KVM_XICS
5304 static void init_default_hcalls(void)
5309 for (i = 0; default_hcall_list[i]; ++i) {
5310 hcall = default_hcall_list[i];
5311 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5312 __set_bit(hcall / 4, default_enabled_hcalls);
5316 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5322 /* If not on a POWER9, reject it */
5323 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5326 /* If any unknown flags set, reject it */
5327 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5330 /* GR (guest radix) bit in process_table field must match */
5331 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5332 if (!!(cfg->process_table & PATB_GR) != radix)
5335 /* Process table size field must be reasonable, i.e. <= 24 */
5336 if ((cfg->process_table & PRTS_MASK) > 24)
5339 /* We can change a guest to/from radix now, if the host is radix */
5340 if (radix && !radix_enabled())
5343 /* If we're a nested hypervisor, we currently only support radix */
5344 if (kvmhv_on_pseries() && !radix)
5347 mutex_lock(&kvm->arch.mmu_setup_lock);
5348 if (radix != kvm_is_radix(kvm)) {
5349 if (kvm->arch.mmu_ready) {
5350 kvm->arch.mmu_ready = 0;
5351 /* order mmu_ready vs. vcpus_running */
5353 if (atomic_read(&kvm->arch.vcpus_running)) {
5354 kvm->arch.mmu_ready = 1;
5360 err = kvmppc_switch_mmu_to_radix(kvm);
5362 err = kvmppc_switch_mmu_to_hpt(kvm);
5367 kvm->arch.process_table = cfg->process_table;
5368 kvmppc_setup_partition_table(kvm);
5370 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5371 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5375 mutex_unlock(&kvm->arch.mmu_setup_lock);
5379 static int kvmhv_enable_nested(struct kvm *kvm)
5383 if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5386 /* kvm == NULL means the caller is testing if the capability exists */
5388 kvm->arch.nested_enable = true;
5392 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5397 if (kvmhv_vcpu_is_radix(vcpu)) {
5398 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5404 /* For now quadrants are the only way to access nested guest memory */
5405 if (rc && vcpu->arch.nested)
5411 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5416 if (kvmhv_vcpu_is_radix(vcpu)) {
5417 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5423 /* For now quadrants are the only way to access nested guest memory */
5424 if (rc && vcpu->arch.nested)
5430 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5432 unpin_vpa(kvm, vpa);
5434 vpa->pinned_addr = NULL;
5436 vpa->update_pending = 0;
5440 * IOCTL handler to turn off secure mode of guest
5442 * - Release all device pages
5443 * - Issue ucall to terminate the guest on the UV side
5444 * - Unpin the VPA pages.
5445 * - Reinit the partition scoped page tables
5447 static int kvmhv_svm_off(struct kvm *kvm)
5449 struct kvm_vcpu *vcpu;
5455 if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5458 mutex_lock(&kvm->arch.mmu_setup_lock);
5459 mmu_was_ready = kvm->arch.mmu_ready;
5460 if (kvm->arch.mmu_ready) {
5461 kvm->arch.mmu_ready = 0;
5462 /* order mmu_ready vs. vcpus_running */
5464 if (atomic_read(&kvm->arch.vcpus_running)) {
5465 kvm->arch.mmu_ready = 1;
5471 srcu_idx = srcu_read_lock(&kvm->srcu);
5472 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5473 struct kvm_memory_slot *memslot;
5474 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5479 kvm_for_each_memslot(memslot, slots) {
5480 kvmppc_uvmem_drop_pages(memslot, kvm);
5481 uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5484 srcu_read_unlock(&kvm->srcu, srcu_idx);
5486 ret = uv_svm_terminate(kvm->arch.lpid);
5487 if (ret != U_SUCCESS) {
5493 * When secure guest is reset, all the guest pages are sent
5494 * to UV via UV_PAGE_IN before the non-boot vcpus get a
5495 * chance to run and unpin their VPA pages. Unpinning of all
5496 * VPA pages is done here explicitly so that VPA pages
5497 * can be migrated to the secure side.
