2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched/signal.h>
26 #include <linux/sched/stat.h>
27 #include <linux/delay.h>
28 #include <linux/export.h>
30 #include <linux/anon_inodes.h>
31 #include <linux/cpu.h>
32 #include <linux/cpumask.h>
33 #include <linux/spinlock.h>
34 #include <linux/page-flags.h>
35 #include <linux/srcu.h>
36 #include <linux/miscdevice.h>
37 #include <linux/debugfs.h>
38 #include <linux/gfp.h>
39 #include <linux/vmalloc.h>
40 #include <linux/highmem.h>
41 #include <linux/hugetlb.h>
42 #include <linux/kvm_irqfd.h>
43 #include <linux/irqbypass.h>
44 #include <linux/module.h>
45 #include <linux/compiler.h>
49 #include <asm/ppc-opcode.h>
50 #include <asm/asm-prototypes.h>
51 #include <asm/disassemble.h>
52 #include <asm/cputable.h>
53 #include <asm/cacheflush.h>
54 #include <asm/tlbflush.h>
55 #include <linux/uaccess.h>
57 #include <asm/kvm_ppc.h>
58 #include <asm/kvm_book3s.h>
59 #include <asm/mmu_context.h>
60 #include <asm/lppaca.h>
61 #include <asm/processor.h>
62 #include <asm/cputhreads.h>
64 #include <asm/hvcall.h>
65 #include <asm/switch_to.h>
67 #include <asm/dbell.h>
69 #include <asm/pnv-pci.h>
77 #define CREATE_TRACE_POINTS
80 /* #define EXIT_DEBUG */
81 /* #define EXIT_DEBUG_SIMPLE */
82 /* #define EXIT_DEBUG_INT */
84 /* Used to indicate that a guest page fault needs to be handled */
85 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
86 /* Used to indicate that a guest passthrough interrupt needs to be handled */
87 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
89 /* Used as a "null" value for timebase values */
90 #define TB_NIL (~(u64)0)
92 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
94 static int dynamic_mt_modes = 6;
95 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
96 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
97 static int target_smt_mode;
98 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
99 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
101 #ifdef CONFIG_KVM_XICS
102 static struct kernel_param_ops module_param_ops = {
103 .set = param_set_int,
104 .get = param_get_int,
107 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
109 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
111 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
113 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
116 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
117 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
119 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
123 struct kvm_vcpu *vcpu;
125 while (++i < MAX_SMT_THREADS) {
126 vcpu = READ_ONCE(vc->runnable_threads[i]);
135 /* Used to traverse the list of runnable threads for a given vcore */
136 #define for_each_runnable_thread(i, vcpu, vc) \
137 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
139 static bool kvmppc_ipi_thread(int cpu)
141 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
143 /* On POWER9 we can use msgsnd to IPI any cpu */
144 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
145 msg |= get_hard_smp_processor_id(cpu);
147 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
151 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
152 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
154 if (cpu_first_thread_sibling(cpu) ==
155 cpu_first_thread_sibling(smp_processor_id())) {
156 msg |= cpu_thread_in_core(cpu);
158 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
165 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
166 if (cpu >= 0 && cpu < nr_cpu_ids) {
167 if (paca[cpu].kvm_hstate.xics_phys) {
171 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
179 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
182 struct swait_queue_head *wqp;
184 wqp = kvm_arch_vcpu_wq(vcpu);
185 if (swq_has_sleeper(wqp)) {
187 ++vcpu->stat.halt_wakeup;
190 cpu = READ_ONCE(vcpu->arch.thread_cpu);
191 if (cpu >= 0 && kvmppc_ipi_thread(cpu))
194 /* CPU points to the first thread of the core */
196 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
197 smp_send_reschedule(cpu);
201 * We use the vcpu_load/put functions to measure stolen time.
202 * Stolen time is counted as time when either the vcpu is able to
203 * run as part of a virtual core, but the task running the vcore
204 * is preempted or sleeping, or when the vcpu needs something done
205 * in the kernel by the task running the vcpu, but that task is
206 * preempted or sleeping. Those two things have to be counted
207 * separately, since one of the vcpu tasks will take on the job
208 * of running the core, and the other vcpu tasks in the vcore will
209 * sleep waiting for it to do that, but that sleep shouldn't count
212 * Hence we accumulate stolen time when the vcpu can run as part of
213 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
214 * needs its task to do other things in the kernel (for example,
215 * service a page fault) in busy_stolen. We don't accumulate
216 * stolen time for a vcore when it is inactive, or for a vcpu
217 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
218 * a misnomer; it means that the vcpu task is not executing in
219 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
220 * the kernel. We don't have any way of dividing up that time
221 * between time that the vcpu is genuinely stopped, time that
222 * the task is actively working on behalf of the vcpu, and time
223 * that the task is preempted, so we don't count any of it as
226 * Updates to busy_stolen are protected by arch.tbacct_lock;
227 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
228 * lock. The stolen times are measured in units of timebase ticks.
229 * (Note that the != TB_NIL checks below are purely defensive;
230 * they should never fail.)
