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>
56 #include <asm/interrupt.h>
58 #include <asm/kvm_ppc.h>
59 #include <asm/kvm_book3s.h>
60 #include <asm/mmu_context.h>
61 #include <asm/lppaca.h>
62 #include <asm/processor.h>
63 #include <asm/cputhreads.h>
65 #include <asm/hvcall.h>
66 #include <asm/switch_to.h>
68 #include <asm/dbell.h>
70 #include <asm/pnv-pci.h>
75 #include <asm/hw_breakpoint.h>
76 #include <asm/kvm_book3s_uvmem.h>
77 #include <asm/ultravisor.h>
79 #include <asm/plpar_wrappers.h>
83 #define CREATE_TRACE_POINTS
86 /* #define EXIT_DEBUG */
87 /* #define EXIT_DEBUG_SIMPLE */
88 /* #define EXIT_DEBUG_INT */
90 /* Used to indicate that a guest page fault needs to be handled */
91 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
92 /* Used to indicate that a guest passthrough interrupt needs to be handled */
93 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
95 /* Used as a "null" value for timebase values */
96 #define TB_NIL (~(u64)0)
98 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
100 static int dynamic_mt_modes = 6;
101 module_param(dynamic_mt_modes, int, 0644);
102 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
103 static int target_smt_mode;
104 module_param(target_smt_mode, int, 0644);
105 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
107 static bool one_vm_per_core;
108 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
109 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
111 #ifdef CONFIG_KVM_XICS
112 static const struct kernel_param_ops module_param_ops = {
113 .set = param_set_int,
114 .get = param_get_int,
117 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
118 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
120 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
121 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
124 /* If set, guests are allowed to create and control nested guests */
125 static bool nested = true;
126 module_param(nested, bool, S_IRUGO | S_IWUSR);
127 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
129 static inline bool nesting_enabled(struct kvm *kvm)
131 return kvm->arch.nested_enable && kvm_is_radix(kvm);
134 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
137 * RWMR values for POWER8. These control the rate at which PURR
138 * and SPURR count and should be set according to the number of
139 * online threads in the vcore being run.
141 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL
142 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL
143 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL
144 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL
145 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL
146 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL
147 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL
148 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL
150 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
162 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
166 struct kvm_vcpu *vcpu;
168 while (++i < MAX_SMT_THREADS) {
169 vcpu = READ_ONCE(vc->runnable_threads[i]);
178 /* Used to traverse the list of runnable threads for a given vcore */
179 #define for_each_runnable_thread(i, vcpu, vc) \
180 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
182 static bool kvmppc_ipi_thread(int cpu)
184 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
186 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
187 if (kvmhv_on_pseries())
190 /* On POWER9 we can use msgsnd to IPI any cpu */
191 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
192 msg |= get_hard_smp_processor_id(cpu);
194 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
198 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
199 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
201 if (cpu_first_thread_sibling(cpu) ==
202 cpu_first_thread_sibling(smp_processor_id())) {
203 msg |= cpu_thread_in_core(cpu);
205 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
212 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
213 if (cpu >= 0 && cpu < nr_cpu_ids) {
214 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
218 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
226 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
229 struct rcuwait *waitp;
231 waitp = kvm_arch_vcpu_get_wait(vcpu);
232 if (rcuwait_wake_up(waitp))
233 ++vcpu->stat.generic.halt_wakeup;
235 cpu = READ_ONCE(vcpu->arch.thread_cpu);
236 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
239 /* CPU points to the first thread of the core */
241 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
242 smp_send_reschedule(cpu);
246 * We use the vcpu_load/put functions to measure stolen time.
247 * Stolen time is counted as time when either the vcpu is able to
248 * run as part of a virtual core, but the task running the vcore
249 * is preempted or sleeping, or when the vcpu needs something done
250 * in the kernel by the task running the vcpu, but that task is
251 * preempted or sleeping. Those two things have to be counted
252 * separately, since one of the vcpu tasks will take on the job
253 * of running the core, and the other vcpu tasks in the vcore will
254 * sleep waiting for it to do that, but that sleep shouldn't count
257 * Hence we accumulate stolen time when the vcpu can run as part of
258 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
259 * needs its task to do other things in the kernel (for example,
260 * service a page fault) in busy_stolen. We don't accumulate
261 * stolen time for a vcore when it is inactive, or for a vcpu
262 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
263 * a misnomer; it means that the vcpu task is not executing in
264 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
265 * the kernel. We don't have any way of dividing up that time
266 * between time that the vcpu is genuinely stopped, time that
267 * the task is actively working on behalf of the vcpu, and time
268 * that the task is preempted, so we don't count any of it as
271 * Updates to busy_stolen are protected by arch.tbacct_lock;
272 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
273 * lock. The stolen times are measured in units of timebase ticks.
274 * (Note that the != TB_NIL checks below are purely defensive;
275 * they should never fail.)
278 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
282 spin_lock_irqsave(&vc->stoltb_lock, flags);
283 vc->preempt_tb = mftb();
284 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
287 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
291 spin_lock_irqsave(&vc->stoltb_lock, flags);
292 if (vc->preempt_tb != TB_NIL) {
293 vc->stolen_tb += mftb() - vc->preempt_tb;
294 vc->preempt_tb = TB_NIL;
296 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
299 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
301 struct kvmppc_vcore *vc = vcpu->arch.vcore;
305 * We can test vc->runner without taking the vcore lock,
306 * because only this task ever sets vc->runner to this
307 * vcpu, and once it is set to this vcpu, only this task
308 * ever sets it to NULL.
310 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
311 kvmppc_core_end_stolen(vc);
313 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
314 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
315 vcpu->arch.busy_preempt != TB_NIL) {
316 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
317 vcpu->arch.busy_preempt = TB_NIL;
319 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
322 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
324 struct kvmppc_vcore *vc = vcpu->arch.vcore;
327 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
328 kvmppc_core_start_stolen(vc);
330 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
331 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
332 vcpu->arch.busy_preempt = mftb();
333 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
336 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
338 vcpu->arch.pvr = pvr;
341 /* Dummy value used in computing PCR value below */
342 #define PCR_ARCH_31 (PCR_ARCH_300 << 1)
344 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
346 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
347 struct kvmppc_vcore *vc = vcpu->arch.vcore;
349 /* We can (emulate) our own architecture version and anything older */
350 if (cpu_has_feature(CPU_FTR_ARCH_31))
351 host_pcr_bit = PCR_ARCH_31;
352 else if (cpu_has_feature(CPU_FTR_ARCH_300))
353 host_pcr_bit = PCR_ARCH_300;
354 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
355 host_pcr_bit = PCR_ARCH_207;
356 else if (cpu_has_feature(CPU_FTR_ARCH_206))
357 host_pcr_bit = PCR_ARCH_206;
359 host_pcr_bit = PCR_ARCH_205;
361 /* Determine lowest PCR bit needed to run guest in given PVR level */
362 guest_pcr_bit = host_pcr_bit;
364 switch (arch_compat) {
366 guest_pcr_bit = PCR_ARCH_205;
370 guest_pcr_bit = PCR_ARCH_206;
373 guest_pcr_bit = PCR_ARCH_207;
376 guest_pcr_bit = PCR_ARCH_300;
379 guest_pcr_bit = PCR_ARCH_31;
386 /* Check requested PCR bits don't exceed our capabilities */
387 if (guest_pcr_bit > host_pcr_bit)
390 spin_lock(&vc->lock);
391 vc->arch_compat = arch_compat;
393 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
394 * Also set all reserved PCR bits
396 vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
397 spin_unlock(&vc->lock);
402 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
406 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
407 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
408 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
409 for (r = 0; r < 16; ++r)
410 pr_err("r%2d = %.16lx r%d = %.16lx\n",
411 r, kvmppc_get_gpr(vcpu, r),
412 r+16, kvmppc_get_gpr(vcpu, r+16));
413 pr_err("ctr = %.16lx lr = %.16lx\n",
414 vcpu->arch.regs.ctr, vcpu->arch.regs.link);
415 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
416 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
417 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
418 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
419 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
420 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
421 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n",
422 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
423 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
424 pr_err("fault dar = %.16lx dsisr = %.8x\n",
425 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
426 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
427 for (r = 0; r < vcpu->arch.slb_max; ++r)
428 pr_err(" ESID = %.16llx VSID = %.16llx\n",
429 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
430 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
431 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
432 vcpu->arch.last_inst);
435 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
437 return kvm_get_vcpu_by_id(kvm, id);
440 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
442 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
443 vpa->yield_count = cpu_to_be32(1);
446 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
447 unsigned long addr, unsigned long len)
449 /* check address is cacheline aligned */
450 if (addr & (L1_CACHE_BYTES - 1))
452 spin_lock(&vcpu->arch.vpa_update_lock);
453 if (v->next_gpa != addr || v->len != len) {
455 v->len = addr ? len : 0;
456 v->update_pending = 1;
458 spin_unlock(&vcpu->arch.vpa_update_lock);
462 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
471 static int vpa_is_registered(struct kvmppc_vpa *vpap)
473 if (vpap->update_pending)
474 return vpap->next_gpa != 0;
475 return vpap->pinned_addr != NULL;
478 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
480 unsigned long vcpuid, unsigned long vpa)
482 struct kvm *kvm = vcpu->kvm;
483 unsigned long len, nb;
485 struct kvm_vcpu *tvcpu;
488 struct kvmppc_vpa *vpap;
490 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
494 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
495 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
496 subfunc == H_VPA_REG_SLB) {
497 /* Registering new area - address must be cache-line aligned */
498 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
501 /* convert logical addr to kernel addr and read length */
502 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
505 if (subfunc == H_VPA_REG_VPA)
506 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
508 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
509 kvmppc_unpin_guest_page(kvm, va, vpa, false);
512 if (len > nb || len < sizeof(struct reg_vpa))
521 spin_lock(&tvcpu->arch.vpa_update_lock);
524 case H_VPA_REG_VPA: /* register VPA */
526 * The size of our lppaca is 1kB because of the way we align
527 * it for the guest to avoid crossing a 4kB boundary. We only
528 * use 640 bytes of the structure though, so we should accept
529 * clients that set a size of 640.
531 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
532 if (len < sizeof(struct lppaca))
534 vpap = &tvcpu->arch.vpa;
538 case H_VPA_REG_DTL: /* register DTL */
539 if (len < sizeof(struct dtl_entry))
541 len -= len % sizeof(struct dtl_entry);
543 /* Check that they have previously registered a VPA */
545 if (!vpa_is_registered(&tvcpu->arch.vpa))
548 vpap = &tvcpu->arch.dtl;
552 case H_VPA_REG_SLB: /* register SLB shadow buffer */
553 /* Check that they have previously registered a VPA */
555 if (!vpa_is_registered(&tvcpu->arch.vpa))
558 vpap = &tvcpu->arch.slb_shadow;
562 case H_VPA_DEREG_VPA: /* deregister VPA */
563 /* Check they don't still have a DTL or SLB buf registered */
565 if (vpa_is_registered(&tvcpu->arch.dtl) ||
566 vpa_is_registered(&tvcpu->arch.slb_shadow))
569 vpap = &tvcpu->arch.vpa;
573 case H_VPA_DEREG_DTL: /* deregister DTL */
574 vpap = &tvcpu->arch.dtl;
578 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
579 vpap = &tvcpu->arch.slb_shadow;
585 vpap->next_gpa = vpa;
587 vpap->update_pending = 1;
590 spin_unlock(&tvcpu->arch.vpa_update_lock);
595 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
597 struct kvm *kvm = vcpu->kvm;
603 * We need to pin the page pointed to by vpap->next_gpa,
604 * but we can't call kvmppc_pin_guest_page under the lock
605 * as it does get_user_pages() and down_read(). So we
606 * have to drop the lock, pin the page, then get the lock
607 * again and check that a new area didn't get registered
611 gpa = vpap->next_gpa;
612 spin_unlock(&vcpu->arch.vpa_update_lock);
616 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
617 spin_lock(&vcpu->arch.vpa_update_lock);
618 if (gpa == vpap->next_gpa)
620 /* sigh... unpin that one and try again */
622 kvmppc_unpin_guest_page(kvm, va, gpa, false);
625 vpap->update_pending = 0;
626 if (va && nb < vpap->len) {
628 * If it's now too short, it must be that userspace
629 * has changed the mappings underlying guest memory,
630 * so unregister the region.
632 kvmppc_unpin_guest_page(kvm, va, gpa, false);
635 if (vpap->pinned_addr)
636 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
639 vpap->pinned_addr = va;
642 vpap->pinned_end = va + vpap->len;
645 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
647 if (!(vcpu->arch.vpa.update_pending ||
648 vcpu->arch.slb_shadow.update_pending ||
649 vcpu->arch.dtl.update_pending))
652 spin_lock(&vcpu->arch.vpa_update_lock);
653 if (vcpu->arch.vpa.update_pending) {
654 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
655 if (vcpu->arch.vpa.pinned_addr)
656 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
658 if (vcpu->arch.dtl.update_pending) {
659 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
660 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
661 vcpu->arch.dtl_index = 0;
663 if (vcpu->arch.slb_shadow.update_pending)
664 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
665 spin_unlock(&vcpu->arch.vpa_update_lock);
669 * Return the accumulated stolen time for the vcore up until `now'.
670 * The caller should hold the vcore lock.
672 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
677 spin_lock_irqsave(&vc->stoltb_lock, flags);
679 if (vc->vcore_state != VCORE_INACTIVE &&
680 vc->preempt_tb != TB_NIL)
681 p += now - vc->preempt_tb;
682 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
686 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
687 struct kvmppc_vcore *vc)
689 struct dtl_entry *dt;
691 unsigned long stolen;
692 unsigned long core_stolen;
696 dt = vcpu->arch.dtl_ptr;
697 vpa = vcpu->arch.vpa.pinned_addr;
699 core_stolen = vcore_stolen_time(vc, now);
700 stolen = core_stolen - vcpu->arch.stolen_logged;
701 vcpu->arch.stolen_logged = core_stolen;
702 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
703 stolen += vcpu->arch.busy_stolen;
704 vcpu->arch.busy_stolen = 0;
705 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
708 memset(dt, 0, sizeof(struct dtl_entry));
709 dt->dispatch_reason = 7;
710 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
711 dt->timebase = cpu_to_be64(now + vc->tb_offset);
712 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
713 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
714 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
716 if (dt == vcpu->arch.dtl.pinned_end)
717 dt = vcpu->arch.dtl.pinned_addr;
718 vcpu->arch.dtl_ptr = dt;
719 /* order writing *dt vs. writing vpa->dtl_idx */
721 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
722 vcpu->arch.dtl.dirty = true;
725 /* See if there is a doorbell interrupt pending for a vcpu */
726 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
729 struct kvmppc_vcore *vc;
731 if (vcpu->arch.doorbell_request)
734 * Ensure that the read of vcore->dpdes comes after the read
735 * of vcpu->doorbell_request. This barrier matches the
736 * smp_wmb() in kvmppc_guest_entry_inject().
739 vc = vcpu->arch.vcore;
740 thr = vcpu->vcpu_id - vc->first_vcpuid;
741 return !!(vc->dpdes & (1 << thr));
744 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
746 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
748 if ((!vcpu->arch.vcore->arch_compat) &&
749 cpu_has_feature(CPU_FTR_ARCH_207S))
754 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
755 unsigned long resource, unsigned long value1,
756 unsigned long value2)
759 case H_SET_MODE_RESOURCE_SET_CIABR:
760 if (!kvmppc_power8_compatible(vcpu))
765 return H_UNSUPPORTED_FLAG_START;
766 /* Guests can't breakpoint the hypervisor */
767 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
769 vcpu->arch.ciabr = value1;
771 case H_SET_MODE_RESOURCE_SET_DAWR0:
772 if (!kvmppc_power8_compatible(vcpu))
774 if (!ppc_breakpoint_available())
777 return H_UNSUPPORTED_FLAG_START;
778 if (value2 & DABRX_HYP)
780 vcpu->arch.dawr0 = value1;
781 vcpu->arch.dawrx0 = value2;
783 case H_SET_MODE_RESOURCE_SET_DAWR1:
784 if (!kvmppc_power8_compatible(vcpu))
786 if (!ppc_breakpoint_available())
788 if (!cpu_has_feature(CPU_FTR_DAWR1))
790 if (!vcpu->kvm->arch.dawr1_enabled)
793 return H_UNSUPPORTED_FLAG_START;
794 if (value2 & DABRX_HYP)
796 vcpu->arch.dawr1 = value1;
797 vcpu->arch.dawrx1 = value2;
799 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
801 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
802 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
804 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
805 kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
806 return H_UNSUPPORTED_FLAG_START;
813 /* Copy guest memory in place - must reside within a single memslot */
814 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
817 struct kvm_memory_slot *to_memslot = NULL;
818 struct kvm_memory_slot *from_memslot = NULL;
819 unsigned long to_addr, from_addr;
822 /* Get HPA for from address */
823 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
826 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
829 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
830 if (kvm_is_error_hva(from_addr))
832 from_addr |= (from & (PAGE_SIZE - 1));
834 /* Get HPA for to address */
835 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
838 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
841 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
842 if (kvm_is_error_hva(to_addr))
844 to_addr |= (to & (PAGE_SIZE - 1));
847 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
851 mark_page_dirty(kvm, to >> PAGE_SHIFT);
855 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
856 unsigned long dest, unsigned long src)
858 u64 pg_sz = SZ_4K; /* 4K page size */
859 u64 pg_mask = SZ_4K - 1;
862 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
863 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
864 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
867 /* dest (and src if copy_page flag set) must be page aligned */
868 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
871 /* zero and/or copy the page as determined by the flags */
872 if (flags & H_COPY_PAGE) {
873 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
876 } else if (flags & H_ZERO_PAGE) {
877 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
882 /* We can ignore the remaining flags */
887 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
889 struct kvmppc_vcore *vcore = target->arch.vcore;
892 * We expect to have been called by the real mode handler
893 * (kvmppc_rm_h_confer()) which would have directly returned
894 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
895 * have useful work to do and should not confer) so we don't
898 * In the case of the P9 single vcpu per vcore case, the real
899 * mode handler is not called but no other threads are in the
903 spin_lock(&vcore->lock);
904 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
905 vcore->vcore_state != VCORE_INACTIVE &&
907 target = vcore->runner;
908 spin_unlock(&vcore->lock);
910 return kvm_vcpu_yield_to(target);
913 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
916 struct lppaca *lppaca;
918 spin_lock(&vcpu->arch.vpa_update_lock);
919 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
921 yield_count = be32_to_cpu(lppaca->yield_count);
922 spin_unlock(&vcpu->arch.vpa_update_lock);
927 * H_RPT_INVALIDATE hcall handler for nested guests.