5499 * This is required to for the secure SMP guest to reboot
5502 kvm_for_each_vcpu(i, vcpu, kvm) {
5503 spin_lock(&vcpu->arch.vpa_update_lock);
5504 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5505 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5506 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5507 spin_unlock(&vcpu->arch.vpa_update_lock);
5510 kvmppc_setup_partition_table(kvm);
5511 kvm->arch.secure_guest = 0;
5512 kvm->arch.mmu_ready = mmu_was_ready;
5514 mutex_unlock(&kvm->arch.mmu_setup_lock);
5518 static struct kvmppc_ops kvm_ops_hv = {
5519 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5520 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5521 .get_one_reg = kvmppc_get_one_reg_hv,
5522 .set_one_reg = kvmppc_set_one_reg_hv,
5523 .vcpu_load = kvmppc_core_vcpu_load_hv,
5524 .vcpu_put = kvmppc_core_vcpu_put_hv,
5525 .inject_interrupt = kvmppc_inject_interrupt_hv,
5526 .set_msr = kvmppc_set_msr_hv,
5527 .vcpu_run = kvmppc_vcpu_run_hv,
5528 .vcpu_create = kvmppc_core_vcpu_create_hv,
5529 .vcpu_free = kvmppc_core_vcpu_free_hv,
5530 .check_requests = kvmppc_core_check_requests_hv,
5531 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
5532 .flush_memslot = kvmppc_core_flush_memslot_hv,
5533 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5534 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
5535 .unmap_hva_range = kvm_unmap_hva_range_hv,
5536 .age_hva = kvm_age_hva_hv,
5537 .test_age_hva = kvm_test_age_hva_hv,
5538 .set_spte_hva = kvm_set_spte_hva_hv,
5539 .mmu_destroy = kvmppc_mmu_destroy_hv,
5540 .free_memslot = kvmppc_core_free_memslot_hv,
5541 .create_memslot = kvmppc_core_create_memslot_hv,
5542 .init_vm = kvmppc_core_init_vm_hv,
5543 .destroy_vm = kvmppc_core_destroy_vm_hv,
5544 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5545 .emulate_op = kvmppc_core_emulate_op_hv,
5546 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5547 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5548 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5549 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
5550 .hcall_implemented = kvmppc_hcall_impl_hv,
5551 #ifdef CONFIG_KVM_XICS
5552 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5553 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5555 .configure_mmu = kvmhv_configure_mmu,
5556 .get_rmmu_info = kvmhv_get_rmmu_info,
5557 .set_smt_mode = kvmhv_set_smt_mode,
5558 .enable_nested = kvmhv_enable_nested,
5559 .load_from_eaddr = kvmhv_load_from_eaddr,
5560 .store_to_eaddr = kvmhv_store_to_eaddr,
5561 .svm_off = kvmhv_svm_off,
5564 static int kvm_init_subcore_bitmap(void)
5567 int nr_cores = cpu_nr_cores();
5568 struct sibling_subcore_state *sibling_subcore_state;
5570 for (i = 0; i < nr_cores; i++) {
5571 int first_cpu = i * threads_per_core;
5572 int node = cpu_to_node(first_cpu);
5574 /* Ignore if it is already allocated. */
5575 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5578 sibling_subcore_state =
5579 kzalloc_node(sizeof(struct sibling_subcore_state),
5581 if (!sibling_subcore_state)
5585 for (j = 0; j < threads_per_core; j++) {
5586 int cpu = first_cpu + j;
5588 paca_ptrs[cpu]->sibling_subcore_state =
5589 sibling_subcore_state;
5595 static int kvmppc_radix_possible(void)
5597 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5600 static int kvmppc_book3s_init_hv(void)
5604 if (!tlbie_capable) {
5605 pr_err("KVM-HV: Host does not support TLBIE\n");
5610 * FIXME!! Do we need to check on all cpus ?
5612 r = kvmppc_core_check_processor_compat_hv();
5616 r = kvmhv_nested_init();
5620 r = kvm_init_subcore_bitmap();
5625 * We need a way of accessing the XICS interrupt controller,
5626 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5627 * indirectly, via OPAL.
5630 if (!xics_on_xive() && !kvmhv_on_pseries() &&
5631 !local_paca->kvm_hstate.xics_phys) {
5632 struct device_node *np;
5634 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5636 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5639 /* presence of intc confirmed - node can be dropped again */
5644 kvm_ops_hv.owner = THIS_MODULE;
5645 kvmppc_hv_ops = &kvm_ops_hv;
5647 init_default_hcalls();
5651 r = kvmppc_mmu_hv_init();
5655 if (kvmppc_radix_possible())
5656 r = kvmppc_radix_init();
5659 * POWER9 chips before version 2.02 can't have some threads in
5660 * HPT mode and some in radix mode on the same core.
5662 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5663 unsigned int pvr = mfspr(SPRN_PVR);
5664 if ((pvr >> 16) == PVR_POWER9 &&
5665 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5666 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5667 no_mixing_hpt_and_radix = true;
5670 r = kvmppc_uvmem_init();
5672 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5677 static void kvmppc_book3s_exit_hv(void)
5679 kvmppc_uvmem_free();
5680 kvmppc_free_host_rm_ops();
5681 if (kvmppc_radix_possible())
5682 kvmppc_radix_exit();
5683 kvmppc_hv_ops = NULL;
5684 kvmhv_nested_exit();
5687 module_init(kvmppc_book3s_init_hv);
5688 module_exit(kvmppc_book3s_exit_hv);
5689 MODULE_LICENSE("GPL");
5690 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5691 MODULE_ALIAS("devname:kvm");