233 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
237 spin_lock_irqsave(&vc->stoltb_lock, flags);
238 vc->preempt_tb = mftb();
239 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
242 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
246 spin_lock_irqsave(&vc->stoltb_lock, flags);
247 if (vc->preempt_tb != TB_NIL) {
248 vc->stolen_tb += mftb() - vc->preempt_tb;
249 vc->preempt_tb = TB_NIL;
251 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
254 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
256 struct kvmppc_vcore *vc = vcpu->arch.vcore;
260 * We can test vc->runner without taking the vcore lock,
261 * because only this task ever sets vc->runner to this
262 * vcpu, and once it is set to this vcpu, only this task
263 * ever sets it to NULL.
265 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
266 kvmppc_core_end_stolen(vc);
268 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
269 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
270 vcpu->arch.busy_preempt != TB_NIL) {
271 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
272 vcpu->arch.busy_preempt = TB_NIL;
274 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
277 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
279 struct kvmppc_vcore *vc = vcpu->arch.vcore;
282 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
283 kvmppc_core_start_stolen(vc);
285 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
286 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
287 vcpu->arch.busy_preempt = mftb();
288 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
291 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
294 * Check for illegal transactional state bit combination
295 * and if we find it, force the TS field to a safe state.
297 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
299 vcpu->arch.shregs.msr = msr;
300 kvmppc_end_cede(vcpu);
303 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
305 vcpu->arch.pvr = pvr;
308 /* Dummy value used in computing PCR value below */
309 #define PCR_ARCH_300 (PCR_ARCH_207 << 1)
311 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
313 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
314 struct kvmppc_vcore *vc = vcpu->arch.vcore;
316 /* We can (emulate) our own architecture version and anything older */
317 if (cpu_has_feature(CPU_FTR_ARCH_300))
318 host_pcr_bit = PCR_ARCH_300;
319 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
320 host_pcr_bit = PCR_ARCH_207;
321 else if (cpu_has_feature(CPU_FTR_ARCH_206))
322 host_pcr_bit = PCR_ARCH_206;
324 host_pcr_bit = PCR_ARCH_205;
326 /* Determine lowest PCR bit needed to run guest in given PVR level */
327 guest_pcr_bit = host_pcr_bit;
329 switch (arch_compat) {
331 guest_pcr_bit = PCR_ARCH_205;
335 guest_pcr_bit = PCR_ARCH_206;
338 guest_pcr_bit = PCR_ARCH_207;
341 guest_pcr_bit = PCR_ARCH_300;
348 /* Check requested PCR bits don't exceed our capabilities */
349 if (guest_pcr_bit > host_pcr_bit)
352 spin_lock(&vc->lock);
353 vc->arch_compat = arch_compat;
354 /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
355 vc->pcr = host_pcr_bit - guest_pcr_bit;
356 spin_unlock(&vc->lock);
361 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
365 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
366 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
367 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
368 for (r = 0; r < 16; ++r)
369 pr_err("r%2d = %.16lx r%d = %.16lx\n",
370 r, kvmppc_get_gpr(vcpu, r),
371 r+16, kvmppc_get_gpr(vcpu, r+16));
372 pr_err("ctr = %.16lx lr = %.16lx\n",
373 vcpu->arch.ctr, vcpu->arch.lr);
374 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
375 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
376 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
377 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
378 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
379 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
380 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
381 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
382 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
383 pr_err("fault dar = %.16lx dsisr = %.8x\n",
384 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
385 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
386 for (r = 0; r < vcpu->arch.slb_max; ++r)
387 pr_err(" ESID = %.16llx VSID = %.16llx\n",
388 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
389 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
390 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
391 vcpu->arch.last_inst);
394 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
396 struct kvm_vcpu *ret;
398 mutex_lock(&kvm->lock);
399 ret = kvm_get_vcpu_by_id(kvm, id);
400 mutex_unlock(&kvm->lock);
404 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
406 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
407 vpa->yield_count = cpu_to_be32(1);
410 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
411 unsigned long addr, unsigned long len)
413 /* check address is cacheline aligned */
414 if (addr & (L1_CACHE_BYTES - 1))
416 spin_lock(&vcpu->arch.vpa_update_lock);
417 if (v->next_gpa != addr || v->len != len) {
419 v->len = addr ? len : 0;
420 v->update_pending = 1;
422 spin_unlock(&vcpu->arch.vpa_update_lock);
426 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
435 static int vpa_is_registered(struct kvmppc_vpa *vpap)
437 if (vpap->update_pending)
438 return vpap->next_gpa != 0;
439 return vpap->pinned_addr != NULL;
442 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
444 unsigned long vcpuid, unsigned long vpa)
446 struct kvm *kvm = vcpu->kvm;
447 unsigned long len, nb;
449 struct kvm_vcpu *tvcpu;
452 struct kvmppc_vpa *vpap;
454 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
458 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
459 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
460 subfunc == H_VPA_REG_SLB) {
461 /* Registering new area - address must be cache-line aligned */
462 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
465 /* convert logical addr to kernel addr and read length */
466 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
469 if (subfunc == H_VPA_REG_VPA)
470 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
472 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
473 kvmppc_unpin_guest_page(kvm, va, vpa, false);
476 if (len > nb || len < sizeof(struct reg_vpa))
485 spin_lock(&tvcpu->arch.vpa_update_lock);
488 case H_VPA_REG_VPA: /* register VPA */
490 * The size of our lppaca is 1kB because of the way we align
491 * it for the guest to avoid crossing a 4kB boundary. We only
492 * use 640 bytes of the structure though, so we should accept
493 * clients that set a size of 640.