929 * Handles only nested process-scoped invalidation requests in L0.
931 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
933 unsigned long type = kvmppc_get_gpr(vcpu, 6);
934 unsigned long pid, pg_sizes, start, end;
937 * The partition-scoped invalidations aren't handled here in L0.
939 if (type & H_RPTI_TYPE_NESTED)
942 pid = kvmppc_get_gpr(vcpu, 4);
943 pg_sizes = kvmppc_get_gpr(vcpu, 7);
944 start = kvmppc_get_gpr(vcpu, 8);
945 end = kvmppc_get_gpr(vcpu, 9);
947 do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
948 type, pg_sizes, start, end);
950 kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
954 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
955 unsigned long id, unsigned long target,
956 unsigned long type, unsigned long pg_sizes,
957 unsigned long start, unsigned long end)
959 if (!kvm_is_radix(vcpu->kvm))
960 return H_UNSUPPORTED;
966 * Partition-scoped invalidation for nested guests.
968 if (type & H_RPTI_TYPE_NESTED) {
969 if (!nesting_enabled(vcpu->kvm))
972 /* Support only cores as target */
973 if (target != H_RPTI_TARGET_CMMU)
976 return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
981 * Process-scoped invalidation for L1 guests.
983 do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
984 type, pg_sizes, start, end);
988 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
990 struct kvm *kvm = vcpu->kvm;
991 unsigned long req = kvmppc_get_gpr(vcpu, 3);
992 unsigned long target, ret = H_SUCCESS;
994 struct kvm_vcpu *tvcpu;
997 if (req <= MAX_HCALL_OPCODE &&
998 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1003 ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1004 kvmppc_get_gpr(vcpu, 5),
1005 kvmppc_get_gpr(vcpu, 6));
1006 if (ret == H_TOO_HARD)
1010 ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1011 kvmppc_get_gpr(vcpu, 5),
1012 kvmppc_get_gpr(vcpu, 6),
1013 kvmppc_get_gpr(vcpu, 7));
1014 if (ret == H_TOO_HARD)
1018 ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1019 kvmppc_get_gpr(vcpu, 5));
1020 if (ret == H_TOO_HARD)
1024 ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1025 kvmppc_get_gpr(vcpu, 5));
1026 if (ret == H_TOO_HARD)
1030 ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1031 kvmppc_get_gpr(vcpu, 5));
1032 if (ret == H_TOO_HARD)
1036 ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1037 kvmppc_get_gpr(vcpu, 5),
1038 kvmppc_get_gpr(vcpu, 6));
1039 if (ret == H_TOO_HARD)
1043 ret = kvmppc_h_bulk_remove(vcpu);
1044 if (ret == H_TOO_HARD)
1051 target = kvmppc_get_gpr(vcpu, 4);
1052 tvcpu = kvmppc_find_vcpu(kvm, target);
1057 tvcpu->arch.prodded = 1;
1059 if (tvcpu->arch.ceded)
1060 kvmppc_fast_vcpu_kick_hv(tvcpu);
1063 target = kvmppc_get_gpr(vcpu, 4);
1066 tvcpu = kvmppc_find_vcpu(kvm, target);
1071 yield_count = kvmppc_get_gpr(vcpu, 5);
1072 if (kvmppc_get_yield_count(tvcpu) != yield_count)
1074 kvm_arch_vcpu_yield_to(tvcpu);
1076 case H_REGISTER_VPA:
1077 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1078 kvmppc_get_gpr(vcpu, 5),
1079 kvmppc_get_gpr(vcpu, 6));
1082 if (list_empty(&kvm->arch.rtas_tokens))
1085 idx = srcu_read_lock(&kvm->srcu);
1086 rc = kvmppc_rtas_hcall(vcpu);
1087 srcu_read_unlock(&kvm->srcu, idx);
1094 /* Send the error out to userspace via KVM_RUN */
1096 case H_LOGICAL_CI_LOAD:
1097 ret = kvmppc_h_logical_ci_load(vcpu);
1098 if (ret == H_TOO_HARD)
1101 case H_LOGICAL_CI_STORE:
1102 ret = kvmppc_h_logical_ci_store(vcpu);
1103 if (ret == H_TOO_HARD)
1107 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1108 kvmppc_get_gpr(vcpu, 5),
1109 kvmppc_get_gpr(vcpu, 6),
1110 kvmppc_get_gpr(vcpu, 7));
1111 if (ret == H_TOO_HARD)
1120 if (kvmppc_xics_enabled(vcpu)) {
1121 if (xics_on_xive()) {
1122 ret = H_NOT_AVAILABLE;
1123 return RESUME_GUEST;
1125 ret = kvmppc_xics_hcall(vcpu, req);
1130 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1133 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1134 kvmppc_get_gpr(vcpu, 5));
1136 #ifdef CONFIG_SPAPR_TCE_IOMMU
1138 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1139 kvmppc_get_gpr(vcpu, 5));
1140 if (ret == H_TOO_HARD)
1144 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1145 kvmppc_get_gpr(vcpu, 5),
1146 kvmppc_get_gpr(vcpu, 6));
1147 if (ret == H_TOO_HARD)
1150 case H_PUT_TCE_INDIRECT:
1151 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1152 kvmppc_get_gpr(vcpu, 5),
1153 kvmppc_get_gpr(vcpu, 6),
1154 kvmppc_get_gpr(vcpu, 7));
1155 if (ret == H_TOO_HARD)
1159 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1160 kvmppc_get_gpr(vcpu, 5),
1161 kvmppc_get_gpr(vcpu, 6),
1162 kvmppc_get_gpr(vcpu, 7));
1163 if (ret == H_TOO_HARD)
1168 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1171 case H_RPT_INVALIDATE:
1172 ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1173 kvmppc_get_gpr(vcpu, 5),
1174 kvmppc_get_gpr(vcpu, 6),
1175 kvmppc_get_gpr(vcpu, 7),
1176 kvmppc_get_gpr(vcpu, 8),
1177 kvmppc_get_gpr(vcpu, 9));
1180 case H_SET_PARTITION_TABLE:
1182 if (nesting_enabled(kvm))
1183 ret = kvmhv_set_partition_table(vcpu);
1185 case H_ENTER_NESTED:
1187 if (!nesting_enabled(kvm))
1189 ret = kvmhv_enter_nested_guest(vcpu);
1190 if (ret == H_INTERRUPT) {
1191 kvmppc_set_gpr(vcpu, 3, 0);
1192 vcpu->arch.hcall_needed = 0;
1194 } else if (ret == H_TOO_HARD) {
1195 kvmppc_set_gpr(vcpu, 3, 0);
1196 vcpu->arch.hcall_needed = 0;
1200 case H_TLB_INVALIDATE:
1202 if (nesting_enabled(kvm))
1203 ret = kvmhv_do_nested_tlbie(vcpu);
1205 case H_COPY_TOFROM_GUEST:
1207 if (nesting_enabled(kvm))
1208 ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1211 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1212 kvmppc_get_gpr(vcpu, 5),
1213 kvmppc_get_gpr(vcpu, 6));
1216 ret = H_UNSUPPORTED;
1217 if (kvmppc_get_srr1(vcpu) & MSR_S)
1218 ret = kvmppc_h_svm_page_in(kvm,
1219 kvmppc_get_gpr(vcpu, 4),
1220 kvmppc_get_gpr(vcpu, 5),
1221 kvmppc_get_gpr(vcpu, 6));
1223 case H_SVM_PAGE_OUT:
1224 ret = H_UNSUPPORTED;
1225 if (kvmppc_get_srr1(vcpu) & MSR_S)
1226 ret = kvmppc_h_svm_page_out(kvm,
1227 kvmppc_get_gpr(vcpu, 4),
1228 kvmppc_get_gpr(vcpu, 5),
1229 kvmppc_get_gpr(vcpu, 6));
1231 case H_SVM_INIT_START:
1232 ret = H_UNSUPPORTED;
1233 if (kvmppc_get_srr1(vcpu) & MSR_S)
1234 ret = kvmppc_h_svm_init_start(kvm);
1236 case H_SVM_INIT_DONE:
1237 ret = H_UNSUPPORTED;
1238 if (kvmppc_get_srr1(vcpu) & MSR_S)
1239 ret = kvmppc_h_svm_init_done(kvm);
1241 case H_SVM_INIT_ABORT:
1243 * Even if that call is made by the Ultravisor, the SSR1 value
1244 * is the guest context one, with the secure bit clear as it has
1245 * not yet been secured. So we can't check it here.
1246 * Instead the kvm->arch.secure_guest flag is checked inside
1247 * kvmppc_h_svm_init_abort().
1249 ret = kvmppc_h_svm_init_abort(kvm);
1255 WARN_ON_ONCE(ret == H_TOO_HARD);
1256 kvmppc_set_gpr(vcpu, 3, ret);
1257 vcpu->arch.hcall_needed = 0;
1258 return RESUME_GUEST;
1262 * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1263 * handlers in book3s_hv_rmhandlers.S.
1265 * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1266 * that the cede logic in kvmppc_run_single_vcpu() works properly.
1268 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1270 vcpu->arch.shregs.msr |= MSR_EE;
1271 vcpu->arch.ceded = 1;
1273 if (vcpu->arch.prodded) {
1274 vcpu->arch.prodded = 0;
1276 vcpu->arch.ceded = 0;
1280 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1286 case H_REGISTER_VPA:
1288 case H_LOGICAL_CI_LOAD:
1289 case H_LOGICAL_CI_STORE:
1290 #ifdef CONFIG_KVM_XICS
1299 case H_RPT_INVALIDATE:
1303 /* See if it's in the real-mode table */
1304 return kvmppc_hcall_impl_hv_realmode(cmd);
1307 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1311 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1314 * Fetch failed, so return to guest and
1315 * try executing it again.
1317 return RESUME_GUEST;
1320 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1321 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1322 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1325 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1326 return RESUME_GUEST;
1330 static void do_nothing(void *x)
1334 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1336 int thr, cpu, pcpu, nthreads;
1338 unsigned long dpdes;
1340 nthreads = vcpu->kvm->arch.emul_smt_mode;
1342 cpu = vcpu->vcpu_id & ~(nthreads - 1);
1343 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1344 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1348 * If the vcpu is currently running on a physical cpu thread,
1349 * interrupt it in order to pull it out of the guest briefly,
1350 * which will update its vcore->dpdes value.
1352 pcpu = READ_ONCE(v->cpu);
1354 smp_call_function_single(pcpu, do_nothing, NULL, 1);
1355 if (kvmppc_doorbell_pending(v))
1362 * On POWER9, emulate doorbell-related instructions in order to
1363 * give the guest the illusion of running on a multi-threaded core.
1364 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1367 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1371 struct kvm *kvm = vcpu->kvm;
1372 struct kvm_vcpu *tvcpu;
1374 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1375 return RESUME_GUEST;
1376 if (get_op(inst) != 31)
1377 return EMULATE_FAIL;
1379 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1380 switch (get_xop(inst)) {
1381 case OP_31_XOP_MSGSNDP:
1382 arg = kvmppc_get_gpr(vcpu, rb);
1383 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1386 if (arg >= kvm->arch.emul_smt_mode)
1388 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1391 if (!tvcpu->arch.doorbell_request) {
1392 tvcpu->arch.doorbell_request = 1;
1393 kvmppc_fast_vcpu_kick_hv(tvcpu);
1396 case OP_31_XOP_MSGCLRP:
1397 arg = kvmppc_get_gpr(vcpu, rb);
1398 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1400 vcpu->arch.vcore->dpdes = 0;
1401 vcpu->arch.doorbell_request = 0;
1403 case OP_31_XOP_MFSPR:
1404 switch (get_sprn(inst)) {
1409 arg = kvmppc_read_dpdes(vcpu);
1412 return EMULATE_FAIL;
1414 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1417 return EMULATE_FAIL;
1419 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1420 return RESUME_GUEST;
1423 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1424 struct task_struct *tsk)
1426 struct kvm_run *run = vcpu->run;
1427 int r = RESUME_HOST;
1429 vcpu->stat.sum_exits++;
1432 * This can happen if an interrupt occurs in the last stages
1433 * of guest entry or the first stages of guest exit (i.e. after
1434 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1435 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1436 * That can happen due to a bug, or due to a machine check
1437 * occurring at just the wrong time.
1439 if (vcpu->arch.shregs.msr & MSR_HV) {
1440 printk(KERN_EMERG "KVM trap in HV mode!\n");
1441 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1442 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1443 vcpu->arch.shregs.msr);
1444 kvmppc_dump_regs(vcpu);
1445 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1446 run->hw.hardware_exit_reason = vcpu->arch.trap;
1449 run->exit_reason = KVM_EXIT_UNKNOWN;
1450 run->ready_for_interrupt_injection = 1;
1451 switch (vcpu->arch.trap) {
1452 /* We're good on these - the host merely wanted to get our attention */
1453 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1454 vcpu->stat.dec_exits++;
1457 case BOOK3S_INTERRUPT_EXTERNAL:
1458 case BOOK3S_INTERRUPT_H_DOORBELL:
1459 case BOOK3S_INTERRUPT_H_VIRT:
1460 vcpu->stat.ext_intr_exits++;
1463 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1464 case BOOK3S_INTERRUPT_HMI:
1465 case BOOK3S_INTERRUPT_PERFMON:
1466 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1469 case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1470 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1471 DEFAULT_RATELIMIT_BURST);
1473 * Print the MCE event to host console. Ratelimit so the guest
1474 * can't flood the host log.
1476 if (__ratelimit(&rs))
1477 machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1480 * If the guest can do FWNMI, exit to userspace so it can
1481 * deliver a FWNMI to the guest.
1482 * Otherwise we synthesize a machine check for the guest
1483 * so that it knows that the machine check occurred.
1485 if (!vcpu->kvm->arch.fwnmi_enabled) {
1486 ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1487 kvmppc_core_queue_machine_check(vcpu, flags);
1492 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1493 run->exit_reason = KVM_EXIT_NMI;
1494 run->hw.hardware_exit_reason = vcpu->arch.trap;
1495 /* Clear out the old NMI status from run->flags */
1496 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1497 /* Now set the NMI status */
1498 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1499 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1501 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1506 case BOOK3S_INTERRUPT_PROGRAM:
1510 * Normally program interrupts are delivered directly
1511 * to the guest by the hardware, but we can get here
1512 * as a result of a hypervisor emulation interrupt
1513 * (e40) getting turned into a 700 by BML RTAS.
1515 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1516 kvmppc_core_queue_program(vcpu, flags);
1520 case BOOK3S_INTERRUPT_SYSCALL:
1524 if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1526 * Guest userspace executed sc 1. This can only be
1527 * reached by the P9 path because the old path
1528 * handles this case in realmode hcall handlers.
1530 if (!kvmhv_vcpu_is_radix(vcpu)) {
1532 * A guest could be running PR KVM, so this
1533 * may be a PR KVM hcall. It must be reflected
1534 * to the guest kernel as a sc interrupt.
1536 kvmppc_core_queue_syscall(vcpu);
1539 * Radix guests can not run PR KVM or nested HV
1540 * hash guests which might run PR KVM, so this
1541 * is always a privilege fault. Send a program
1542 * check to guest kernel.
1544 kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1551 * hcall - gather args and set exit_reason. This will next be
1552 * handled by kvmppc_pseries_do_hcall which may be able to deal
1553 * with it and resume guest, or may punt to userspace.
1555 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1556 for (i = 0; i < 9; ++i)
1557 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1558 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1559 vcpu->arch.hcall_needed = 1;
1564 * We get these next two if the guest accesses a page which it thinks
1565 * it has mapped but which is not actually present, either because
1566 * it is for an emulated I/O device or because the corresonding
1567 * host page has been paged out.
1569 * Any other HDSI/HISI interrupts have been handled already for P7/8
1570 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1571 * fault handling is done here.
1573 case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1577 if (vcpu->arch.fault_dsisr == HDSISR_CANARY) {
1578 r = RESUME_GUEST; /* Just retry if it's the canary */
1582 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1584 * Radix doesn't require anything, and pre-ISAv3.0 hash
1585 * already attempted to handle this in rmhandlers. The
1586 * hash fault handling below is v3 only (it uses ASDR
1589 r = RESUME_PAGE_FAULT;
1593 if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1594 kvmppc_core_queue_data_storage(vcpu,
1595 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1600 if (!(vcpu->arch.shregs.msr & MSR_DR))
1601 vsid = vcpu->kvm->arch.vrma_slb_v;
1603 vsid = vcpu->arch.fault_gpa;
1605 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1606 vsid, vcpu->arch.fault_dsisr, true);
1609 } else if (err == -1 || err == -2) {
1610 r = RESUME_PAGE_FAULT;
1612 kvmppc_core_queue_data_storage(vcpu,
1613 vcpu->arch.fault_dar, err);
1618 case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1622 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1623 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1624 DSISR_SRR1_MATCH_64S;
1625 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1627 * Radix doesn't require anything, and pre-ISAv3.0 hash
1628 * already attempted to handle this in rmhandlers. The
1629 * hash fault handling below is v3 only (it uses ASDR
1632 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1633 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1634 r = RESUME_PAGE_FAULT;
1638 if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1639 kvmppc_core_queue_inst_storage(vcpu,
1640 vcpu->arch.fault_dsisr);
1645 if (!(vcpu->arch.shregs.msr & MSR_IR))
1646 vsid = vcpu->kvm->arch.vrma_slb_v;
1648 vsid = vcpu->arch.fault_gpa;
1650 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1651 vsid, vcpu->arch.fault_dsisr, false);
1654 } else if (err == -1) {
1655 r = RESUME_PAGE_FAULT;
1657 kvmppc_core_queue_inst_storage(vcpu, err);
1664 * This occurs if the guest executes an illegal instruction.
1665 * If the guest debug is disabled, generate a program interrupt
1666 * to the guest. If guest debug is enabled, we need to check
1667 * whether the instruction is a software breakpoint instruction.
1668 * Accordingly return to Guest or Host.
1670 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1671 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1672 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1673 swab32(vcpu->arch.emul_inst) :
1674 vcpu->arch.emul_inst;
1675 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1676 r = kvmppc_emulate_debug_inst(vcpu);
1678 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1683 * This occurs if the guest (kernel or userspace), does something that
1684 * is prohibited by HFSCR.
1685 * On POWER9, this could be a doorbell instruction that we need
1687 * Otherwise, we just generate a program interrupt to the guest.
1689 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1691 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1692 cpu_has_feature(CPU_FTR_ARCH_300))
1693 r = kvmppc_emulate_doorbell_instr(vcpu);
1694 if (r == EMULATE_FAIL) {
1695 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1700 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1701 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1703 * This occurs for various TM-related instructions that
1704 * we need to emulate on POWER9 DD2.2. We have already
1705 * handled the cases where the guest was in real-suspend
1706 * mode and was transitioning to transactional state.
1708 r = kvmhv_p9_tm_emulation(vcpu);
1712 case BOOK3S_INTERRUPT_HV_RM_HARD:
1713 r = RESUME_PASSTHROUGH;
1716 kvmppc_dump_regs(vcpu);
1717 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1718 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1719 vcpu->arch.shregs.msr);
1720 run->hw.hardware_exit_reason = vcpu->arch.trap;
1728 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1733 vcpu->stat.sum_exits++;
1736 * This can happen if an interrupt occurs in the last stages
1737 * of guest entry or the first stages of guest exit (i.e. after
1738 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1739 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1740 * That can happen due to a bug, or due to a machine check
1741 * occurring at just the wrong time.
1743 if (vcpu->arch.shregs.msr & MSR_HV) {
1744 pr_emerg("KVM trap in HV mode while nested!\n");
1745 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1746 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1747 vcpu->arch.shregs.msr);
1748 kvmppc_dump_regs(vcpu);
1751 switch (vcpu->arch.trap) {
1752 /* We're good on these - the host merely wanted to get our attention */
1753 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1754 vcpu->stat.dec_exits++;
1757 case BOOK3S_INTERRUPT_EXTERNAL:
1758 vcpu->stat.ext_intr_exits++;
1761 case BOOK3S_INTERRUPT_H_DOORBELL:
1762 case BOOK3S_INTERRUPT_H_VIRT:
1763 vcpu->stat.ext_intr_exits++;
1766 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1767 case BOOK3S_INTERRUPT_HMI:
1768 case BOOK3S_INTERRUPT_PERFMON:
1769 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1772 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1774 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1775 DEFAULT_RATELIMIT_BURST);
1776 /* Pass the machine check to the L1 guest */
1778 /* Print the MCE event to host console. */
1779 if (__ratelimit(&rs))
1780 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1784 * We get these next two if the guest accesses a page which it thinks
1785 * it has mapped but which is not actually present, either because
1786 * it is for an emulated I/O device or because the corresonding
1787 * host page has been paged out.
1789 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1790 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1791 r = kvmhv_nested_page_fault(vcpu);
1792 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1794 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1795 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1796 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1797 DSISR_SRR1_MATCH_64S;
1798 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1799 vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1800 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1801 r = kvmhv_nested_page_fault(vcpu);
1802 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1805 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1806 case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1808 * This occurs for various TM-related instructions that
1809 * we need to emulate on POWER9 DD2.2. We have already
1810 * handled the cases where the guest was in real-suspend
1811 * mode and was transitioning to transactional state.
1813 r = kvmhv_p9_tm_emulation(vcpu);
1817 case BOOK3S_INTERRUPT_HV_RM_HARD:
1818 vcpu->arch.trap = 0;
1820 if (!xics_on_xive())
1821 kvmppc_xics_rm_complete(vcpu, 0);
1823 case BOOK3S_INTERRUPT_SYSCALL:
1825 unsigned long req = kvmppc_get_gpr(vcpu, 3);
1828 * The H_RPT_INVALIDATE hcalls issued by nested
1829 * guests for process-scoped invalidations when
1830 * GTSE=0, are handled here in L0.
1832 if (req == H_RPT_INVALIDATE) {
1833 r = kvmppc_nested_h_rpt_invalidate(vcpu);
1848 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1849 struct kvm_sregs *sregs)
1853 memset(sregs, 0, sizeof(struct kvm_sregs));
1854 sregs->pvr = vcpu->arch.pvr;
1855 for (i = 0; i < vcpu->arch.slb_max; i++) {
1856 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1857 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1863 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1864 struct kvm_sregs *sregs)
1868 /* Only accept the same PVR as the host's, since we can't spoof it */
1869 if (sregs->pvr != vcpu->arch.pvr)
1873 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1874 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1875 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1876 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1880 vcpu->arch.slb_max = j;
1886 * Enforce limits on guest LPCR values based on hardware availability,
1887 * guest configuration, and possibly hypervisor support and security
1890 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
1892 /* LPCR_TC only applies to HPT guests */
1893 if (kvm_is_radix(kvm))
1896 /* On POWER8 and above, userspace can modify AIL */
1897 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
1899 if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
1900 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
1902 * On some POWER9s we force AIL off for radix guests to prevent
1903 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
1904 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
1905 * be cached, which the host TLB management does not expect.
1907 if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
1911 * On POWER9, allow userspace to enable large decrementer for the
1912 * guest, whether or not the host has it enabled.
1914 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1920 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
1922 if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
1923 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
1924 lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
1928 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1929 bool preserve_top32)
1931 struct kvm *kvm = vcpu->kvm;
1932 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1935 spin_lock(&vc->lock);
1938 * Userspace can only modify
1939 * DPFD (default prefetch depth), ILE (interrupt little-endian),
1940 * TC (translation control), AIL (alternate interrupt location),
1941 * LD (large decrementer).
1942 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
1944 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
1946 /* Broken 32-bit version of LPCR must not clear top bits */
1950 new_lpcr = kvmppc_filter_lpcr_hv(kvm,
1951 (vc->lpcr & ~mask) | (new_lpcr & mask));
1954 * If ILE (interrupt little-endian) has changed, update the
1955 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1957 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1958 struct kvm_vcpu *vcpu;
1961 kvm_for_each_vcpu(i, vcpu, kvm) {
1962 if (vcpu->arch.vcore != vc)
1964 if (new_lpcr & LPCR_ILE)
1965 vcpu->arch.intr_msr |= MSR_LE;
1967 vcpu->arch.intr_msr &= ~MSR_LE;
1971 vc->lpcr = new_lpcr;
1973 spin_unlock(&vc->lock);
1976 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1977 union kvmppc_one_reg *val)
1983 case KVM_REG_PPC_DEBUG_INST:
1984 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1986 case KVM_REG_PPC_HIOR:
1987 *val = get_reg_val(id, 0);
1989 case KVM_REG_PPC_DABR:
1990 *val = get_reg_val(id, vcpu->arch.dabr);
1992 case KVM_REG_PPC_DABRX:
1993 *val = get_reg_val(id, vcpu->arch.dabrx);
1995 case KVM_REG_PPC_DSCR:
1996 *val = get_reg_val(id, vcpu->arch.dscr);
1998 case KVM_REG_PPC_PURR:
1999 *val = get_reg_val(id, vcpu->arch.purr);
2001 case KVM_REG_PPC_SPURR:
2002 *val = get_reg_val(id, vcpu->arch.spurr);
2004 case KVM_REG_PPC_AMR:
2005 *val = get_reg_val(id, vcpu->arch.amr);
2007 case KVM_REG_PPC_UAMOR:
2008 *val = get_reg_val(id, vcpu->arch.uamor);
2010 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2011 i = id - KVM_REG_PPC_MMCR0;
2012 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
2014 case KVM_REG_PPC_MMCR2:
2015 *val = get_reg_val(id, vcpu->arch.mmcr[2]);
2017 case KVM_REG_PPC_MMCRA:
2018 *val = get_reg_val(id, vcpu->arch.mmcra);
2020 case KVM_REG_PPC_MMCRS:
2021 *val = get_reg_val(id, vcpu->arch.mmcrs);
2023 case KVM_REG_PPC_MMCR3:
2024 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2026 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2027 i = id - KVM_REG_PPC_PMC1;
2028 *val = get_reg_val(id, vcpu->arch.pmc[i]);
2030 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2031 i = id - KVM_REG_PPC_SPMC1;
2032 *val = get_reg_val(id, vcpu->arch.spmc[i]);
2034 case KVM_REG_PPC_SIAR:
2035 *val = get_reg_val(id, vcpu->arch.siar);
2037 case KVM_REG_PPC_SDAR:
2038 *val = get_reg_val(id, vcpu->arch.sdar);
2040 case KVM_REG_PPC_SIER:
2041 *val = get_reg_val(id, vcpu->arch.sier[0]);
2043 case KVM_REG_PPC_SIER2:
2044 *val = get_reg_val(id, vcpu->arch.sier[1]);
2046 case KVM_REG_PPC_SIER3:
2047 *val = get_reg_val(id, vcpu->arch.sier[2]);
2049 case KVM_REG_PPC_IAMR:
2050 *val = get_reg_val(id, vcpu->arch.iamr);
2052 case KVM_REG_PPC_PSPB:
2053 *val = get_reg_val(id, vcpu->arch.pspb);
2055 case KVM_REG_PPC_DPDES:
2057 * On POWER9, where we are emulating msgsndp etc.,
2058 * we return 1 bit for each vcpu, which can come from
2059 * either vcore->dpdes or doorbell_request.
2060 * On POWER8, doorbell_request is 0.
2062 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
2063 vcpu->arch.doorbell_request);
2065 case KVM_REG_PPC_VTB:
2066 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
2068 case KVM_REG_PPC_DAWR:
2069 *val = get_reg_val(id, vcpu->arch.dawr0);
2071 case KVM_REG_PPC_DAWRX:
2072 *val = get_reg_val(id, vcpu->arch.dawrx0);
2074 case KVM_REG_PPC_DAWR1:
2075 *val = get_reg_val(id, vcpu->arch.dawr1);
2077 case KVM_REG_PPC_DAWRX1:
2078 *val = get_reg_val(id, vcpu->arch.dawrx1);
2080 case KVM_REG_PPC_CIABR:
2081 *val = get_reg_val(id, vcpu->arch.ciabr);
2083 case KVM_REG_PPC_CSIGR:
2084 *val = get_reg_val(id, vcpu->arch.csigr);
2086 case KVM_REG_PPC_TACR:
2087 *val = get_reg_val(id, vcpu->arch.tacr);
2089 case KVM_REG_PPC_TCSCR:
2090 *val = get_reg_val(id, vcpu->arch.tcscr);
2092 case KVM_REG_PPC_PID:
2093 *val = get_reg_val(id, vcpu->arch.pid);
2095 case KVM_REG_PPC_ACOP:
2096 *val = get_reg_val(id, vcpu->arch.acop);
2098 case KVM_REG_PPC_WORT:
2099 *val = get_reg_val(id, vcpu->arch.wort);
2101 case KVM_REG_PPC_TIDR:
2102 *val = get_reg_val(id, vcpu->arch.tid);
2104 case KVM_REG_PPC_PSSCR:
2105 *val = get_reg_val(id, vcpu->arch.psscr);
2107 case KVM_REG_PPC_VPA_ADDR:
2108 spin_lock(&vcpu->arch.vpa_update_lock);
2109 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2110 spin_unlock(&vcpu->arch.vpa_update_lock);
2112 case KVM_REG_PPC_VPA_SLB:
2113 spin_lock(&vcpu->arch.vpa_update_lock);
2114 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2115 val->vpaval.length = vcpu->arch.slb_shadow.len;
2116 spin_unlock(&vcpu->arch.vpa_update_lock);
2118 case KVM_REG_PPC_VPA_DTL:
2119 spin_lock(&vcpu->arch.vpa_update_lock);
2120 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2121 val->vpaval.length = vcpu->arch.dtl.len;
2122 spin_unlock(&vcpu->arch.vpa_update_lock);
2124 case KVM_REG_PPC_TB_OFFSET:
2125 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
2127 case KVM_REG_PPC_LPCR:
2128 case KVM_REG_PPC_LPCR_64:
2129 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
2131 case KVM_REG_PPC_PPR:
2132 *val = get_reg_val(id, vcpu->arch.ppr);
2134 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2135 case KVM_REG_PPC_TFHAR:
2136 *val = get_reg_val(id, vcpu->arch.tfhar);
2138 case KVM_REG_PPC_TFIAR:
2139 *val = get_reg_val(id, vcpu->arch.tfiar);
2141 case KVM_REG_PPC_TEXASR:
2142 *val = get_reg_val(id, vcpu->arch.texasr);
2144 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2145 i = id - KVM_REG_PPC_TM_GPR0;
2146 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2148 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2151 i = id - KVM_REG_PPC_TM_VSR0;
2153 for (j = 0; j < TS_FPRWIDTH; j++)
2154 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2156 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2157 val->vval = vcpu->arch.vr_tm.vr[i-32];
2163 case KVM_REG_PPC_TM_CR:
2164 *val = get_reg_val(id, vcpu->arch.cr_tm);
2166 case KVM_REG_PPC_TM_XER:
2167 *val = get_reg_val(id, vcpu->arch.xer_tm);
2169 case KVM_REG_PPC_TM_LR:
2170 *val = get_reg_val(id, vcpu->arch.lr_tm);
2172 case KVM_REG_PPC_TM_CTR:
2173 *val = get_reg_val(id, vcpu->arch.ctr_tm);
2175 case KVM_REG_PPC_TM_FPSCR:
2176 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2178 case KVM_REG_PPC_TM_AMR:
2179 *val = get_reg_val(id, vcpu->arch.amr_tm);
2181 case KVM_REG_PPC_TM_PPR:
2182 *val = get_reg_val(id, vcpu->arch.ppr_tm);
2184 case KVM_REG_PPC_TM_VRSAVE:
2185 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
2187 case KVM_REG_PPC_TM_VSCR:
2188 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2189 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2193 case KVM_REG_PPC_TM_DSCR:
2194 *val = get_reg_val(id, vcpu->arch.dscr_tm);
2196 case KVM_REG_PPC_TM_TAR:
2197 *val = get_reg_val(id, vcpu->arch.tar_tm);
2200 case KVM_REG_PPC_ARCH_COMPAT:
2201 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
2203 case KVM_REG_PPC_DEC_EXPIRY:
2204 *val = get_reg_val(id, vcpu->arch.dec_expires +
2205 vcpu->arch.vcore->tb_offset);
2207 case KVM_REG_PPC_ONLINE:
2208 *val = get_reg_val(id, vcpu->arch.online);
2210 case KVM_REG_PPC_PTCR:
2211 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2221 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2222 union kvmppc_one_reg *val)
2226 unsigned long addr, len;
2229 case KVM_REG_PPC_HIOR:
2230 /* Only allow this to be set to zero */
2231 if (set_reg_val(id, *val))
2234 case KVM_REG_PPC_DABR:
2235 vcpu->arch.dabr = set_reg_val(id, *val);
2237 case KVM_REG_PPC_DABRX:
2238 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2240 case KVM_REG_PPC_DSCR:
2241 vcpu->arch.dscr = set_reg_val(id, *val);
2243 case KVM_REG_PPC_PURR:
2244 vcpu->arch.purr = set_reg_val(id, *val);
2246 case KVM_REG_PPC_SPURR:
2247 vcpu->arch.spurr = set_reg_val(id, *val);
2249 case KVM_REG_PPC_AMR:
2250 vcpu->arch.amr = set_reg_val(id, *val);
2252 case KVM_REG_PPC_UAMOR:
2253 vcpu->arch.uamor = set_reg_val(id, *val);
2255 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2256 i = id - KVM_REG_PPC_MMCR0;
2257 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2259 case KVM_REG_PPC_MMCR2:
2260 vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2262 case KVM_REG_PPC_MMCRA:
2263 vcpu->arch.mmcra = set_reg_val(id, *val);
2265 case KVM_REG_PPC_MMCRS:
2266 vcpu->arch.mmcrs = set_reg_val(id, *val);
2268 case KVM_REG_PPC_MMCR3:
2269 *val = get_reg_val(id, vcpu->arch.mmcr[3]);
2271 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2272 i = id - KVM_REG_PPC_PMC1;
2273 vcpu->arch.pmc[i] = set_reg_val(id, *val);
2275 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2276 i = id - KVM_REG_PPC_SPMC1;
2277 vcpu->arch.spmc[i] = set_reg_val(id, *val);
2279 case KVM_REG_PPC_SIAR:
2280 vcpu->arch.siar = set_reg_val(id, *val);
2282 case KVM_REG_PPC_SDAR:
2283 vcpu->arch.sdar = set_reg_val(id, *val);
2285 case KVM_REG_PPC_SIER:
2286 vcpu->arch.sier[0] = set_reg_val(id, *val);
2288 case KVM_REG_PPC_SIER2:
2289 vcpu->arch.sier[1] = set_reg_val(id, *val);
2291 case KVM_REG_PPC_SIER3:
2292 vcpu->arch.sier[2] = set_reg_val(id, *val);
2294 case KVM_REG_PPC_IAMR:
2295 vcpu->arch.iamr = set_reg_val(id, *val);
2297 case KVM_REG_PPC_PSPB:
2298 vcpu->arch.pspb = set_reg_val(id, *val);
2300 case KVM_REG_PPC_DPDES:
2301 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2303 case KVM_REG_PPC_VTB:
2304 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2306 case KVM_REG_PPC_DAWR:
2307 vcpu->arch.dawr0 = set_reg_val(id, *val);
2309 case KVM_REG_PPC_DAWRX:
2310 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2312 case KVM_REG_PPC_DAWR1:
2313 vcpu->arch.dawr1 = set_reg_val(id, *val);
2315 case KVM_REG_PPC_DAWRX1:
2316 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2318 case KVM_REG_PPC_CIABR:
2319 vcpu->arch.ciabr = set_reg_val(id, *val);
2320 /* Don't allow setting breakpoints in hypervisor code */
2321 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2322 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
2324 case KVM_REG_PPC_CSIGR:
2325 vcpu->arch.csigr = set_reg_val(id, *val);
2327 case KVM_REG_PPC_TACR:
2328 vcpu->arch.tacr = set_reg_val(id, *val);
2330 case KVM_REG_PPC_TCSCR:
2331 vcpu->arch.tcscr = set_reg_val(id, *val);
2333 case KVM_REG_PPC_PID:
2334 vcpu->arch.pid = set_reg_val(id, *val);
2336 case KVM_REG_PPC_ACOP:
2337 vcpu->arch.acop = set_reg_val(id, *val);
2339 case KVM_REG_PPC_WORT:
2340 vcpu->arch.wort = set_reg_val(id, *val);
2342 case KVM_REG_PPC_TIDR:
2343 vcpu->arch.tid = set_reg_val(id, *val);
2345 case KVM_REG_PPC_PSSCR:
2346 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2348 case KVM_REG_PPC_VPA_ADDR:
2349 addr = set_reg_val(id, *val);
2351 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2352 vcpu->arch.dtl.next_gpa))
2354 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2356 case KVM_REG_PPC_VPA_SLB:
2357 addr = val->vpaval.addr;
2358 len = val->vpaval.length;
2360 if (addr && !vcpu->arch.vpa.next_gpa)
2362 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2364 case KVM_REG_PPC_VPA_DTL:
2365 addr = val->vpaval.addr;
2366 len = val->vpaval.length;
2368 if (addr && (len < sizeof(struct dtl_entry) ||
2369 !vcpu->arch.vpa.next_gpa))
2371 len -= len % sizeof(struct dtl_entry);
2372 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2374 case KVM_REG_PPC_TB_OFFSET:
2375 /* round up to multiple of 2^24 */
2376 vcpu->arch.vcore->tb_offset =
2377 ALIGN(set_reg_val(id, *val), 1UL << 24);
2379 case KVM_REG_PPC_LPCR:
2380 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2382 case KVM_REG_PPC_LPCR_64:
2383 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2385 case KVM_REG_PPC_PPR:
2386 vcpu->arch.ppr = set_reg_val(id, *val);
2388 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2389 case KVM_REG_PPC_TFHAR:
2390 vcpu->arch.tfhar = set_reg_val(id, *val);
2392 case KVM_REG_PPC_TFIAR:
2393 vcpu->arch.tfiar = set_reg_val(id, *val);
2395 case KVM_REG_PPC_TEXASR:
2396 vcpu->arch.texasr = set_reg_val(id, *val);
2398 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2399 i = id - KVM_REG_PPC_TM_GPR0;
2400 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2402 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2405 i = id - KVM_REG_PPC_TM_VSR0;
2407 for (j = 0; j < TS_FPRWIDTH; j++)
2408 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2410 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2411 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2416 case KVM_REG_PPC_TM_CR:
2417 vcpu->arch.cr_tm = set_reg_val(id, *val);
2419 case KVM_REG_PPC_TM_XER:
2420 vcpu->arch.xer_tm = set_reg_val(id, *val);
2422 case KVM_REG_PPC_TM_LR:
2423 vcpu->arch.lr_tm = set_reg_val(id, *val);
2425 case KVM_REG_PPC_TM_CTR:
2426 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2428 case KVM_REG_PPC_TM_FPSCR:
2429 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2431 case KVM_REG_PPC_TM_AMR:
2432 vcpu->arch.amr_tm = set_reg_val(id, *val);
2434 case KVM_REG_PPC_TM_PPR:
2435 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2437 case KVM_REG_PPC_TM_VRSAVE:
2438 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2440 case KVM_REG_PPC_TM_VSCR:
2441 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2442 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2446 case KVM_REG_PPC_TM_DSCR:
2447 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2449 case KVM_REG_PPC_TM_TAR:
2450 vcpu->arch.tar_tm = set_reg_val(id, *val);
2453 case KVM_REG_PPC_ARCH_COMPAT:
2454 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2456 case KVM_REG_PPC_DEC_EXPIRY:
2457 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2458 vcpu->arch.vcore->tb_offset;
2460 case KVM_REG_PPC_ONLINE:
2461 i = set_reg_val(id, *val);
2462 if (i && !vcpu->arch.online)
2463 atomic_inc(&vcpu->arch.vcore->online_count);
2464 else if (!i && vcpu->arch.online)
2465 atomic_dec(&vcpu->arch.vcore->online_count);
2466 vcpu->arch.online = i;
2468 case KVM_REG_PPC_PTCR:
2469 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2480 * On POWER9, threads are independent and can be in different partitions.