497 vpap = &tvcpu->arch.vpa;
501 case H_VPA_REG_DTL: /* register DTL */
502 if (len < sizeof(struct dtl_entry))
504 len -= len % sizeof(struct dtl_entry);
506 /* Check that they have previously registered a VPA */
508 if (!vpa_is_registered(&tvcpu->arch.vpa))
511 vpap = &tvcpu->arch.dtl;
515 case H_VPA_REG_SLB: /* register SLB shadow buffer */
516 /* Check that they have previously registered a VPA */
518 if (!vpa_is_registered(&tvcpu->arch.vpa))
521 vpap = &tvcpu->arch.slb_shadow;
525 case H_VPA_DEREG_VPA: /* deregister VPA */
526 /* Check they don't still have a DTL or SLB buf registered */
528 if (vpa_is_registered(&tvcpu->arch.dtl) ||
529 vpa_is_registered(&tvcpu->arch.slb_shadow))
532 vpap = &tvcpu->arch.vpa;
536 case H_VPA_DEREG_DTL: /* deregister DTL */
537 vpap = &tvcpu->arch.dtl;
541 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
542 vpap = &tvcpu->arch.slb_shadow;
548 vpap->next_gpa = vpa;
550 vpap->update_pending = 1;
553 spin_unlock(&tvcpu->arch.vpa_update_lock);
558 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
560 struct kvm *kvm = vcpu->kvm;
566 * We need to pin the page pointed to by vpap->next_gpa,
567 * but we can't call kvmppc_pin_guest_page under the lock
568 * as it does get_user_pages() and down_read(). So we
569 * have to drop the lock, pin the page, then get the lock
570 * again and check that a new area didn't get registered
574 gpa = vpap->next_gpa;
575 spin_unlock(&vcpu->arch.vpa_update_lock);
579 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
580 spin_lock(&vcpu->arch.vpa_update_lock);
581 if (gpa == vpap->next_gpa)
583 /* sigh... unpin that one and try again */
585 kvmppc_unpin_guest_page(kvm, va, gpa, false);
588 vpap->update_pending = 0;
589 if (va && nb < vpap->len) {
591 * If it's now too short, it must be that userspace
592 * has changed the mappings underlying guest memory,
593 * so unregister the region.
595 kvmppc_unpin_guest_page(kvm, va, gpa, false);
598 if (vpap->pinned_addr)
599 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
602 vpap->pinned_addr = va;
605 vpap->pinned_end = va + vpap->len;
608 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
610 if (!(vcpu->arch.vpa.update_pending ||
611 vcpu->arch.slb_shadow.update_pending ||
612 vcpu->arch.dtl.update_pending))
615 spin_lock(&vcpu->arch.vpa_update_lock);
616 if (vcpu->arch.vpa.update_pending) {
617 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
618 if (vcpu->arch.vpa.pinned_addr)
619 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
621 if (vcpu->arch.dtl.update_pending) {
622 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
623 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
624 vcpu->arch.dtl_index = 0;
626 if (vcpu->arch.slb_shadow.update_pending)
627 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
628 spin_unlock(&vcpu->arch.vpa_update_lock);
632 * Return the accumulated stolen time for the vcore up until `now'.
633 * The caller should hold the vcore lock.
635 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
640 spin_lock_irqsave(&vc->stoltb_lock, flags);
642 if (vc->vcore_state != VCORE_INACTIVE &&
643 vc->preempt_tb != TB_NIL)
644 p += now - vc->preempt_tb;
645 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
649 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
650 struct kvmppc_vcore *vc)
652 struct dtl_entry *dt;
654 unsigned long stolen;
655 unsigned long core_stolen;
659 dt = vcpu->arch.dtl_ptr;
660 vpa = vcpu->arch.vpa.pinned_addr;
662 core_stolen = vcore_stolen_time(vc, now);
663 stolen = core_stolen - vcpu->arch.stolen_logged;
664 vcpu->arch.stolen_logged = core_stolen;
665 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
666 stolen += vcpu->arch.busy_stolen;
667 vcpu->arch.busy_stolen = 0;
668 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
671 memset(dt, 0, sizeof(struct dtl_entry));
672 dt->dispatch_reason = 7;
673 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
674 dt->timebase = cpu_to_be64(now + vc->tb_offset);
675 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
676 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
677 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
679 if (dt == vcpu->arch.dtl.pinned_end)
680 dt = vcpu->arch.dtl.pinned_addr;
681 vcpu->arch.dtl_ptr = dt;
682 /* order writing *dt vs. writing vpa->dtl_idx */
684 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
685 vcpu->arch.dtl.dirty = true;
688 /* See if there is a doorbell interrupt pending for a vcpu */
689 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
692 struct kvmppc_vcore *vc;
694 if (vcpu->arch.doorbell_request)
697 * Ensure that the read of vcore->dpdes comes after the read
698 * of vcpu->doorbell_request. This barrier matches the
699 * lwsync in book3s_hv_rmhandlers.S just before the
700 * fast_guest_return label.