2481 * Therefore we consider each thread to be a subcore.
2482 * There is a restriction that all threads have to be in the same
2483 * MMU mode (radix or HPT), unfortunately, but since we only support
2484 * HPT guests on a HPT host so far, that isn't an impediment yet.
2486 static int threads_per_vcore(struct kvm *kvm)
2488 if (cpu_has_feature(CPU_FTR_ARCH_300))
2490 return threads_per_subcore;
2493 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2495 struct kvmppc_vcore *vcore;
2497 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2502 spin_lock_init(&vcore->lock);
2503 spin_lock_init(&vcore->stoltb_lock);
2504 rcuwait_init(&vcore->wait);
2505 vcore->preempt_tb = TB_NIL;
2506 vcore->lpcr = kvm->arch.lpcr;
2507 vcore->first_vcpuid = id;
2509 INIT_LIST_HEAD(&vcore->preempt_list);
2514 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2515 static struct debugfs_timings_element {
2519 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
2520 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
2521 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
2522 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
2523 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
2526 #define N_TIMINGS (ARRAY_SIZE(timings))
2528 struct debugfs_timings_state {
2529 struct kvm_vcpu *vcpu;
2530 unsigned int buflen;
2531 char buf[N_TIMINGS * 100];
2534 static int debugfs_timings_open(struct inode *inode, struct file *file)
2536 struct kvm_vcpu *vcpu = inode->i_private;
2537 struct debugfs_timings_state *p;
2539 p = kzalloc(sizeof(*p), GFP_KERNEL);
2543 kvm_get_kvm(vcpu->kvm);
2545 file->private_data = p;
2547 return nonseekable_open(inode, file);
2550 static int debugfs_timings_release(struct inode *inode, struct file *file)
2552 struct debugfs_timings_state *p = file->private_data;
2554 kvm_put_kvm(p->vcpu->kvm);
2559 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2560 size_t len, loff_t *ppos)
2562 struct debugfs_timings_state *p = file->private_data;
2563 struct kvm_vcpu *vcpu = p->vcpu;
2565 struct kvmhv_tb_accumulator tb;
2574 buf_end = s + sizeof(p->buf);
2575 for (i = 0; i < N_TIMINGS; ++i) {
2576 struct kvmhv_tb_accumulator *acc;
2578 acc = (struct kvmhv_tb_accumulator *)
2579 ((unsigned long)vcpu + timings[i].offset);
2581 for (loops = 0; loops < 1000; ++loops) {
2582 count = acc->seqcount;
2587 if (count == acc->seqcount) {
2595 snprintf(s, buf_end - s, "%s: stuck\n",
2598 snprintf(s, buf_end - s,
2599 "%s: %llu %llu %llu %llu\n",
2600 timings[i].name, count / 2,
2601 tb_to_ns(tb.tb_total),
2602 tb_to_ns(tb.tb_min),
2603 tb_to_ns(tb.tb_max));
2606 p->buflen = s - p->buf;
2610 if (pos >= p->buflen)
2612 if (len > p->buflen - pos)
2613 len = p->buflen - pos;
2614 n = copy_to_user(buf, p->buf + pos, len);
2624 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2625 size_t len, loff_t *ppos)
2630 static const struct file_operations debugfs_timings_ops = {
2631 .owner = THIS_MODULE,
2632 .open = debugfs_timings_open,
2633 .release = debugfs_timings_release,
2634 .read = debugfs_timings_read,
2635 .write = debugfs_timings_write,
2636 .llseek = generic_file_llseek,
2639 /* Create a debugfs directory for the vcpu */
2640 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2643 struct kvm *kvm = vcpu->kvm;
2645 snprintf(buf, sizeof(buf), "vcpu%u", id);
2646 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2647 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2648 &debugfs_timings_ops);
2651 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2652 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2655 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2657 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2661 struct kvmppc_vcore *vcore;
2668 vcpu->arch.shared = &vcpu->arch.shregs;
2669 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2671 * The shared struct is never shared on HV,
2672 * so we can always use host endianness
2674 #ifdef __BIG_ENDIAN__
2675 vcpu->arch.shared_big_endian = true;
2677 vcpu->arch.shared_big_endian = false;
2680 vcpu->arch.mmcr[0] = MMCR0_FC;
2681 vcpu->arch.ctrl = CTRL_RUNLATCH;
2682 /* default to host PVR, since we can't spoof it */
2683 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2684 spin_lock_init(&vcpu->arch.vpa_update_lock);
2685 spin_lock_init(&vcpu->arch.tbacct_lock);
2686 vcpu->arch.busy_preempt = TB_NIL;
2687 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2690 * Set the default HFSCR for the guest from the host value.
2691 * This value is only used on POWER9.
2692 * On POWER9, we want to virtualize the doorbell facility, so we
2693 * don't set the HFSCR_MSGP bit, and that causes those instructions
2694 * to trap and then we emulate them.
2696 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2697 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP | HFSCR_PREFIX;
2698 if (cpu_has_feature(CPU_FTR_HVMODE)) {
2699 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2700 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2701 vcpu->arch.hfscr |= HFSCR_TM;
2703 if (cpu_has_feature(CPU_FTR_TM_COMP))
2704 vcpu->arch.hfscr |= HFSCR_TM;
2706 kvmppc_mmu_book3s_hv_init(vcpu);
2708 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2710 init_waitqueue_head(&vcpu->arch.cpu_run);
2712 mutex_lock(&kvm->lock);
2715 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2716 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2717 pr_devel("KVM: VCPU ID too high\n");
2718 core = KVM_MAX_VCORES;
2720 BUG_ON(kvm->arch.smt_mode != 1);
2721 core = kvmppc_pack_vcpu_id(kvm, id);
2724 core = id / kvm->arch.smt_mode;
2726 if (core < KVM_MAX_VCORES) {
2727 vcore = kvm->arch.vcores[core];
2728 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2729 pr_devel("KVM: collision on id %u", id);
2731 } else if (!vcore) {
2733 * Take mmu_setup_lock for mutual exclusion
2734 * with kvmppc_update_lpcr().
2737 vcore = kvmppc_vcore_create(kvm,
2738 id & ~(kvm->arch.smt_mode - 1));
2739 mutex_lock(&kvm->arch.mmu_setup_lock);
2740 kvm->arch.vcores[core] = vcore;
2741 kvm->arch.online_vcores++;
2742 mutex_unlock(&kvm->arch.mmu_setup_lock);
2745 mutex_unlock(&kvm->lock);
2750 spin_lock(&vcore->lock);
2751 ++vcore->num_threads;
2752 spin_unlock(&vcore->lock);
2753 vcpu->arch.vcore = vcore;
2754 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2755 vcpu->arch.thread_cpu = -1;
2756 vcpu->arch.prev_cpu = -1;
2758 vcpu->arch.cpu_type = KVM_CPU_3S_64;
2759 kvmppc_sanity_check(vcpu);
2761 debugfs_vcpu_init(vcpu, id);
2766 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2767 unsigned long flags)
2774 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2776 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2778 * On POWER8 (or POWER7), the threading mode is "strict",
2779 * so we pack smt_mode vcpus per vcore.
2781 if (smt_mode > threads_per_subcore)
2785 * On POWER9, the threading mode is "loose",
2786 * so each vcpu gets its own vcore.
2791 mutex_lock(&kvm->lock);
2793 if (!kvm->arch.online_vcores) {
2794 kvm->arch.smt_mode = smt_mode;
2795 kvm->arch.emul_smt_mode = esmt;
2798 mutex_unlock(&kvm->lock);
2803 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2805 if (vpa->pinned_addr)
2806 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2810 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2812 spin_lock(&vcpu->arch.vpa_update_lock);
2813 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2814 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2815 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2816 spin_unlock(&vcpu->arch.vpa_update_lock);
2819 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2821 /* Indicate we want to get back into the guest */
2825 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2827 unsigned long dec_nsec, now;
2830 if (now > vcpu->arch.dec_expires) {
2831 /* decrementer has already gone negative */
2832 kvmppc_core_queue_dec(vcpu);
2833 kvmppc_core_prepare_to_enter(vcpu);
2836 dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2837 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2838 vcpu->arch.timer_running = 1;
2841 extern int __kvmppc_vcore_entry(void);
2843 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2844 struct kvm_vcpu *vcpu)
2848 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2850 spin_lock_irq(&vcpu->arch.tbacct_lock);
2852 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2853 vcpu->arch.stolen_logged;
2854 vcpu->arch.busy_preempt = now;
2855 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2856 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2858 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2861 static int kvmppc_grab_hwthread(int cpu)
2863 struct paca_struct *tpaca;
2864 long timeout = 10000;
2866 tpaca = paca_ptrs[cpu];
2868 /* Ensure the thread won't go into the kernel if it wakes */
2869 tpaca->kvm_hstate.kvm_vcpu = NULL;
2870 tpaca->kvm_hstate.kvm_vcore = NULL;
2871 tpaca->kvm_hstate.napping = 0;
2873 tpaca->kvm_hstate.hwthread_req = 1;
2876 * If the thread is already executing in the kernel (e.g. handling
2877 * a stray interrupt), wait for it to get back to nap mode.
2878 * The smp_mb() is to ensure that our setting of hwthread_req
2879 * is visible before we look at hwthread_state, so if this
2880 * races with the code at system_reset_pSeries and the thread
2881 * misses our setting of hwthread_req, we are sure to see its
2882 * setting of hwthread_state, and vice versa.
2885 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2886 if (--timeout <= 0) {
2887 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2895 static void kvmppc_release_hwthread(int cpu)
2897 struct paca_struct *tpaca;
2899 tpaca = paca_ptrs[cpu];
2900 tpaca->kvm_hstate.hwthread_req = 0;
2901 tpaca->kvm_hstate.kvm_vcpu = NULL;
2902 tpaca->kvm_hstate.kvm_vcore = NULL;
2903 tpaca->kvm_hstate.kvm_split_mode = NULL;
2906 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2908 struct kvm_nested_guest *nested = vcpu->arch.nested;
2909 cpumask_t *cpu_in_guest;
2912 cpu = cpu_first_tlb_thread_sibling(cpu);
2914 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2915 cpu_in_guest = &nested->cpu_in_guest;
2917 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2918 cpu_in_guest = &kvm->arch.cpu_in_guest;
2921 * Make sure setting of bit in need_tlb_flush precedes
2922 * testing of cpu_in_guest bits. The matching barrier on
2923 * the other side is the first smp_mb() in kvmppc_run_core().
2926 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
2927 i += cpu_tlb_thread_sibling_step())
2928 if (cpumask_test_cpu(i, cpu_in_guest))
2929 smp_call_function_single(i, do_nothing, NULL, 1);
2932 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2934 struct kvm_nested_guest *nested = vcpu->arch.nested;
2935 struct kvm *kvm = vcpu->kvm;
2938 if (!cpu_has_feature(CPU_FTR_HVMODE))
2942 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2944 prev_cpu = vcpu->arch.prev_cpu;
2947 * With radix, the guest can do TLB invalidations itself,
2948 * and it could choose to use the local form (tlbiel) if
2949 * it is invalidating a translation that has only ever been
2950 * used on one vcpu. However, that doesn't mean it has
2951 * only ever been used on one physical cpu, since vcpus
2952 * can move around between pcpus. To cope with this, when
2953 * a vcpu moves from one pcpu to another, we need to tell
2954 * any vcpus running on the same core as this vcpu previously
2955 * ran to flush the TLB. The TLB is shared between threads,
2956 * so we use a single bit in .need_tlb_flush for all 4 threads.
2958 if (prev_cpu != pcpu) {
2959 if (prev_cpu >= 0 &&
2960 cpu_first_tlb_thread_sibling(prev_cpu) !=
2961 cpu_first_tlb_thread_sibling(pcpu))
2962 radix_flush_cpu(kvm, prev_cpu, vcpu);
2964 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2966 vcpu->arch.prev_cpu = pcpu;
2970 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2973 struct paca_struct *tpaca;
2974 struct kvm *kvm = vc->kvm;
2978 if (vcpu->arch.timer_running) {
2979 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2980 vcpu->arch.timer_running = 0;
2982 cpu += vcpu->arch.ptid;
2983 vcpu->cpu = vc->pcpu;
2984 vcpu->arch.thread_cpu = cpu;
2985 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2987 tpaca = paca_ptrs[cpu];
2988 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2989 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2990 tpaca->kvm_hstate.fake_suspend = 0;
2991 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2993 tpaca->kvm_hstate.kvm_vcore = vc;
2994 if (cpu != smp_processor_id())
2995 kvmppc_ipi_thread(cpu);
2998 static void kvmppc_wait_for_nap(int n_threads)
3000 int cpu = smp_processor_id();
3005 for (loops = 0; loops < 1000000; ++loops) {
3007 * Check if all threads are finished.
3008 * We set the vcore pointer when starting a thread
3009 * and the thread clears it when finished, so we look
3010 * for any threads that still have a non-NULL vcore ptr.
3012 for (i = 1; i < n_threads; ++i)
3013 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3015 if (i == n_threads) {
3022 for (i = 1; i < n_threads; ++i)
3023 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3024 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3028 * Check that we are on thread 0 and that any other threads in
3029 * this core are off-line. Then grab the threads so they can't
3032 static int on_primary_thread(void)
3034 int cpu = smp_processor_id();
3037 /* Are we on a primary subcore? */
3038 if (cpu_thread_in_subcore(cpu))
3042 while (++thr < threads_per_subcore)
3043 if (cpu_online(cpu + thr))
3046 /* Grab all hw threads so they can't go into the kernel */
3047 for (thr = 1; thr < threads_per_subcore; ++thr) {
3048 if (kvmppc_grab_hwthread(cpu + thr)) {
3049 /* Couldn't grab one; let the others go */
3051 kvmppc_release_hwthread(cpu + thr);
3052 } while (--thr > 0);
3060 * A list of virtual cores for each physical CPU.
3061 * These are vcores that could run but their runner VCPU tasks are
3062 * (or may be) preempted.
3064 struct preempted_vcore_list {
3065 struct list_head list;
3069 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3071 static void init_vcore_lists(void)
3075 for_each_possible_cpu(cpu) {
3076 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3077 spin_lock_init(&lp->lock);
3078 INIT_LIST_HEAD(&lp->list);
3082 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3084 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3086 vc->vcore_state = VCORE_PREEMPT;
3087 vc->pcpu = smp_processor_id();
3088 if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3089 spin_lock(&lp->lock);
3090 list_add_tail(&vc->preempt_list, &lp->list);
3091 spin_unlock(&lp->lock);
3094 /* Start accumulating stolen time */
3095 kvmppc_core_start_stolen(vc);
3098 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3100 struct preempted_vcore_list *lp;
3102 kvmppc_core_end_stolen(vc);
3103 if (!list_empty(&vc->preempt_list)) {
3104 lp = &per_cpu(preempted_vcores, vc->pcpu);
3105 spin_lock(&lp->lock);
3106 list_del_init(&vc->preempt_list);
3107 spin_unlock(&lp->lock);
3109 vc->vcore_state = VCORE_INACTIVE;
3113 * This stores information about the virtual cores currently
3114 * assigned to a physical core.
3118 int max_subcore_threads;
3120 int subcore_threads[MAX_SUBCORES];
3121 struct kvmppc_vcore *vc[MAX_SUBCORES];
3125 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3126 * respectively in 2-way micro-threading (split-core) mode on POWER8.
3128 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3130 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3132 memset(cip, 0, sizeof(*cip));
3133 cip->n_subcores = 1;
3134 cip->max_subcore_threads = vc->num_threads;
3135 cip->total_threads = vc->num_threads;
3136 cip->subcore_threads[0] = vc->num_threads;
3140 static bool subcore_config_ok(int n_subcores, int n_threads)
3143 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3144 * split-core mode, with one thread per subcore.
3146 if (cpu_has_feature(CPU_FTR_ARCH_300))
3147 return n_subcores <= 4 && n_threads == 1;
3149 /* On POWER8, can only dynamically split if unsplit to begin with */
3150 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3152 if (n_subcores > MAX_SUBCORES)
3154 if (n_subcores > 1) {
3155 if (!(dynamic_mt_modes & 2))
3157 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3161 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3164 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3166 vc->entry_exit_map = 0;
3168 vc->napping_threads = 0;
3169 vc->conferring_threads = 0;
3170 vc->tb_offset_applied = 0;
3173 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3175 int n_threads = vc->num_threads;
3178 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3181 /* In one_vm_per_core mode, require all vcores to be from the same vm */
3182 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3185 if (n_threads < cip->max_subcore_threads)
3186 n_threads = cip->max_subcore_threads;
3187 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3189 cip->max_subcore_threads = n_threads;
3191 sub = cip->n_subcores;
3193 cip->total_threads += vc->num_threads;
3194 cip->subcore_threads[sub] = vc->num_threads;
3196 init_vcore_to_run(vc);
3197 list_del_init(&vc->preempt_list);
3203 * Work out whether it is possible to piggyback the execution of
3204 * vcore *pvc onto the execution of the other vcores described in *cip.
3206 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3209 if (cip->total_threads + pvc->num_threads > target_threads)
3212 return can_dynamic_split(pvc, cip);
3215 static void prepare_threads(struct kvmppc_vcore *vc)
3218 struct kvm_vcpu *vcpu;
3220 for_each_runnable_thread(i, vcpu, vc) {
3221 if (signal_pending(vcpu->arch.run_task))
3222 vcpu->arch.ret = -EINTR;
3223 else if (vcpu->arch.vpa.update_pending ||
3224 vcpu->arch.slb_shadow.update_pending ||
3225 vcpu->arch.dtl.update_pending)
3226 vcpu->arch.ret = RESUME_GUEST;
3229 kvmppc_remove_runnable(vc, vcpu);
3230 wake_up(&vcpu->arch.cpu_run);
3234 static void collect_piggybacks(struct core_info *cip, int target_threads)
3236 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3237 struct kvmppc_vcore *pvc, *vcnext;
3239 spin_lock(&lp->lock);
3240 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3241 if (!spin_trylock(&pvc->lock))
3243 prepare_threads(pvc);
3244 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3245 list_del_init(&pvc->preempt_list);
3246 if (pvc->runner == NULL) {
3247 pvc->vcore_state = VCORE_INACTIVE;
3248 kvmppc_core_end_stolen(pvc);
3250 spin_unlock(&pvc->lock);
3253 if (!can_piggyback(pvc, cip, target_threads)) {
3254 spin_unlock(&pvc->lock);
3257 kvmppc_core_end_stolen(pvc);
3258 pvc->vcore_state = VCORE_PIGGYBACK;
3259 if (cip->total_threads >= target_threads)
3262 spin_unlock(&lp->lock);
3265 static bool recheck_signals_and_mmu(struct core_info *cip)
3268 struct kvm_vcpu *vcpu;
3269 struct kvmppc_vcore *vc;
3271 for (sub = 0; sub < cip->n_subcores; ++sub) {
3273 if (!vc->kvm->arch.mmu_ready)
3275 for_each_runnable_thread(i, vcpu, vc)
3276 if (signal_pending(vcpu->arch.run_task))
3282 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3284 int still_running = 0, i;
3287 struct kvm_vcpu *vcpu;
3289 spin_lock(&vc->lock);
3291 for_each_runnable_thread(i, vcpu, vc) {
3293 * It's safe to unlock the vcore in the loop here, because
3294 * for_each_runnable_thread() is safe against removal of
3295 * the vcpu, and the vcore state is VCORE_EXITING here,
3296 * so any vcpus becoming runnable will have their arch.trap
3297 * set to zero and can't actually run in the guest.
3299 spin_unlock(&vc->lock);
3300 /* cancel pending dec exception if dec is positive */
3301 if (now < vcpu->arch.dec_expires &&
3302 kvmppc_core_pending_dec(vcpu))
3303 kvmppc_core_dequeue_dec(vcpu);
3305 trace_kvm_guest_exit(vcpu);
3308 if (vcpu->arch.trap)
3309 ret = kvmppc_handle_exit_hv(vcpu,
3310 vcpu->arch.run_task);
3312 vcpu->arch.ret = ret;
3313 vcpu->arch.trap = 0;
3315 spin_lock(&vc->lock);
3316 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3317 if (vcpu->arch.pending_exceptions)
3318 kvmppc_core_prepare_to_enter(vcpu);
3319 if (vcpu->arch.ceded)
3320 kvmppc_set_timer(vcpu);
3324 kvmppc_remove_runnable(vc, vcpu);
3325 wake_up(&vcpu->arch.cpu_run);
3329 if (still_running > 0) {
3330 kvmppc_vcore_preempt(vc);
3331 } else if (vc->runner) {
3332 vc->vcore_state = VCORE_PREEMPT;
3333 kvmppc_core_start_stolen(vc);
3335 vc->vcore_state = VCORE_INACTIVE;
3337 if (vc->n_runnable > 0 && vc->runner == NULL) {
3338 /* make sure there's a candidate runner awake */
3340 vcpu = next_runnable_thread(vc, &i);
3341 wake_up(&vcpu->arch.cpu_run);
3344 spin_unlock(&vc->lock);
3348 * Clear core from the list of active host cores as we are about to
3349 * enter the guest. Only do this if it is the primary thread of the
3350 * core (not if a subcore) that is entering the guest.
3352 static inline int kvmppc_clear_host_core(unsigned int cpu)
3356 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3359 * Memory barrier can be omitted here as we will do a smp_wmb()
3360 * later in kvmppc_start_thread and we need ensure that state is
3361 * visible to other CPUs only after we enter guest.
3363 core = cpu >> threads_shift;
3364 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3369 * Advertise this core as an active host core since we exited the guest
3370 * Only need to do this if it is the primary thread of the core that is
3373 static inline int kvmppc_set_host_core(unsigned int cpu)
3377 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3381 * Memory barrier can be omitted here because we do a spin_unlock
3382 * immediately after this which provides the memory barrier.
3384 core = cpu >> threads_shift;
3385 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3389 static void set_irq_happened(int trap)
3392 case BOOK3S_INTERRUPT_EXTERNAL:
3393 local_paca->irq_happened |= PACA_IRQ_EE;
3395 case BOOK3S_INTERRUPT_H_DOORBELL:
3396 local_paca->irq_happened |= PACA_IRQ_DBELL;
3398 case BOOK3S_INTERRUPT_HMI:
3399 local_paca->irq_happened |= PACA_IRQ_HMI;
3401 case BOOK3S_INTERRUPT_SYSTEM_RESET:
3402 replay_system_reset();
3408 * Run a set of guest threads on a physical core.
3409 * Called with vc->lock held.
3411 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3413 struct kvm_vcpu *vcpu;
3416 struct core_info core_info;
3417 struct kvmppc_vcore *pvc;
3418 struct kvm_split_mode split_info, *sip;
3419 int split, subcore_size, active;
3422 unsigned long cmd_bit, stat_bit;
3425 int controlled_threads;
3429 if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3433 * Remove from the list any threads that have a signal pending
3434 * or need a VPA update done
3436 prepare_threads(vc);
3438 /* if the runner is no longer runnable, let the caller pick a new one */
3439 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3445 init_vcore_to_run(vc);
3446 vc->preempt_tb = TB_NIL;
3449 * Number of threads that we will be controlling: the same as
3450 * the number of threads per subcore, except on POWER9,
3451 * where it's 1 because the threads are (mostly) independent.
3453 controlled_threads = threads_per_vcore(vc->kvm);
3456 * Make sure we are running on primary threads, and that secondary
3457 * threads are offline. Also check if the number of threads in this
3458 * guest are greater than the current system threads per guest.
3460 if ((controlled_threads > 1) &&
3461 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3462 for_each_runnable_thread(i, vcpu, vc) {
3463 vcpu->arch.ret = -EBUSY;
3464 kvmppc_remove_runnable(vc, vcpu);
3465 wake_up(&vcpu->arch.cpu_run);
3471 * See if we could run any other vcores on the physical core
3472 * along with this one.
3474 init_core_info(&core_info, vc);
3475 pcpu = smp_processor_id();
3476 target_threads = controlled_threads;
3477 if (target_smt_mode && target_smt_mode < target_threads)
3478 target_threads = target_smt_mode;
3479 if (vc->num_threads < target_threads)
3480 collect_piggybacks(&core_info, target_threads);
3483 * Hard-disable interrupts, and check resched flag and signals.
3484 * If we need to reschedule or deliver a signal, clean up
3485 * and return without going into the guest(s).
3486 * If the mmu_ready flag has been cleared, don't go into the
3487 * guest because that means a HPT resize operation is in progress.
3489 local_irq_disable();
3491 if (lazy_irq_pending() || need_resched() ||
3492 recheck_signals_and_mmu(&core_info)) {
3494 vc->vcore_state = VCORE_INACTIVE;
3495 /* Unlock all except the primary vcore */
3496 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3497 pvc = core_info.vc[sub];
3498 /* Put back on to the preempted vcores list */
3499 kvmppc_vcore_preempt(pvc);
3500 spin_unlock(&pvc->lock);
3502 for (i = 0; i < controlled_threads; ++i)
3503 kvmppc_release_hwthread(pcpu + i);
3507 kvmppc_clear_host_core(pcpu);
3509 /* Decide on micro-threading (split-core) mode */
3510 subcore_size = threads_per_subcore;
3511 cmd_bit = stat_bit = 0;
3512 split = core_info.n_subcores;
3514 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3518 memset(&split_info, 0, sizeof(split_info));
3519 for (sub = 0; sub < core_info.n_subcores; ++sub)
3520 split_info.vc[sub] = core_info.vc[sub];
3523 if (split == 2 && (dynamic_mt_modes & 2)) {
3524 cmd_bit = HID0_POWER8_1TO2LPAR;
3525 stat_bit = HID0_POWER8_2LPARMODE;
3528 cmd_bit = HID0_POWER8_1TO4LPAR;
3529 stat_bit = HID0_POWER8_4LPARMODE;
3531 subcore_size = MAX_SMT_THREADS / split;
3532 split_info.rpr = mfspr(SPRN_RPR);
3533 split_info.pmmar = mfspr(SPRN_PMMAR);
3534 split_info.ldbar = mfspr(SPRN_LDBAR);
3535 split_info.subcore_size = subcore_size;
3537 split_info.subcore_size = 1;
3540 /* order writes to split_info before kvm_split_mode pointer */
3544 for (thr = 0; thr < controlled_threads; ++thr) {
3545 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3547 paca->kvm_hstate.napping = 0;
3548 paca->kvm_hstate.kvm_split_mode = sip;
3551 /* Initiate micro-threading (split-core) on POWER8 if required */
3553 unsigned long hid0 = mfspr(SPRN_HID0);
3555 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3557 mtspr(SPRN_HID0, hid0);
3560 hid0 = mfspr(SPRN_HID0);
3561 if (hid0 & stat_bit)
3568 * On POWER8, set RWMR register.
3569 * Since it only affects PURR and SPURR, it doesn't affect
3570 * the host, so we don't save/restore the host value.
3573 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3574 int n_online = atomic_read(&vc->online_count);
3577 * Use the 8-thread value if we're doing split-core
3578 * or if the vcore's online count looks bogus.
3580 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3581 n_online >= 1 && n_online <= MAX_SMT_THREADS)
3582 rwmr_val = p8_rwmr_values[n_online];
3583 mtspr(SPRN_RWMR, rwmr_val);
3586 /* Start all the threads */
3588 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3589 thr = is_power8 ? subcore_thread_map[sub] : sub;
3592 pvc = core_info.vc[sub];
3593 pvc->pcpu = pcpu + thr;
3594 for_each_runnable_thread(i, vcpu, pvc) {
3595 kvmppc_start_thread(vcpu, pvc);
3596 kvmppc_create_dtl_entry(vcpu, pvc);
3597 trace_kvm_guest_enter(vcpu);
3598 if (!vcpu->arch.ptid)
3600 active |= 1 << (thr + vcpu->arch.ptid);
3603 * We need to start the first thread of each subcore
3604 * even if it doesn't have a vcpu.
3607 kvmppc_start_thread(NULL, pvc);
3611 * Ensure that split_info.do_nap is set after setting
3612 * the vcore pointer in the PACA of the secondaries.
3617 * When doing micro-threading, poke the inactive threads as well.
3618 * This gets them to the nap instruction after kvm_do_nap,
3619 * which reduces the time taken to unsplit later.
3622 split_info.do_nap = 1; /* ask secondaries to nap when done */
3623 for (thr = 1; thr < threads_per_subcore; ++thr)
3624 if (!(active & (1 << thr)))
3625 kvmppc_ipi_thread(pcpu + thr);
3628 vc->vcore_state = VCORE_RUNNING;
3631 trace_kvmppc_run_core(vc, 0);
3633 for (sub = 0; sub < core_info.n_subcores; ++sub)
3634 spin_unlock(&core_info.vc[sub]->lock);
3636 guest_enter_irqoff();
3638 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3640 this_cpu_disable_ftrace();
3643 * Interrupts will be enabled once we get into the guest,
3644 * so tell lockdep that we're about to enable interrupts.
3646 trace_hardirqs_on();
3648 trap = __kvmppc_vcore_entry();
3650 trace_hardirqs_off();
3652 this_cpu_enable_ftrace();
3654 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3656 set_irq_happened(trap);
3658 spin_lock(&vc->lock);
3659 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3660 vc->vcore_state = VCORE_EXITING;
3662 /* wait for secondary threads to finish writing their state to memory */
3663 kvmppc_wait_for_nap(controlled_threads);
3665 /* Return to whole-core mode if we split the core earlier */
3667 unsigned long hid0 = mfspr(SPRN_HID0);
3668 unsigned long loops = 0;
3670 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3671 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3673 mtspr(SPRN_HID0, hid0);
3676 hid0 = mfspr(SPRN_HID0);
3677 if (!(hid0 & stat_bit))
3682 split_info.do_nap = 0;
3685 kvmppc_set_host_core(pcpu);
3687 guest_exit_irqoff();
3691 /* Let secondaries go back to the offline loop */
3692 for (i = 0; i < controlled_threads; ++i) {
3693 kvmppc_release_hwthread(pcpu + i);
3694 if (sip && sip->napped[i])
3695 kvmppc_ipi_thread(pcpu + i);
3696 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3699 spin_unlock(&vc->lock);
3701 /* make sure updates to secondary vcpu structs are visible now */
3706 for (sub = 0; sub < core_info.n_subcores; ++sub) {
3707 pvc = core_info.vc[sub];
3708 post_guest_process(pvc, pvc == vc);
3711 spin_lock(&vc->lock);
3714 vc->vcore_state = VCORE_INACTIVE;
3715 trace_kvmppc_run_core(vc, 1);
3718 static void load_spr_state(struct kvm_vcpu *vcpu)
3720 mtspr(SPRN_DSCR, vcpu->arch.dscr);
3721 mtspr(SPRN_IAMR, vcpu->arch.iamr);
3722 mtspr(SPRN_PSPB, vcpu->arch.pspb);
3723 mtspr(SPRN_FSCR, vcpu->arch.fscr);
3724 mtspr(SPRN_TAR, vcpu->arch.tar);
3725 mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3726 mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3727 mtspr(SPRN_BESCR, vcpu->arch.bescr);
3728 mtspr(SPRN_WORT, vcpu->arch.wort);
3729 mtspr(SPRN_TIDR, vcpu->arch.tid);
3730 mtspr(SPRN_AMR, vcpu->arch.amr);
3731 mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3734 * DAR, DSISR, and for nested HV, SPRGs must be set with MSR[RI]
3735 * clear (or hstate set appropriately to catch those registers
3736 * being clobbered if we take a MCE or SRESET), so those are done
3740 if (!(vcpu->arch.ctrl & 1))
3741 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3744 static void store_spr_state(struct kvm_vcpu *vcpu)
3746 vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3748 vcpu->arch.iamr = mfspr(SPRN_IAMR);
3749 vcpu->arch.pspb = mfspr(SPRN_PSPB);
3750 vcpu->arch.fscr = mfspr(SPRN_FSCR);
3751 vcpu->arch.tar = mfspr(SPRN_TAR);
3752 vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3753 vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3754 vcpu->arch.bescr = mfspr(SPRN_BESCR);
3755 vcpu->arch.wort = mfspr(SPRN_WORT);
3756 vcpu->arch.tid = mfspr(SPRN_TIDR);
3757 vcpu->arch.amr = mfspr(SPRN_AMR);
3758 vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3759 vcpu->arch.dscr = mfspr(SPRN_DSCR);
3763 * Privileged (non-hypervisor) host registers to save.