703 vc = vcpu->arch.vcore;
704 thr = vcpu->vcpu_id - vc->first_vcpuid;
705 return !!(vc->dpdes & (1 << thr));
708 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
710 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
712 if ((!vcpu->arch.vcore->arch_compat) &&
713 cpu_has_feature(CPU_FTR_ARCH_207S))
718 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
719 unsigned long resource, unsigned long value1,
720 unsigned long value2)
723 case H_SET_MODE_RESOURCE_SET_CIABR:
724 if (!kvmppc_power8_compatible(vcpu))
729 return H_UNSUPPORTED_FLAG_START;
730 /* Guests can't breakpoint the hypervisor */
731 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
733 vcpu->arch.ciabr = value1;
735 case H_SET_MODE_RESOURCE_SET_DAWR:
736 if (!kvmppc_power8_compatible(vcpu))
739 return H_UNSUPPORTED_FLAG_START;
740 if (value2 & DABRX_HYP)
742 vcpu->arch.dawr = value1;
743 vcpu->arch.dawrx = value2;
750 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
752 struct kvmppc_vcore *vcore = target->arch.vcore;
755 * We expect to have been called by the real mode handler
756 * (kvmppc_rm_h_confer()) which would have directly returned
757 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
758 * have useful work to do and should not confer) so we don't
762 spin_lock(&vcore->lock);
763 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
764 vcore->vcore_state != VCORE_INACTIVE &&
766 target = vcore->runner;
767 spin_unlock(&vcore->lock);
769 return kvm_vcpu_yield_to(target);
772 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
775 struct lppaca *lppaca;
777 spin_lock(&vcpu->arch.vpa_update_lock);
778 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
780 yield_count = be32_to_cpu(lppaca->yield_count);
781 spin_unlock(&vcpu->arch.vpa_update_lock);
785 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
787 unsigned long req = kvmppc_get_gpr(vcpu, 3);
788 unsigned long target, ret = H_SUCCESS;
790 struct kvm_vcpu *tvcpu;
793 if (req <= MAX_HCALL_OPCODE &&
794 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
801 target = kvmppc_get_gpr(vcpu, 4);
802 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
807 tvcpu->arch.prodded = 1;
809 if (tvcpu->arch.ceded)
810 kvmppc_fast_vcpu_kick_hv(tvcpu);
813 target = kvmppc_get_gpr(vcpu, 4);
816 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
821 yield_count = kvmppc_get_gpr(vcpu, 5);
822 if (kvmppc_get_yield_count(tvcpu) != yield_count)
824 kvm_arch_vcpu_yield_to(tvcpu);
827 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
828 kvmppc_get_gpr(vcpu, 5),
829 kvmppc_get_gpr(vcpu, 6));
832 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
835 idx = srcu_read_lock(&vcpu->kvm->srcu);
836 rc = kvmppc_rtas_hcall(vcpu);
837 srcu_read_unlock(&vcpu->kvm->srcu, idx);
844 /* Send the error out to userspace via KVM_RUN */
846 case H_LOGICAL_CI_LOAD:
847 ret = kvmppc_h_logical_ci_load(vcpu);
848 if (ret == H_TOO_HARD)
851 case H_LOGICAL_CI_STORE:
852 ret = kvmppc_h_logical_ci_store(vcpu);
853 if (ret == H_TOO_HARD)
857 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
858 kvmppc_get_gpr(vcpu, 5),
859 kvmppc_get_gpr(vcpu, 6),
860 kvmppc_get_gpr(vcpu, 7));
861 if (ret == H_TOO_HARD)
870 if (kvmppc_xics_enabled(vcpu)) {
871 if (xive_enabled()) {
872 ret = H_NOT_AVAILABLE;
875 ret = kvmppc_xics_hcall(vcpu, req);
880 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
881 kvmppc_get_gpr(vcpu, 5),
882 kvmppc_get_gpr(vcpu, 6));
883 if (ret == H_TOO_HARD)
886 case H_PUT_TCE_INDIRECT:
887 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
888 kvmppc_get_gpr(vcpu, 5),
889 kvmppc_get_gpr(vcpu, 6),
890 kvmppc_get_gpr(vcpu, 7));
891 if (ret == H_TOO_HARD)
895 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
896 kvmppc_get_gpr(vcpu, 5),
897 kvmppc_get_gpr(vcpu, 6),
898 kvmppc_get_gpr(vcpu, 7));
899 if (ret == H_TOO_HARD)
905 kvmppc_set_gpr(vcpu, 3, ret);
906 vcpu->arch.hcall_needed = 0;
910 static int kvmppc_hcall_impl_hv(unsigned long cmd)
918 case H_LOGICAL_CI_LOAD:
919 case H_LOGICAL_CI_STORE:
920 #ifdef CONFIG_KVM_XICS
931 /* See if it's in the real-mode table */
932 return kvmppc_hcall_impl_hv_realmode(cmd);
935 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
936 struct kvm_vcpu *vcpu)
940 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
943 * Fetch failed, so return to guest and
944 * try executing it again.