3765 struct p9_host_os_sprs {
3773 static void save_p9_host_os_sprs(struct p9_host_os_sprs *host_os_sprs)
3775 host_os_sprs->dscr = mfspr(SPRN_DSCR);
3776 host_os_sprs->tidr = mfspr(SPRN_TIDR);
3777 host_os_sprs->iamr = mfspr(SPRN_IAMR);
3778 host_os_sprs->amr = mfspr(SPRN_AMR);
3779 host_os_sprs->fscr = mfspr(SPRN_FSCR);
3782 /* vcpu guest regs must already be saved */
3783 static void restore_p9_host_os_sprs(struct kvm_vcpu *vcpu,
3784 struct p9_host_os_sprs *host_os_sprs)
3786 mtspr(SPRN_PSPB, 0);
3787 mtspr(SPRN_WORT, 0);
3788 mtspr(SPRN_UAMOR, 0);
3790 mtspr(SPRN_DSCR, host_os_sprs->dscr);
3791 mtspr(SPRN_TIDR, host_os_sprs->tidr);
3792 mtspr(SPRN_IAMR, host_os_sprs->iamr);
3794 if (host_os_sprs->amr != vcpu->arch.amr)
3795 mtspr(SPRN_AMR, host_os_sprs->amr);
3797 if (host_os_sprs->fscr != vcpu->arch.fscr)
3798 mtspr(SPRN_FSCR, host_os_sprs->fscr);
3800 /* Save guest CTRL register, set runlatch to 1 */
3801 if (!(vcpu->arch.ctrl & 1))
3802 mtspr(SPRN_CTRLT, 1);
3805 static inline bool hcall_is_xics(unsigned long req)
3807 return req == H_EOI || req == H_CPPR || req == H_IPI ||
3808 req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
3812 * Guest entry for POWER9 and later CPUs.
3814 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3817 struct kvmppc_vcore *vc = vcpu->arch.vcore;
3818 struct p9_host_os_sprs host_os_sprs;
3823 WARN_ON_ONCE(vcpu->arch.ceded);
3825 dec = mfspr(SPRN_DEC);
3828 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3829 local_paca->kvm_hstate.dec_expires = dec + tb;
3830 if (local_paca->kvm_hstate.dec_expires < time_limit)
3831 time_limit = local_paca->kvm_hstate.dec_expires;
3833 save_p9_host_os_sprs(&host_os_sprs);
3835 kvmhv_save_host_pmu(); /* saves it to PACA kvm_hstate */
3837 kvmppc_subcore_enter_guest();
3839 vc->entry_exit_map = 1;
3842 if (vcpu->arch.vpa.pinned_addr) {
3843 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3844 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3845 lp->yield_count = cpu_to_be32(yield_count);
3846 vcpu->arch.vpa.dirty = 1;
3849 if (cpu_has_feature(CPU_FTR_TM) ||
3850 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3851 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3853 kvmhv_load_guest_pmu(vcpu);
3855 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3856 load_fp_state(&vcpu->arch.fp);
3857 #ifdef CONFIG_ALTIVEC
3858 load_vr_state(&vcpu->arch.vr);
3860 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3862 load_spr_state(vcpu);
3865 * When setting DEC, we must always deal with irq_work_raise via NMI vs
3866 * setting DEC. The problem occurs right as we switch into guest mode
3867 * if a NMI hits and sets pending work and sets DEC, then that will
3868 * apply to the guest and not bring us back to the host.
3870 * irq_work_raise could check a flag (or possibly LPCR[HDICE] for
3871 * example) and set HDEC to 1? That wouldn't solve the nested hv
3872 * case which needs to abort the hcall or zero the time limit.
3874 * XXX: Another day's problem.
3876 mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3878 if (kvmhv_on_pseries()) {
3880 * We need to save and restore the guest visible part of the
3881 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3882 * doesn't do this for us. Note only required if pseries since
3883 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
3885 unsigned long host_psscr;
3886 /* call our hypervisor to load up HV regs and go */
3887 struct hv_guest_state hvregs;
3889 host_psscr = mfspr(SPRN_PSSCR_PR);
3890 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3891 kvmhv_save_hv_regs(vcpu, &hvregs);
3893 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3894 hvregs.version = HV_GUEST_STATE_VERSION;
3895 if (vcpu->arch.nested) {
3896 hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3897 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3899 hvregs.lpid = vcpu->kvm->arch.lpid;
3900 hvregs.vcpu_token = vcpu->vcpu_id;
3902 hvregs.hdec_expiry = time_limit;
3903 mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3904 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3905 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3906 __pa(&vcpu->arch.regs));
3907 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3908 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3909 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3910 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3911 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3912 mtspr(SPRN_PSSCR_PR, host_psscr);
3914 /* H_CEDE has to be handled now, not later */
3915 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3916 kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3918 kvmppc_set_gpr(vcpu, 3, 0);
3922 kvmppc_xive_push_vcpu(vcpu);
3923 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr);
3924 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3925 !(vcpu->arch.shregs.msr & MSR_PR)) {
3926 unsigned long req = kvmppc_get_gpr(vcpu, 3);
3928 /* H_CEDE has to be handled now, not later */
3929 if (req == H_CEDE) {
3931 kvmppc_xive_rearm_escalation(vcpu); /* may un-cede */
3932 kvmppc_set_gpr(vcpu, 3, 0);
3935 /* XICS hcalls must be handled before xive is pulled */
3936 } else if (hcall_is_xics(req)) {
3939 ret = kvmppc_xive_xics_hcall(vcpu, req);
3940 if (ret != H_TOO_HARD) {
3941 kvmppc_set_gpr(vcpu, 3, ret);
3946 kvmppc_xive_pull_vcpu(vcpu);
3948 if (kvm_is_radix(vcpu->kvm))
3949 vcpu->arch.slb_max = 0;
3952 dec = mfspr(SPRN_DEC);
3953 if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3956 vcpu->arch.dec_expires = dec + tb;
3958 vcpu->arch.thread_cpu = -1;
3960 store_spr_state(vcpu);
3962 restore_p9_host_os_sprs(vcpu, &host_os_sprs);
3964 msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3965 store_fp_state(&vcpu->arch.fp);
3966 #ifdef CONFIG_ALTIVEC
3967 store_vr_state(&vcpu->arch.vr);
3969 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3971 if (cpu_has_feature(CPU_FTR_TM) ||
3972 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3973 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3976 if (vcpu->arch.vpa.pinned_addr) {
3977 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3978 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3979 lp->yield_count = cpu_to_be32(yield_count);
3980 vcpu->arch.vpa.dirty = 1;
3981 save_pmu = lp->pmcregs_in_use;
3983 /* Must save pmu if this guest is capable of running nested guests */
3984 save_pmu |= nesting_enabled(vcpu->kvm);
3986 kvmhv_save_guest_pmu(vcpu, save_pmu);
3988 vc->entry_exit_map = 0x101;
3991 mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3992 /* We may have raced with new irq work */
3993 if (test_irq_work_pending())
3995 mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3997 kvmhv_load_host_pmu();
3999 kvmppc_subcore_exit_guest();
4005 * Wait for some other vcpu thread to execute us, and
4006 * wake us up when we need to handle something in the host.
4008 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4009 struct kvm_vcpu *vcpu, int wait_state)
4013 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4014 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4015 spin_unlock(&vc->lock);
4017 spin_lock(&vc->lock);
4019 finish_wait(&vcpu->arch.cpu_run, &wait);
4022 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4024 if (!halt_poll_ns_grow)
4027 vc->halt_poll_ns *= halt_poll_ns_grow;
4028 if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4029 vc->halt_poll_ns = halt_poll_ns_grow_start;
4032 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4034 if (halt_poll_ns_shrink == 0)
4035 vc->halt_poll_ns = 0;
4037 vc->halt_poll_ns /= halt_poll_ns_shrink;
4040 #ifdef CONFIG_KVM_XICS
4041 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4043 if (!xics_on_xive())
4045 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4046 vcpu->arch.xive_saved_state.cppr;
4049 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4053 #endif /* CONFIG_KVM_XICS */
4055 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4057 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4058 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4065 * Check to see if any of the runnable vcpus on the vcore have pending
4066 * exceptions or are no longer ceded
4068 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4070 struct kvm_vcpu *vcpu;
4073 for_each_runnable_thread(i, vcpu, vc) {
4074 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4082 * All the vcpus in this vcore are idle, so wait for a decrementer
4083 * or external interrupt to one of the vcpus. vc->lock is held.
4085 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4087 ktime_t cur, start_poll, start_wait;
4091 /* Poll for pending exceptions and ceded state */
4092 cur = start_poll = ktime_get();
4093 if (vc->halt_poll_ns) {
4094 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4095 ++vc->runner->stat.generic.halt_attempted_poll;
4097 vc->vcore_state = VCORE_POLLING;
4098 spin_unlock(&vc->lock);
4101 if (kvmppc_vcore_check_block(vc)) {
4106 } while (kvm_vcpu_can_poll(cur, stop));
4108 spin_lock(&vc->lock);
4109 vc->vcore_state = VCORE_INACTIVE;
4112 ++vc->runner->stat.generic.halt_successful_poll;
4117 prepare_to_rcuwait(&vc->wait);
4118 set_current_state(TASK_INTERRUPTIBLE);
4119 if (kvmppc_vcore_check_block(vc)) {
4120 finish_rcuwait(&vc->wait);
4122 /* If we polled, count this as a successful poll */
4123 if (vc->halt_poll_ns)
4124 ++vc->runner->stat.generic.halt_successful_poll;
4128 start_wait = ktime_get();
4130 vc->vcore_state = VCORE_SLEEPING;
4131 trace_kvmppc_vcore_blocked(vc, 0);
4132 spin_unlock(&vc->lock);
4134 finish_rcuwait(&vc->wait);
4135 spin_lock(&vc->lock);
4136 vc->vcore_state = VCORE_INACTIVE;
4137 trace_kvmppc_vcore_blocked(vc, 1);
4138 ++vc->runner->stat.halt_successful_wait;
4143 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4145 /* Attribute wait time */
4147 vc->runner->stat.halt_wait_ns +=
4148 ktime_to_ns(cur) - ktime_to_ns(start_wait);
4149 /* Attribute failed poll time */
4150 if (vc->halt_poll_ns)
4151 vc->runner->stat.generic.halt_poll_fail_ns +=
4152 ktime_to_ns(start_wait) -
4153 ktime_to_ns(start_poll);
4155 /* Attribute successful poll time */
4156 if (vc->halt_poll_ns)
4157 vc->runner->stat.generic.halt_poll_success_ns +=
4159 ktime_to_ns(start_poll);
4162 /* Adjust poll time */
4164 if (block_ns <= vc->halt_poll_ns)
4166 /* We slept and blocked for longer than the max halt time */
4167 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4168 shrink_halt_poll_ns(vc);
4169 /* We slept and our poll time is too small */
4170 else if (vc->halt_poll_ns < halt_poll_ns &&
4171 block_ns < halt_poll_ns)
4172 grow_halt_poll_ns(vc);
4173 if (vc->halt_poll_ns > halt_poll_ns)
4174 vc->halt_poll_ns = halt_poll_ns;
4176 vc->halt_poll_ns = 0;
4178 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4182 * This never fails for a radix guest, as none of the operations it does
4183 * for a radix guest can fail or have a way to report failure.
4185 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4188 struct kvm *kvm = vcpu->kvm;
4190 mutex_lock(&kvm->arch.mmu_setup_lock);
4191 if (!kvm->arch.mmu_ready) {
4192 if (!kvm_is_radix(kvm))
4193 r = kvmppc_hv_setup_htab_rma(vcpu);
4195 if (cpu_has_feature(CPU_FTR_ARCH_300))
4196 kvmppc_setup_partition_table(kvm);
4197 kvm->arch.mmu_ready = 1;
4200 mutex_unlock(&kvm->arch.mmu_setup_lock);
4204 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4206 struct kvm_run *run = vcpu->run;
4208 struct kvmppc_vcore *vc;
4211 trace_kvmppc_run_vcpu_enter(vcpu);
4213 run->exit_reason = 0;
4214 vcpu->arch.ret = RESUME_GUEST;
4215 vcpu->arch.trap = 0;
4216 kvmppc_update_vpas(vcpu);
4219 * Synchronize with other threads in this virtual core
4221 vc = vcpu->arch.vcore;
4222 spin_lock(&vc->lock);
4223 vcpu->arch.ceded = 0;
4224 vcpu->arch.run_task = current;
4225 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4226 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4227 vcpu->arch.busy_preempt = TB_NIL;
4228 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4232 * This happens the first time this is called for a vcpu.
4233 * If the vcore is already running, we may be able to start
4234 * this thread straight away and have it join in.
4236 if (!signal_pending(current)) {
4237 if ((vc->vcore_state == VCORE_PIGGYBACK ||
4238 vc->vcore_state == VCORE_RUNNING) &&
4239 !VCORE_IS_EXITING(vc)) {
4240 kvmppc_create_dtl_entry(vcpu, vc);
4241 kvmppc_start_thread(vcpu, vc);
4242 trace_kvm_guest_enter(vcpu);
4243 } else if (vc->vcore_state == VCORE_SLEEPING) {
4244 rcuwait_wake_up(&vc->wait);
4249 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4250 !signal_pending(current)) {
4251 /* See if the MMU is ready to go */
4252 if (!vcpu->kvm->arch.mmu_ready) {
4253 spin_unlock(&vc->lock);
4254 r = kvmhv_setup_mmu(vcpu);
4255 spin_lock(&vc->lock);
4257 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4259 hardware_entry_failure_reason = 0;
4265 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4266 kvmppc_vcore_end_preempt(vc);
4268 if (vc->vcore_state != VCORE_INACTIVE) {
4269 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4272 for_each_runnable_thread(i, v, vc) {
4273 kvmppc_core_prepare_to_enter(v);
4274 if (signal_pending(v->arch.run_task)) {
4275 kvmppc_remove_runnable(vc, v);
4276 v->stat.signal_exits++;
4277 v->run->exit_reason = KVM_EXIT_INTR;
4278 v->arch.ret = -EINTR;
4279 wake_up(&v->arch.cpu_run);
4282 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4285 for_each_runnable_thread(i, v, vc) {
4286 if (!kvmppc_vcpu_woken(v))
4287 n_ceded += v->arch.ceded;
4292 if (n_ceded == vc->n_runnable) {
4293 kvmppc_vcore_blocked(vc);
4294 } else if (need_resched()) {
4295 kvmppc_vcore_preempt(vc);
4296 /* Let something else run */
4297 cond_resched_lock(&vc->lock);
4298 if (vc->vcore_state == VCORE_PREEMPT)
4299 kvmppc_vcore_end_preempt(vc);
4301 kvmppc_run_core(vc);
4306 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4307 (vc->vcore_state == VCORE_RUNNING ||
4308 vc->vcore_state == VCORE_EXITING ||
4309 vc->vcore_state == VCORE_PIGGYBACK))
4310 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4312 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4313 kvmppc_vcore_end_preempt(vc);
4315 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4316 kvmppc_remove_runnable(vc, vcpu);
4317 vcpu->stat.signal_exits++;
4318 run->exit_reason = KVM_EXIT_INTR;
4319 vcpu->arch.ret = -EINTR;
4322 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4323 /* Wake up some vcpu to run the core */
4325 v = next_runnable_thread(vc, &i);
4326 wake_up(&v->arch.cpu_run);
4329 trace_kvmppc_run_vcpu_exit(vcpu);
4330 spin_unlock(&vc->lock);
4331 return vcpu->arch.ret;
4334 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4337 struct kvm_run *run = vcpu->run;
4340 struct kvmppc_vcore *vc;
4341 struct kvm *kvm = vcpu->kvm;
4342 struct kvm_nested_guest *nested = vcpu->arch.nested;
4344 trace_kvmppc_run_vcpu_enter(vcpu);
4346 run->exit_reason = 0;
4347 vcpu->arch.ret = RESUME_GUEST;
4348 vcpu->arch.trap = 0;
4350 vc = vcpu->arch.vcore;
4351 vcpu->arch.ceded = 0;
4352 vcpu->arch.run_task = current;
4353 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4354 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4355 vcpu->arch.busy_preempt = TB_NIL;
4356 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4357 vc->runnable_threads[0] = vcpu;
4361 /* See if the MMU is ready to go */
4362 if (!kvm->arch.mmu_ready) {
4363 r = kvmhv_setup_mmu(vcpu);
4365 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4366 run->fail_entry.hardware_entry_failure_reason = 0;
4375 kvmppc_update_vpas(vcpu);
4377 init_vcore_to_run(vc);
4378 vc->preempt_tb = TB_NIL;
4381 pcpu = smp_processor_id();
4383 if (kvm_is_radix(kvm))
4384 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4386 local_irq_disable();
4388 if (signal_pending(current))
4390 if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4394 kvmppc_core_prepare_to_enter(vcpu);
4395 if (vcpu->arch.doorbell_request) {
4398 vcpu->arch.doorbell_request = 0;
4400 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4401 &vcpu->arch.pending_exceptions))
4403 } else if (vcpu->arch.pending_exceptions ||
4404 vcpu->arch.doorbell_request ||
4405 xive_interrupt_pending(vcpu)) {
4406 vcpu->arch.ret = RESUME_HOST;
4410 kvmppc_clear_host_core(pcpu);
4412 local_paca->kvm_hstate.napping = 0;
4413 local_paca->kvm_hstate.kvm_split_mode = NULL;
4414 kvmppc_start_thread(vcpu, vc);
4415 kvmppc_create_dtl_entry(vcpu, vc);
4416 trace_kvm_guest_enter(vcpu);
4418 vc->vcore_state = VCORE_RUNNING;
4419 trace_kvmppc_run_core(vc, 0);
4421 guest_enter_irqoff();
4423 srcu_idx = srcu_read_lock(&kvm->srcu);
4425 this_cpu_disable_ftrace();
4427 /* Tell lockdep that we're about to enable interrupts */
4428 trace_hardirqs_on();
4430 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4431 vcpu->arch.trap = trap;
4433 trace_hardirqs_off();
4435 this_cpu_enable_ftrace();
4437 srcu_read_unlock(&kvm->srcu, srcu_idx);
4439 set_irq_happened(trap);
4441 kvmppc_set_host_core(pcpu);
4443 guest_exit_irqoff();
4447 cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4452 * cancel pending decrementer exception if DEC is now positive, or if
4453 * entering a nested guest in which case the decrementer is now owned
4454 * by L2 and the L1 decrementer is provided in hdec_expires
4456 if (kvmppc_core_pending_dec(vcpu) &&
4457 ((get_tb() < vcpu->arch.dec_expires) ||
4458 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4459 kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4460 kvmppc_core_dequeue_dec(vcpu);
4462 trace_kvm_guest_exit(vcpu);
4466 r = kvmppc_handle_exit_hv(vcpu, current);
4468 r = kvmppc_handle_nested_exit(vcpu);
4472 if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4473 !kvmppc_vcpu_woken(vcpu)) {
4474 kvmppc_set_timer(vcpu);
4475 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4476 if (signal_pending(current)) {
4477 vcpu->stat.signal_exits++;
4478 run->exit_reason = KVM_EXIT_INTR;
4479 vcpu->arch.ret = -EINTR;
4482 spin_lock(&vc->lock);
4483 kvmppc_vcore_blocked(vc);
4484 spin_unlock(&vc->lock);
4487 vcpu->arch.ceded = 0;
4489 vc->vcore_state = VCORE_INACTIVE;
4490 trace_kvmppc_run_core(vc, 1);
4493 kvmppc_remove_runnable(vc, vcpu);
4494 trace_kvmppc_run_vcpu_exit(vcpu);
4496 return vcpu->arch.ret;
4499 vcpu->stat.signal_exits++;
4500 run->exit_reason = KVM_EXIT_INTR;
4501 vcpu->arch.ret = -EINTR;
4508 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4510 struct kvm_run *run = vcpu->run;
4513 unsigned long ebb_regs[3] = {}; /* shut up GCC */
4514 unsigned long user_tar = 0;
4515 unsigned int user_vrsave;
4518 if (!vcpu->arch.sane) {
4519 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4524 * Don't allow entry with a suspended transaction, because
4525 * the guest entry/exit code will lose it.