949 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
950 run->exit_reason = KVM_EXIT_DEBUG;
951 run->debug.arch.address = kvmppc_get_pc(vcpu);
954 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
959 static void do_nothing(void *x)
963 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
965 int thr, cpu, pcpu, nthreads;
969 nthreads = vcpu->kvm->arch.emul_smt_mode;
971 cpu = vcpu->vcpu_id & ~(nthreads - 1);
972 for (thr = 0; thr < nthreads; ++thr, ++cpu) {
973 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
977 * If the vcpu is currently running on a physical cpu thread,
978 * interrupt it in order to pull it out of the guest briefly,
979 * which will update its vcore->dpdes value.
981 pcpu = READ_ONCE(v->cpu);
983 smp_call_function_single(pcpu, do_nothing, NULL, 1);
984 if (kvmppc_doorbell_pending(v))
991 * On POWER9, emulate doorbell-related instructions in order to
992 * give the guest the illusion of running on a multi-threaded core.
993 * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
996 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1000 struct kvm *kvm = vcpu->kvm;
1001 struct kvm_vcpu *tvcpu;
1003 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1004 return EMULATE_FAIL;
1005 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1006 return RESUME_GUEST;
1007 if (get_op(inst) != 31)
1008 return EMULATE_FAIL;
1010 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1011 switch (get_xop(inst)) {
1012 case OP_31_XOP_MSGSNDP:
1013 arg = kvmppc_get_gpr(vcpu, rb);
1014 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1017 if (arg >= kvm->arch.emul_smt_mode)
1019 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1022 if (!tvcpu->arch.doorbell_request) {
1023 tvcpu->arch.doorbell_request = 1;
1024 kvmppc_fast_vcpu_kick_hv(tvcpu);
1027 case OP_31_XOP_MSGCLRP:
1028 arg = kvmppc_get_gpr(vcpu, rb);
1029 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1031 vcpu->arch.vcore->dpdes = 0;
1032 vcpu->arch.doorbell_request = 0;
1034 case OP_31_XOP_MFSPR:
1035 switch (get_sprn(inst)) {
1040 arg = kvmppc_read_dpdes(vcpu);
1043 return EMULATE_FAIL;
1045 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1048 return EMULATE_FAIL;
1050 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1051 return RESUME_GUEST;
1054 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1055 struct task_struct *tsk)
1057 int r = RESUME_HOST;
1059 vcpu->stat.sum_exits++;
1062 * This can happen if an interrupt occurs in the last stages
1063 * of guest entry or the first stages of guest exit (i.e. after
1064 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1065 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1066 * That can happen due to a bug, or due to a machine check
1067 * occurring at just the wrong time.
1069 if (vcpu->arch.shregs.msr & MSR_HV) {
1070 printk(KERN_EMERG "KVM trap in HV mode!\n");
1071 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1072 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1073 vcpu->arch.shregs.msr);
1074 kvmppc_dump_regs(vcpu);
1075 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1076 run->hw.hardware_exit_reason = vcpu->arch.trap;
1079 run->exit_reason = KVM_EXIT_UNKNOWN;
1080 run->ready_for_interrupt_injection = 1;
1081 switch (vcpu->arch.trap) {
1082 /* We're good on these - the host merely wanted to get our attention */
1083 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1084 vcpu->stat.dec_exits++;
1087 case BOOK3S_INTERRUPT_EXTERNAL:
1088 case BOOK3S_INTERRUPT_H_DOORBELL:
1089 case BOOK3S_INTERRUPT_H_VIRT:
1090 vcpu->stat.ext_intr_exits++;
1093 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1094 case BOOK3S_INTERRUPT_HMI:
1095 case BOOK3S_INTERRUPT_PERFMON:
1096 case BOOK3S_INTERRUPT_SYSTEM_RESET:
1099 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1100 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1101 run->exit_reason = KVM_EXIT_NMI;
1102 run->hw.hardware_exit_reason = vcpu->arch.trap;
1103 /* Clear out the old NMI status from run->flags */
1104 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1105 /* Now set the NMI status */
1106 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1107 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1109 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1112 /* Print the MCE event to host console. */
1113 machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1115 case BOOK3S_INTERRUPT_PROGRAM:
1119 * Normally program interrupts are delivered directly
1120 * to the guest by the hardware, but we can get here
1121 * as a result of a hypervisor emulation interrupt
1122 * (e40) getting turned into a 700 by BML RTAS.
1124 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1125 kvmppc_core_queue_program(vcpu, flags);
1129 case BOOK3S_INTERRUPT_SYSCALL:
1131 /* hcall - punt to userspace */
1134 /* hypercall with MSR_PR has already been handled in rmode,
1135 * and never reaches here.
1138 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1139 for (i = 0; i < 9; ++i)
1140 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1141 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1142 vcpu->arch.hcall_needed = 1;
1147 * We get these next two if the guest accesses a page which it thinks
1148 * it has mapped but which is not actually present, either because
1149 * it is for an emulated I/O device or because the corresonding
1150 * host page has been paged out. Any other HDSI/HISI interrupts
1151 * have been handled already.
1153 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1154 r = RESUME_PAGE_FAULT;
1156 case BOOK3S_INTERRUPT_H_INST_STORAGE:
1157 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1158 vcpu->arch.fault_dsisr = 0;
1159 r = RESUME_PAGE_FAULT;
1162 * This occurs if the guest executes an illegal instruction.
1163 * If the guest debug is disabled, generate a program interrupt
1164 * to the guest. If guest debug is enabled, we need to check
1165 * whether the instruction is a software breakpoint instruction.
1166 * Accordingly return to Guest or Host.
1168 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1169 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1170 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1171 swab32(vcpu->arch.emul_inst) :
1172 vcpu->arch.emul_inst;
1173 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1174 r = kvmppc_emulate_debug_inst(run, vcpu);
1176 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1181 * This occurs if the guest (kernel or userspace), does something that
1182 * is prohibited by HFSCR.
1183 * On POWER9, this could be a doorbell instruction that we need
1185 * Otherwise, we just generate a program interrupt to the guest.
1187 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1189 if ((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG)
1190 r = kvmppc_emulate_doorbell_instr(vcpu);
1191 if (r == EMULATE_FAIL) {
1192 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1196 case BOOK3S_INTERRUPT_HV_RM_HARD:
1197 r = RESUME_PASSTHROUGH;
1200 kvmppc_dump_regs(vcpu);
1201 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1202 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1203 vcpu->arch.shregs.msr);
1204 run->hw.hardware_exit_reason = vcpu->arch.trap;
1212 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1213 struct kvm_sregs *sregs)
1217 memset(sregs, 0, sizeof(struct kvm_sregs));
1218 sregs->pvr = vcpu->arch.pvr;
1219 for (i = 0; i < vcpu->arch.slb_max; i++) {
1220 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1221 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1227 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1228 struct kvm_sregs *sregs)
1232 /* Only accept the same PVR as the host's, since we can't spoof it */
1233 if (sregs->pvr != vcpu->arch.pvr)
1237 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1238 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1239 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1240 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1244 vcpu->arch.slb_max = j;
1249 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1250 bool preserve_top32)
1252 struct kvm *kvm = vcpu->kvm;
1253 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1256 mutex_lock(&kvm->lock);
1257 spin_lock(&vc->lock);
1259 * If ILE (interrupt little-endian) has changed, update the
1260 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1262 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1263 struct kvm_vcpu *vcpu;
1266 kvm_for_each_vcpu(i, vcpu, kvm) {
1267 if (vcpu->arch.vcore != vc)
1269 if (new_lpcr & LPCR_ILE)
1270 vcpu->arch.intr_msr |= MSR_LE;
1272 vcpu->arch.intr_msr &= ~MSR_LE;
1277 * Userspace can only modify DPFD (default prefetch depth),
1278 * ILE (interrupt little-endian) and TC (translation control).
1279 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1281 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1282 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1285 * On POWER9, allow userspace to enable large decrementer for the
1286 * guest, whether or not the host has it enabled.
1288 if (cpu_has_feature(CPU_FTR_ARCH_300))
1291 /* Broken 32-bit version of LPCR must not clear top bits */
1294 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1295 spin_unlock(&vc->lock);
1296 mutex_unlock(&kvm->lock);
1299 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1300 union kvmppc_one_reg *val)
1306 case KVM_REG_PPC_DEBUG_INST:
1307 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1309 case KVM_REG_PPC_HIOR:
1310 *val = get_reg_val(id, 0);
1312 case KVM_REG_PPC_DABR:
1313 *val = get_reg_val(id, vcpu->arch.dabr);
1315 case KVM_REG_PPC_DABRX:
1316 *val = get_reg_val(id, vcpu->arch.dabrx);