4526 * If the guest has TM enabled, save away their TM-related SPRs
4527 * (they will get restored by the TM unavailable interrupt).
4529 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4530 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4531 (current->thread.regs->msr & MSR_TM)) {
4532 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4533 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4534 run->fail_entry.hardware_entry_failure_reason = 0;
4537 /* Enable TM so we can read the TM SPRs */
4538 mtmsr(mfmsr() | MSR_TM);
4539 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4540 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4541 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4542 current->thread.regs->msr &= ~MSR_TM;
4547 * Force online to 1 for the sake of old userspace which doesn't
4550 if (!vcpu->arch.online) {
4551 atomic_inc(&vcpu->arch.vcore->online_count);
4552 vcpu->arch.online = 1;
4555 kvmppc_core_prepare_to_enter(vcpu);
4557 /* No need to go into the guest when all we'll do is come back out */
4558 if (signal_pending(current)) {
4559 run->exit_reason = KVM_EXIT_INTR;
4564 atomic_inc(&kvm->arch.vcpus_running);
4565 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4568 flush_all_to_thread(current);
4570 /* Save userspace EBB and other register values */
4571 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4572 ebb_regs[0] = mfspr(SPRN_EBBHR);
4573 ebb_regs[1] = mfspr(SPRN_EBBRR);
4574 ebb_regs[2] = mfspr(SPRN_BESCR);
4575 user_tar = mfspr(SPRN_TAR);
4577 user_vrsave = mfspr(SPRN_VRSAVE);
4579 vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4580 vcpu->arch.pgdir = kvm->mm->pgd;
4581 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4584 if (cpu_has_feature(CPU_FTR_ARCH_300))
4585 r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4586 vcpu->arch.vcore->lpcr);
4588 r = kvmppc_run_vcpu(vcpu);
4590 if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4591 if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4593 * These should have been caught reflected
4594 * into the guest by now. Final sanity check:
4595 * don't allow userspace to execute hcalls in
4601 trace_kvm_hcall_enter(vcpu);
4602 r = kvmppc_pseries_do_hcall(vcpu);
4603 trace_kvm_hcall_exit(vcpu, r);
4604 kvmppc_core_prepare_to_enter(vcpu);
4605 } else if (r == RESUME_PAGE_FAULT) {
4606 srcu_idx = srcu_read_lock(&kvm->srcu);
4607 r = kvmppc_book3s_hv_page_fault(vcpu,
4608 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4609 srcu_read_unlock(&kvm->srcu, srcu_idx);
4610 } else if (r == RESUME_PASSTHROUGH) {
4611 if (WARN_ON(xics_on_xive()))
4614 r = kvmppc_xics_rm_complete(vcpu, 0);
4616 } while (is_kvmppc_resume_guest(r));
4618 /* Restore userspace EBB and other register values */
4619 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4620 mtspr(SPRN_EBBHR, ebb_regs[0]);
4621 mtspr(SPRN_EBBRR, ebb_regs[1]);
4622 mtspr(SPRN_BESCR, ebb_regs[2]);
4623 mtspr(SPRN_TAR, user_tar);
4625 mtspr(SPRN_VRSAVE, user_vrsave);
4627 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4628 atomic_dec(&kvm->arch.vcpus_running);
4632 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4633 int shift, int sllp)
4635 (*sps)->page_shift = shift;
4636 (*sps)->slb_enc = sllp;
4637 (*sps)->enc[0].page_shift = shift;
4638 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4640 * Add 16MB MPSS support (may get filtered out by userspace)
4643 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4645 (*sps)->enc[1].page_shift = 24;
4646 (*sps)->enc[1].pte_enc = penc;
4652 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4653 struct kvm_ppc_smmu_info *info)
4655 struct kvm_ppc_one_seg_page_size *sps;
4658 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4659 * POWER7 doesn't support keys for instruction accesses,
4660 * POWER8 and POWER9 do.
4662 info->data_keys = 32;
4663 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4665 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4666 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4667 info->slb_size = 32;
4669 /* We only support these sizes for now, and no muti-size segments */
4670 sps = &info->sps[0];
4671 kvmppc_add_seg_page_size(&sps, 12, 0);
4672 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4673 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4675 /* If running as a nested hypervisor, we don't support HPT guests */
4676 if (kvmhv_on_pseries())
4677 info->flags |= KVM_PPC_NO_HASH;
4683 * Get (and clear) the dirty memory log for a memory slot.
4685 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4686 struct kvm_dirty_log *log)
4688 struct kvm_memslots *slots;
4689 struct kvm_memory_slot *memslot;
4692 unsigned long *buf, *p;
4693 struct kvm_vcpu *vcpu;
4695 mutex_lock(&kvm->slots_lock);
4698 if (log->slot >= KVM_USER_MEM_SLOTS)
4701 slots = kvm_memslots(kvm);
4702 memslot = id_to_memslot(slots, log->slot);
4704 if (!memslot || !memslot->dirty_bitmap)
4708 * Use second half of bitmap area because both HPT and radix
4709 * accumulate bits in the first half.
4711 n = kvm_dirty_bitmap_bytes(memslot);
4712 buf = memslot->dirty_bitmap + n / sizeof(long);
4715 if (kvm_is_radix(kvm))
4716 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4718 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4723 * We accumulate dirty bits in the first half of the
4724 * memslot's dirty_bitmap area, for when pages are paged
4725 * out or modified by the host directly. Pick up these
4726 * bits and add them to the map.
4728 p = memslot->dirty_bitmap;
4729 for (i = 0; i < n / sizeof(long); ++i)
4730 buf[i] |= xchg(&p[i], 0);
4732 /* Harvest dirty bits from VPA and DTL updates */
4733 /* Note: we never modify the SLB shadow buffer areas */
4734 kvm_for_each_vcpu(i, vcpu, kvm) {
4735 spin_lock(&vcpu->arch.vpa_update_lock);
4736 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4737 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4738 spin_unlock(&vcpu->arch.vpa_update_lock);
4742 if (copy_to_user(log->dirty_bitmap, buf, n))
4747 mutex_unlock(&kvm->slots_lock);
4751 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4753 vfree(slot->arch.rmap);
4754 slot->arch.rmap = NULL;
4757 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4758 struct kvm_memory_slot *slot,
4759 const struct kvm_userspace_memory_region *mem,
4760 enum kvm_mr_change change)
4762 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4764 if (change == KVM_MR_CREATE) {
4765 slot->arch.rmap = vzalloc(array_size(npages,
4766 sizeof(*slot->arch.rmap)));
4767 if (!slot->arch.rmap)
4774 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4775 const struct kvm_userspace_memory_region *mem,
4776 const struct kvm_memory_slot *old,
4777 const struct kvm_memory_slot *new,
4778 enum kvm_mr_change change)
4780 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4783 * If we are making a new memslot, it might make
4784 * some address that was previously cached as emulated
4785 * MMIO be no longer emulated MMIO, so invalidate
4786 * all the caches of emulated MMIO translations.
4789 atomic64_inc(&kvm->arch.mmio_update);
4792 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4793 * have already called kvm_arch_flush_shadow_memslot() to
4794 * flush shadow mappings. For KVM_MR_CREATE we have no
4795 * previous mappings. So the only case to handle is
4796 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4798 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4799 * to get rid of any THP PTEs in the partition-scoped page tables
4800 * so we can track dirtiness at the page level; we flush when
4801 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4804 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4805 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4806 kvmppc_radix_flush_memslot(kvm, old);
4808 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4810 if (!kvm->arch.secure_guest)
4816 * @TODO kvmppc_uvmem_memslot_create() can fail and
4817 * return error. Fix this.
4819 kvmppc_uvmem_memslot_create(kvm, new);
4822 kvmppc_uvmem_memslot_delete(kvm, old);
4825 /* TODO: Handle KVM_MR_MOVE */
4831 * Update LPCR values in kvm->arch and in vcores.
4832 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4833 * of kvm->arch.lpcr update).
4835 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4840 if ((kvm->arch.lpcr & mask) == lpcr)
4843 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4845 for (i = 0; i < KVM_MAX_VCORES; ++i) {
4846 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4850 spin_lock(&vc->lock);
4851 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4852 verify_lpcr(kvm, vc->lpcr);
4853 spin_unlock(&vc->lock);
4854 if (++cores_done >= kvm->arch.online_vcores)
4859 void kvmppc_setup_partition_table(struct kvm *kvm)
4861 unsigned long dw0, dw1;
4863 if (!kvm_is_radix(kvm)) {
4864 /* PS field - page size for VRMA */
4865 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4866 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4867 /* HTABSIZE and HTABORG fields */
4868 dw0 |= kvm->arch.sdr1;
4870 /* Second dword as set by userspace */
4871 dw1 = kvm->arch.process_table;
4873 dw0 = PATB_HR | radix__get_tree_size() |
4874 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4875 dw1 = PATB_GR | kvm->arch.process_table;
4877 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4881 * Set up HPT (hashed page table) and RMA (real-mode area).
4882 * Must be called with kvm->arch.mmu_setup_lock held.
4884 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4887 struct kvm *kvm = vcpu->kvm;
4889 struct kvm_memory_slot *memslot;
4890 struct vm_area_struct *vma;
4891 unsigned long lpcr = 0, senc;
4892 unsigned long psize, porder;
4895 /* Allocate hashed page table (if not done already) and reset it */
4896 if (!kvm->arch.hpt.virt) {
4897 int order = KVM_DEFAULT_HPT_ORDER;
4898 struct kvm_hpt_info info;
4900 err = kvmppc_allocate_hpt(&info, order);
4901 /* If we get here, it means userspace didn't specify a
4902 * size explicitly. So, try successively smaller
4903 * sizes if the default failed. */
4904 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4905 err = kvmppc_allocate_hpt(&info, order);
4908 pr_err("KVM: Couldn't alloc HPT\n");
4912 kvmppc_set_hpt(kvm, &info);
4915 /* Look up the memslot for guest physical address 0 */
4916 srcu_idx = srcu_read_lock(&kvm->srcu);
4917 memslot = gfn_to_memslot(kvm, 0);
4919 /* We must have some memory at 0 by now */
4921 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4924 /* Look up the VMA for the start of this memory slot */
4925 hva = memslot->userspace_addr;
4926 mmap_read_lock(kvm->mm);
4927 vma = find_vma(kvm->mm, hva);
4928 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4931 psize = vma_kernel_pagesize(vma);
4933 mmap_read_unlock(kvm->mm);
4935 /* We can handle 4k, 64k or 16M pages in the VRMA */
4936 if (psize >= 0x1000000)
4938 else if (psize >= 0x10000)
4942 porder = __ilog2(psize);
4944 senc = slb_pgsize_encoding(psize);
4945 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4946 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4947 /* Create HPTEs in the hash page table for the VRMA */
4948 kvmppc_map_vrma(vcpu, memslot, porder);
4950 /* Update VRMASD field in the LPCR */
4951 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4952 /* the -4 is to account for senc values starting at 0x10 */
4953 lpcr = senc << (LPCR_VRMASD_SH - 4);
4954 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4957 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4961 srcu_read_unlock(&kvm->srcu, srcu_idx);
4966 mmap_read_unlock(kvm->mm);
4971 * Must be called with kvm->arch.mmu_setup_lock held and
4972 * mmu_ready = 0 and no vcpus running.
4974 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4976 if (nesting_enabled(kvm))
4977 kvmhv_release_all_nested(kvm);
4978 kvmppc_rmap_reset(kvm);
4979 kvm->arch.process_table = 0;
4980 /* Mutual exclusion with kvm_unmap_gfn_range etc. */
4981 spin_lock(&kvm->mmu_lock);
4982 kvm->arch.radix = 0;
4983 spin_unlock(&kvm->mmu_lock);
4984 kvmppc_free_radix(kvm);
4985 kvmppc_update_lpcr(kvm, LPCR_VPM1,
4986 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4991 * Must be called with kvm->arch.mmu_setup_lock held and
4992 * mmu_ready = 0 and no vcpus running.
4994 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4998 err = kvmppc_init_vm_radix(kvm);
5001 kvmppc_rmap_reset(kvm);
5002 /* Mutual exclusion with kvm_unmap_gfn_range etc. */
5003 spin_lock(&kvm->mmu_lock);
5004 kvm->arch.radix = 1;
5005 spin_unlock(&kvm->mmu_lock);
5006 kvmppc_free_hpt(&kvm->arch.hpt);
5007 kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
5008 LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
5012 #ifdef CONFIG_KVM_XICS
5014 * Allocate a per-core structure for managing state about which cores are
5015 * running in the host versus the guest and for exchanging data between
5016 * real mode KVM and CPU running in the host.
5017 * This is only done for the first VM.
5018 * The allocated structure stays even if all VMs have stopped.
5019 * It is only freed when the kvm-hv module is unloaded.
5020 * It's OK for this routine to fail, we just don't support host
5021 * core operations like redirecting H_IPI wakeups.
5023 void kvmppc_alloc_host_rm_ops(void)
5025 struct kvmppc_host_rm_ops *ops;
5026 unsigned long l_ops;
5030 /* Not the first time here ? */
5031 if (kvmppc_host_rm_ops_hv != NULL)
5034 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5038 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5039 ops->rm_core = kzalloc(size, GFP_KERNEL);
5041 if (!ops->rm_core) {
5048 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5049 if (!cpu_online(cpu))
5052 core = cpu >> threads_shift;
5053 ops->rm_core[core].rm_state.in_host = 1;
5056 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5059 * Make the contents of the kvmppc_host_rm_ops structure visible
5060 * to other CPUs before we assign it to the global variable.
5061 * Do an atomic assignment (no locks used here), but if someone
5062 * beats us to it, just free our copy and return.
5065 l_ops = (unsigned long) ops;
5067 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5069 kfree(ops->rm_core);
5074 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5075 "ppc/kvm_book3s:prepare",
5076 kvmppc_set_host_core,
5077 kvmppc_clear_host_core);
5081 void kvmppc_free_host_rm_ops(void)
5083 if (kvmppc_host_rm_ops_hv) {
5084 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5085 kfree(kvmppc_host_rm_ops_hv->rm_core);
5086 kfree(kvmppc_host_rm_ops_hv);
5087 kvmppc_host_rm_ops_hv = NULL;
5092 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5094 unsigned long lpcr, lpid;
5098 mutex_init(&kvm->arch.uvmem_lock);
5099 INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5100 mutex_init(&kvm->arch.mmu_setup_lock);
5102 /* Allocate the guest's logical partition ID */
5104 lpid = kvmppc_alloc_lpid();
5107 kvm->arch.lpid = lpid;
5109 kvmppc_alloc_host_rm_ops();
5111 kvmhv_vm_nested_init(kvm);
5114 * Since we don't flush the TLB when tearing down a VM,
5115 * and this lpid might have previously been used,
5116 * make sure we flush on each core before running the new VM.
5117 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5118 * does this flush for us.
5120 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5121 cpumask_setall(&kvm->arch.need_tlb_flush);
5123 /* Start out with the default set of hcalls enabled */
5124 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5125 sizeof(kvm->arch.enabled_hcalls));
5127 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5128 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5130 /* Init LPCR for virtual RMA mode */
5131 if (cpu_has_feature(CPU_FTR_HVMODE)) {
5132 kvm->arch.host_lpid = mfspr(SPRN_LPID);
5133 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5134 lpcr &= LPCR_PECE | LPCR_LPES;
5138 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5139 LPCR_VPM0 | LPCR_VPM1;
5140 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5141 (VRMA_VSID << SLB_VSID_SHIFT_1T);
5142 /* On POWER8 turn on online bit to enable PURR/SPURR */
5143 if (cpu_has_feature(CPU_FTR_ARCH_207S))
5146 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5147 * Set HVICE bit to enable hypervisor virtualization interrupts.
5148 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5149 * be unnecessary but better safe than sorry in case we re-enable
5150 * EE in HV mode with this LPCR still set)
5152 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5154 lpcr |= LPCR_HVICE | LPCR_HEIC;
5157 * If xive is enabled, we route 0x500 interrupts directly
5165 * If the host uses radix, the guest starts out as radix.
5167 if (radix_enabled()) {
5168 kvm->arch.radix = 1;
5169 kvm->arch.mmu_ready = 1;
5171 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5172 ret = kvmppc_init_vm_radix(kvm);
5174 kvmppc_free_lpid(kvm->arch.lpid);
5177 kvmppc_setup_partition_table(kvm);
5180 verify_lpcr(kvm, lpcr);
5181 kvm->arch.lpcr = lpcr;
5183 /* Initialization for future HPT resizes */
5184 kvm->arch.resize_hpt = NULL;
5187 * Work out how many sets the TLB has, for the use of
5188 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5190 if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5192 * P10 will flush all the congruence class with a single tlbiel
5194 kvm->arch.tlb_sets = 1;
5195 } else if (radix_enabled())
5196 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
5197 else if (cpu_has_feature(CPU_FTR_ARCH_300))
5198 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
5199 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5200 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
5202 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
5205 * Track that we now have a HV mode VM active. This blocks secondary
5206 * CPU threads from coming online.
5208 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5209 kvm_hv_vm_activated();
5212 * Initialize smt_mode depending on processor.
5213 * POWER8 and earlier have to use "strict" threading, where
5214 * all vCPUs in a vcore have to run on the same (sub)core,
5215 * whereas on POWER9 the threads can each run a different
5218 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5219 kvm->arch.smt_mode = threads_per_subcore;
5221 kvm->arch.smt_mode = 1;
5222 kvm->arch.emul_smt_mode = 1;
5225 * Create a debugfs directory for the VM
5227 snprintf(buf, sizeof(buf), "vm%d", current->pid);
5228 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
5229 kvmppc_mmu_debugfs_init(kvm);
5230 if (radix_enabled())
5231 kvmhv_radix_debugfs_init(kvm);
5236 static void kvmppc_free_vcores(struct kvm *kvm)
5240 for (i = 0; i < KVM_MAX_VCORES; ++i)
5241 kfree(kvm->arch.vcores[i]);
5242 kvm->arch.online_vcores = 0;
5245 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5247 debugfs_remove_recursive(kvm->arch.debugfs_dir);
5249 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5250 kvm_hv_vm_deactivated();
5252 kvmppc_free_vcores(kvm);
5255 if (kvm_is_radix(kvm))
5256 kvmppc_free_radix(kvm);
5258 kvmppc_free_hpt(&kvm->arch.hpt);
5260 /* Perform global invalidation and return lpid to the pool */
5261 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5262 if (nesting_enabled(kvm))
5263 kvmhv_release_all_nested(kvm);
5264 kvm->arch.process_table = 0;
5265 if (kvm->arch.secure_guest)
5266 uv_svm_terminate(kvm->arch.lpid);
5267 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5270 kvmppc_free_lpid(kvm->arch.lpid);
5272 kvmppc_free_pimap(kvm);
5275 /* We don't need to emulate any privileged instructions or dcbz */
5276 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5277 unsigned int inst, int *advance)
5279 return EMULATE_FAIL;
5282 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5285 return EMULATE_FAIL;
5288 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5291 return EMULATE_FAIL;
5294 static int kvmppc_core_check_processor_compat_hv(void)
5296 if (cpu_has_feature(CPU_FTR_HVMODE) &&
5297 cpu_has_feature(CPU_FTR_ARCH_206))
5300 /* POWER9 in radix mode is capable of being a nested hypervisor. */
5301 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5307 #ifdef CONFIG_KVM_XICS
5309 void kvmppc_free_pimap(struct kvm *kvm)
5311 kfree(kvm->arch.pimap);
5314 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5316 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5319 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5321 struct irq_desc *desc;
5322 struct kvmppc_irq_map *irq_map;
5323 struct kvmppc_passthru_irqmap *pimap;
5324 struct irq_chip *chip;
5327 if (!kvm_irq_bypass)
5330 desc = irq_to_desc(host_irq);
5334 mutex_lock(&kvm->lock);
5336 pimap = kvm->arch.pimap;
5337 if (pimap == NULL) {
5338 /* First call, allocate structure to hold IRQ map */
5339 pimap = kvmppc_alloc_pimap();
5340 if (pimap == NULL) {
5341 mutex_unlock(&kvm->lock);
5344 kvm->arch.pimap = pimap;
5348 * For now, we only support interrupts for which the EOI operation
5349 * is an OPAL call followed by a write to XIRR, since that's
5350 * what our real-mode EOI code does, or a XIVE interrupt
5352 chip = irq_data_get_irq_chip(&desc->irq_data);
5353 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5354 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5355 host_irq, guest_gsi);
5356 mutex_unlock(&kvm->lock);
5361 * See if we already have an entry for this guest IRQ number.
5362 * If it's mapped to a hardware IRQ number, that's an error,
5363 * otherwise re-use this entry.
5365 for (i = 0; i < pimap->n_mapped; i++) {
5366 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5367 if (pimap->mapped[i].r_hwirq) {
5368 mutex_unlock(&kvm->lock);
5375 if (i == KVMPPC_PIRQ_MAPPED) {
5376 mutex_unlock(&kvm->lock);
5377 return -EAGAIN; /* table is full */
5380 irq_map = &pimap->mapped[i];
5382 irq_map->v_hwirq = guest_gsi;
5383 irq_map->desc = desc;
5386 * Order the above two stores before the next to serialize with
5387 * the KVM real mode handler.
5390 irq_map->r_hwirq = desc->irq_data.hwirq;
5392 if (i == pimap->n_mapped)
5396 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5398 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5400 irq_map->r_hwirq = 0;
5402 mutex_unlock(&kvm->lock);
5407 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5409 struct irq_desc *desc;
5410 struct kvmppc_passthru_irqmap *pimap;
5413 if (!kvm_irq_bypass)
5416 desc = irq_to_desc(host_irq);
5420 mutex_lock(&kvm->lock);
5421 if (!kvm->arch.pimap)
5424 pimap = kvm->arch.pimap;
5426 for (i = 0; i < pimap->n_mapped; i++) {
5427 if (guest_gsi == pimap->mapped[i].v_hwirq)
5431 if (i == pimap->n_mapped) {
5432 mutex_unlock(&kvm->lock);
5437 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5439 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5441 /* invalidate the entry (what do do on error from the above ?) */
5442 pimap->mapped[i].r_hwirq = 0;
5445 * We don't free this structure even when the count goes to
5446 * zero. The structure is freed when we destroy the VM.
5449 mutex_unlock(&kvm->lock);
5453 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5454 struct irq_bypass_producer *prod)
5457 struct kvm_kernel_irqfd *irqfd =
5458 container_of(cons, struct kvm_kernel_irqfd, consumer);
5460 irqfd->producer = prod;
5462 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5464 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5465 prod->irq, irqfd->gsi, ret);
5470 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5471 struct irq_bypass_producer *prod)
5474 struct kvm_kernel_irqfd *irqfd =
5475 container_of(cons, struct kvm_kernel_irqfd, consumer);
5477 irqfd->producer = NULL;
5480 * When producer of consumer is unregistered, we change back to
5481 * default external interrupt handling mode - KVM real mode
5482 * will switch back to host.
5484 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5486 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5487 prod->irq, irqfd->gsi, ret);
5491 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5492 unsigned int ioctl, unsigned long arg)
5494 struct kvm *kvm __maybe_unused = filp->private_data;
5495 void __user *argp = (void __user *)arg;
5500 case KVM_PPC_ALLOCATE_HTAB: {
5503 /* If we're a nested hypervisor, we currently only support radix */
5504 if (kvmhv_on_pseries()) {
5510 if (get_user(htab_order, (u32 __user *)argp))
5512 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5519 case KVM_PPC_GET_HTAB_FD: {
5520 struct kvm_get_htab_fd ghf;
5523 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5525 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5529 case KVM_PPC_RESIZE_HPT_PREPARE: {
5530 struct kvm_ppc_resize_hpt rhpt;
5533 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5536 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5540 case KVM_PPC_RESIZE_HPT_COMMIT: {
5541 struct kvm_ppc_resize_hpt rhpt;
5544 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5547 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5559 * List of hcall numbers to enable by default.
5560 * For compatibility with old userspace, we enable by default
5561 * all hcalls that were implemented before the hcall-enabling
5562 * facility was added. Note this list should not include H_RTAS.
5564 static unsigned int default_hcall_list[] = {
5570 #ifdef CONFIG_SPAPR_TCE_IOMMU
5580 #ifdef CONFIG_KVM_XICS
5591 static void init_default_hcalls(void)
5596 for (i = 0; default_hcall_list[i]; ++i) {
5597 hcall = default_hcall_list[i];
5598 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5599 __set_bit(hcall / 4, default_enabled_hcalls);
5603 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5609 /* If not on a POWER9, reject it */
5610 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5613 /* If any unknown flags set, reject it */
5614 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5617 /* GR (guest radix) bit in process_table field must match */
5618 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5619 if (!!(cfg->process_table & PATB_GR) != radix)
5622 /* Process table size field must be reasonable, i.e. <= 24 */
5623 if ((cfg->process_table & PRTS_MASK) > 24)
5626 /* We can change a guest to/from radix now, if the host is radix */
5627 if (radix && !radix_enabled())
5630 /* If we're a nested hypervisor, we currently only support radix */
5631 if (kvmhv_on_pseries() && !radix)
5634 mutex_lock(&kvm->arch.mmu_setup_lock);
5635 if (radix != kvm_is_radix(kvm)) {
5636 if (kvm->arch.mmu_ready) {
5637 kvm->arch.mmu_ready = 0;
5638 /* order mmu_ready vs. vcpus_running */
5640 if (atomic_read(&kvm->arch.vcpus_running)) {
5641 kvm->arch.mmu_ready = 1;
5647 err = kvmppc_switch_mmu_to_radix(kvm);
5649 err = kvmppc_switch_mmu_to_hpt(kvm);
5654 kvm->arch.process_table = cfg->process_table;
5655 kvmppc_setup_partition_table(kvm);
5657 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5658 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5662 mutex_unlock(&kvm->arch.mmu_setup_lock);
5666 static int kvmhv_enable_nested(struct kvm *kvm)
5670 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5672 if (!radix_enabled())
5675 /* kvm == NULL means the caller is testing if the capability exists */
5677 kvm->arch.nested_enable = true;
5681 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5686 if (kvmhv_vcpu_is_radix(vcpu)) {
5687 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5693 /* For now quadrants are the only way to access nested guest memory */
5694 if (rc && vcpu->arch.nested)
5700 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5705 if (kvmhv_vcpu_is_radix(vcpu)) {
5706 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5712 /* For now quadrants are the only way to access nested guest memory */
5713 if (rc && vcpu->arch.nested)
5719 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5721 unpin_vpa(kvm, vpa);
5723 vpa->pinned_addr = NULL;
5725 vpa->update_pending = 0;
5729 * Enable a guest to become a secure VM, or test whether
5730 * that could be enabled.
5731 * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5732 * tested (kvm == NULL) or enabled (kvm != NULL).
5734 static int kvmhv_enable_svm(struct kvm *kvm)
5736 if (!kvmppc_uvmem_available())
5739 kvm->arch.svm_enabled = 1;
5744 * IOCTL handler to turn off secure mode of guest
5746 * - Release all device pages
5747 * - Issue ucall to terminate the guest on the UV side
5748 * - Unpin the VPA pages.
5749 * - Reinit the partition scoped page tables
5751 static int kvmhv_svm_off(struct kvm *kvm)
5753 struct kvm_vcpu *vcpu;
5759 if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5762 mutex_lock(&kvm->arch.mmu_setup_lock);
5763 mmu_was_ready = kvm->arch.mmu_ready;
5764 if (kvm->arch.mmu_ready) {
5765 kvm->arch.mmu_ready = 0;
5766 /* order mmu_ready vs. vcpus_running */
5768 if (atomic_read(&kvm->arch.vcpus_running)) {
5769 kvm->arch.mmu_ready = 1;
5775 srcu_idx = srcu_read_lock(&kvm->srcu);
5776 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5777 struct kvm_memory_slot *memslot;
5778 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5783 kvm_for_each_memslot(memslot, slots) {
5784 kvmppc_uvmem_drop_pages(memslot, kvm, true);
5785 uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5788 srcu_read_unlock(&kvm->srcu, srcu_idx);
5790 ret = uv_svm_terminate(kvm->arch.lpid);
5791 if (ret != U_SUCCESS) {
5797 * When secure guest is reset, all the guest pages are sent
5798 * to UV via UV_PAGE_IN before the non-boot vcpus get a
5799 * chance to run and unpin their VPA pages. Unpinning of all
5800 * VPA pages is done here explicitly so that VPA pages
5801 * can be migrated to the secure side.
5803 * This is required to for the secure SMP guest to reboot
5806 kvm_for_each_vcpu(i, vcpu, kvm) {
5807 spin_lock(&vcpu->arch.vpa_update_lock);
5808 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5809 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5810 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5811 spin_unlock(&vcpu->arch.vpa_update_lock);
5814 kvmppc_setup_partition_table(kvm);
5815 kvm->arch.secure_guest = 0;
5816 kvm->arch.mmu_ready = mmu_was_ready;
5818 mutex_unlock(&kvm->arch.mmu_setup_lock);
5822 static int kvmhv_enable_dawr1(struct kvm *kvm)
5824 if (!cpu_has_feature(CPU_FTR_DAWR1))
5827 /* kvm == NULL means the caller is testing if the capability exists */
5829 kvm->arch.dawr1_enabled = true;
5833 static bool kvmppc_hash_v3_possible(void)
5835 if (!cpu_has_feature(CPU_FTR_ARCH_300))
5838 if (!cpu_has_feature(CPU_FTR_HVMODE))
5842 * POWER9 chips before version 2.02 can't have some threads in
5843 * HPT mode and some in radix mode on the same core.
5845 if (radix_enabled()) {
5846 unsigned int pvr = mfspr(SPRN_PVR);
5847 if ((pvr >> 16) == PVR_POWER9 &&
5848 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5849 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5856 static struct kvmppc_ops kvm_ops_hv = {
5857 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5858 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5859 .get_one_reg = kvmppc_get_one_reg_hv,
5860 .set_one_reg = kvmppc_set_one_reg_hv,
5861 .vcpu_load = kvmppc_core_vcpu_load_hv,
5862 .vcpu_put = kvmppc_core_vcpu_put_hv,
5863 .inject_interrupt = kvmppc_inject_interrupt_hv,
5864 .set_msr = kvmppc_set_msr_hv,
5865 .vcpu_run = kvmppc_vcpu_run_hv,
5866 .vcpu_create = kvmppc_core_vcpu_create_hv,
5867 .vcpu_free = kvmppc_core_vcpu_free_hv,
5868 .check_requests = kvmppc_core_check_requests_hv,
5869 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
5870 .flush_memslot = kvmppc_core_flush_memslot_hv,
5871 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5872 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
5873 .unmap_gfn_range = kvm_unmap_gfn_range_hv,
5874 .age_gfn = kvm_age_gfn_hv,
5875 .test_age_gfn = kvm_test_age_gfn_hv,
5876 .set_spte_gfn = kvm_set_spte_gfn_hv,
5877 .free_memslot = kvmppc_core_free_memslot_hv,
5878 .init_vm = kvmppc_core_init_vm_hv,
5879 .destroy_vm = kvmppc_core_destroy_vm_hv,
5880 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5881 .emulate_op = kvmppc_core_emulate_op_hv,
5882 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5883 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5884 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5885 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
5886 .hcall_implemented = kvmppc_hcall_impl_hv,
5887 #ifdef CONFIG_KVM_XICS
5888 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5889 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5891 .configure_mmu = kvmhv_configure_mmu,
5892 .get_rmmu_info = kvmhv_get_rmmu_info,
5893 .set_smt_mode = kvmhv_set_smt_mode,
5894 .enable_nested = kvmhv_enable_nested,
5895 .load_from_eaddr = kvmhv_load_from_eaddr,
5896 .store_to_eaddr = kvmhv_store_to_eaddr,
5897 .enable_svm = kvmhv_enable_svm,
5898 .svm_off = kvmhv_svm_off,
5899 .enable_dawr1 = kvmhv_enable_dawr1,
5900 .hash_v3_possible = kvmppc_hash_v3_possible,
5903 static int kvm_init_subcore_bitmap(void)
5906 int nr_cores = cpu_nr_cores();
5907 struct sibling_subcore_state *sibling_subcore_state;
5909 for (i = 0; i < nr_cores; i++) {
5910 int first_cpu = i * threads_per_core;
5911 int node = cpu_to_node(first_cpu);
5913 /* Ignore if it is already allocated. */
5914 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5917 sibling_subcore_state =
5918 kzalloc_node(sizeof(struct sibling_subcore_state),
5920 if (!sibling_subcore_state)
5924 for (j = 0; j < threads_per_core; j++) {
5925 int cpu = first_cpu + j;
5927 paca_ptrs[cpu]->sibling_subcore_state =
5928 sibling_subcore_state;
5934 static int kvmppc_radix_possible(void)
5936 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5939 static int kvmppc_book3s_init_hv(void)
5943 if (!tlbie_capable) {
5944 pr_err("KVM-HV: Host does not support TLBIE\n");
5949 * FIXME!! Do we need to check on all cpus ?
5951 r = kvmppc_core_check_processor_compat_hv();
5955 r = kvmhv_nested_init();
5959 r = kvm_init_subcore_bitmap();
5964 * We need a way of accessing the XICS interrupt controller,
5965 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5966 * indirectly, via OPAL.
5969 if (!xics_on_xive() && !kvmhv_on_pseries() &&
5970 !local_paca->kvm_hstate.xics_phys) {
5971 struct device_node *np;
5973 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5975 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5978 /* presence of intc confirmed - node can be dropped again */
5983 kvm_ops_hv.owner = THIS_MODULE;
5984 kvmppc_hv_ops = &kvm_ops_hv;
5986 init_default_hcalls();
5990 r = kvmppc_mmu_hv_init();
5994 if (kvmppc_radix_possible())
5995 r = kvmppc_radix_init();
5997 r = kvmppc_uvmem_init();
5999 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6004 static void kvmppc_book3s_exit_hv(void)
6006 kvmppc_uvmem_free();
6007 kvmppc_free_host_rm_ops();
6008 if (kvmppc_radix_possible())
6009 kvmppc_radix_exit();
6010 kvmppc_hv_ops = NULL;
6011 kvmhv_nested_exit();
6014 module_init(kvmppc_book3s_init_hv);
6015 module_exit(kvmppc_book3s_exit_hv);
6016 MODULE_LICENSE("GPL");
6017 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6018 MODULE_ALIAS("devname:kvm");