1318 case KVM_REG_PPC_DSCR:
1319 *val = get_reg_val(id, vcpu->arch.dscr);
1321 case KVM_REG_PPC_PURR:
1322 *val = get_reg_val(id, vcpu->arch.purr);
1324 case KVM_REG_PPC_SPURR:
1325 *val = get_reg_val(id, vcpu->arch.spurr);
1327 case KVM_REG_PPC_AMR:
1328 *val = get_reg_val(id, vcpu->arch.amr);
1330 case KVM_REG_PPC_UAMOR:
1331 *val = get_reg_val(id, vcpu->arch.uamor);
1333 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1334 i = id - KVM_REG_PPC_MMCR0;
1335 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1337 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1338 i = id - KVM_REG_PPC_PMC1;
1339 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1341 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1342 i = id - KVM_REG_PPC_SPMC1;
1343 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1345 case KVM_REG_PPC_SIAR:
1346 *val = get_reg_val(id, vcpu->arch.siar);
1348 case KVM_REG_PPC_SDAR:
1349 *val = get_reg_val(id, vcpu->arch.sdar);
1351 case KVM_REG_PPC_SIER:
1352 *val = get_reg_val(id, vcpu->arch.sier);
1354 case KVM_REG_PPC_IAMR:
1355 *val = get_reg_val(id, vcpu->arch.iamr);
1357 case KVM_REG_PPC_PSPB:
1358 *val = get_reg_val(id, vcpu->arch.pspb);
1360 case KVM_REG_PPC_DPDES:
1361 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1363 case KVM_REG_PPC_VTB:
1364 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1366 case KVM_REG_PPC_DAWR:
1367 *val = get_reg_val(id, vcpu->arch.dawr);
1369 case KVM_REG_PPC_DAWRX:
1370 *val = get_reg_val(id, vcpu->arch.dawrx);
1372 case KVM_REG_PPC_CIABR:
1373 *val = get_reg_val(id, vcpu->arch.ciabr);
1375 case KVM_REG_PPC_CSIGR:
1376 *val = get_reg_val(id, vcpu->arch.csigr);
1378 case KVM_REG_PPC_TACR:
1379 *val = get_reg_val(id, vcpu->arch.tacr);
1381 case KVM_REG_PPC_TCSCR:
1382 *val = get_reg_val(id, vcpu->arch.tcscr);
1384 case KVM_REG_PPC_PID:
1385 *val = get_reg_val(id, vcpu->arch.pid);
1387 case KVM_REG_PPC_ACOP:
1388 *val = get_reg_val(id, vcpu->arch.acop);
1390 case KVM_REG_PPC_WORT:
1391 *val = get_reg_val(id, vcpu->arch.wort);
1393 case KVM_REG_PPC_TIDR:
1394 *val = get_reg_val(id, vcpu->arch.tid);
1396 case KVM_REG_PPC_PSSCR:
1397 *val = get_reg_val(id, vcpu->arch.psscr);
1399 case KVM_REG_PPC_VPA_ADDR:
1400 spin_lock(&vcpu->arch.vpa_update_lock);
1401 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1402 spin_unlock(&vcpu->arch.vpa_update_lock);
1404 case KVM_REG_PPC_VPA_SLB:
1405 spin_lock(&vcpu->arch.vpa_update_lock);
1406 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1407 val->vpaval.length = vcpu->arch.slb_shadow.len;
1408 spin_unlock(&vcpu->arch.vpa_update_lock);
1410 case KVM_REG_PPC_VPA_DTL:
1411 spin_lock(&vcpu->arch.vpa_update_lock);
1412 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1413 val->vpaval.length = vcpu->arch.dtl.len;
1414 spin_unlock(&vcpu->arch.vpa_update_lock);
1416 case KVM_REG_PPC_TB_OFFSET:
1417 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1419 case KVM_REG_PPC_LPCR:
1420 case KVM_REG_PPC_LPCR_64:
1421 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1423 case KVM_REG_PPC_PPR:
1424 *val = get_reg_val(id, vcpu->arch.ppr);
1426 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1427 case KVM_REG_PPC_TFHAR:
1428 *val = get_reg_val(id, vcpu->arch.tfhar);
1430 case KVM_REG_PPC_TFIAR:
1431 *val = get_reg_val(id, vcpu->arch.tfiar);
1433 case KVM_REG_PPC_TEXASR:
1434 *val = get_reg_val(id, vcpu->arch.texasr);
1436 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1437 i = id - KVM_REG_PPC_TM_GPR0;
1438 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1440 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1443 i = id - KVM_REG_PPC_TM_VSR0;
1445 for (j = 0; j < TS_FPRWIDTH; j++)
1446 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1448 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1449 val->vval = vcpu->arch.vr_tm.vr[i-32];
1455 case KVM_REG_PPC_TM_CR:
1456 *val = get_reg_val(id, vcpu->arch.cr_tm);
1458 case KVM_REG_PPC_TM_XER:
1459 *val = get_reg_val(id, vcpu->arch.xer_tm);
1461 case KVM_REG_PPC_TM_LR:
1462 *val = get_reg_val(id, vcpu->arch.lr_tm);
1464 case KVM_REG_PPC_TM_CTR:
1465 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1467 case KVM_REG_PPC_TM_FPSCR:
1468 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1470 case KVM_REG_PPC_TM_AMR:
1471 *val = get_reg_val(id, vcpu->arch.amr_tm);
1473 case KVM_REG_PPC_TM_PPR:
1474 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1476 case KVM_REG_PPC_TM_VRSAVE:
1477 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1479 case KVM_REG_PPC_TM_VSCR:
1480 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1481 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1485 case KVM_REG_PPC_TM_DSCR:
1486 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1488 case KVM_REG_PPC_TM_TAR:
1489 *val = get_reg_val(id, vcpu->arch.tar_tm);
1492 case KVM_REG_PPC_ARCH_COMPAT:
1493 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1503 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1504 union kvmppc_one_reg *val)
1508 unsigned long addr, len;
1511 case KVM_REG_PPC_HIOR:
1512 /* Only allow this to be set to zero */
1513 if (set_reg_val(id, *val))
1516 case KVM_REG_PPC_DABR:
1517 vcpu->arch.dabr = set_reg_val(id, *val);
1519 case KVM_REG_PPC_DABRX:
1520 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1522 case KVM_REG_PPC_DSCR:
1523 vcpu->arch.dscr = set_reg_val(id, *val);
1525 case KVM_REG_PPC_PURR:
1526 vcpu->arch.purr = set_reg_val(id, *val);
1528 case KVM_REG_PPC_SPURR:
1529 vcpu->arch.spurr = set_reg_val(id, *val);
1531 case KVM_REG_PPC_AMR:
1532 vcpu->arch.amr = set_reg_val(id, *val);
1534 case KVM_REG_PPC_UAMOR:
1535 vcpu->arch.uamor = set_reg_val(id, *val);
1537 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1538 i = id - KVM_REG_PPC_MMCR0;
1539 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1541 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1542 i = id - KVM_REG_PPC_PMC1;
1543 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1545 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1546 i = id - KVM_REG_PPC_SPMC1;
1547 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1549 case KVM_REG_PPC_SIAR:
1550 vcpu->arch.siar = set_reg_val(id, *val);
1552 case KVM_REG_PPC_SDAR:
1553 vcpu->arch.sdar = set_reg_val(id, *val);
1555 case KVM_REG_PPC_SIER:
1556 vcpu->arch.sier = set_reg_val(id, *val);
1558 case KVM_REG_PPC_IAMR:
1559 vcpu->arch.iamr = set_reg_val(id, *val);
1561 case KVM_REG_PPC_PSPB:
1562 vcpu->arch.pspb = set_reg_val(id, *val);
1564 case KVM_REG_PPC_DPDES:
1565 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1567 case KVM_REG_PPC_VTB:
1568 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1570 case KVM_REG_PPC_DAWR:
1571 vcpu->arch.dawr = set_reg_val(id, *val);
1573 case KVM_REG_PPC_DAWRX:
1574 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1576 case KVM_REG_PPC_CIABR:
1577 vcpu->arch.ciabr = set_reg_val(id, *val);
1578 /* Don't allow setting breakpoints in hypervisor code */
1579 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1580 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1582 case KVM_REG_PPC_CSIGR:
1583 vcpu->arch.csigr = set_reg_val(id, *val);
1585 case KVM_REG_PPC_TACR:
1586 vcpu->arch.tacr = set_reg_val(id, *val);
1588 case KVM_REG_PPC_TCSCR:
1589 vcpu->arch.tcscr = set_reg_val(id, *val);
1591 case KVM_REG_PPC_PID:
1592 vcpu->arch.pid = set_reg_val(id, *val);
1594 case KVM_REG_PPC_ACOP:
1595 vcpu->arch.acop = set_reg_val(id, *val);
1597 case KVM_REG_PPC_WORT:
1598 vcpu->arch.wort = set_reg_val(id, *val);
1600 case KVM_REG_PPC_TIDR:
1601 vcpu->arch.tid = set_reg_val(id, *val);
1603 case KVM_REG_PPC_PSSCR:
1604 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1606 case KVM_REG_PPC_VPA_ADDR:
1607 addr = set_reg_val(id, *val);
1609 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1610 vcpu->arch.dtl.next_gpa))
1612 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1614 case KVM_REG_PPC_VPA_SLB:
1615 addr = val->vpaval.addr;
1616 len = val->vpaval.length;
1618 if (addr && !vcpu->arch.vpa.next_gpa)
1620 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1622 case KVM_REG_PPC_VPA_DTL:
1623 addr = val->vpaval.addr;
1624 len = val->vpaval.length;
1626 if (addr && (len < sizeof(struct dtl_entry) ||
1627 !vcpu->arch.vpa.next_gpa))
1629 len -= len % sizeof(struct dtl_entry);
1630 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1632 case KVM_REG_PPC_TB_OFFSET:
1634 * POWER9 DD1 has an erratum where writing TBU40 causes
1635 * the timebase to lose ticks. So we don't let the
1636 * timebase offset be changed on P9 DD1. (It is
1637 * initialized to zero.)
1639 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1641 /* round up to multiple of 2^24 */
1642 vcpu->arch.vcore->tb_offset =
1643 ALIGN(set_reg_val(id, *val), 1UL << 24);
1645 case KVM_REG_PPC_LPCR:
1646 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1648 case KVM_REG_PPC_LPCR_64:
1649 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1651 case KVM_REG_PPC_PPR:
1652 vcpu->arch.ppr = set_reg_val(id, *val);
1654 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1655 case KVM_REG_PPC_TFHAR:
1656 vcpu->arch.tfhar = set_reg_val(id, *val);
1658 case KVM_REG_PPC_TFIAR:
1659 vcpu->arch.tfiar = set_reg_val(id, *val);
1661 case KVM_REG_PPC_TEXASR:
1662 vcpu->arch.texasr = set_reg_val(id, *val);
1664 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1665 i = id - KVM_REG_PPC_TM_GPR0;
1666 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1668 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1671 i = id - KVM_REG_PPC_TM_VSR0;
1673 for (j = 0; j < TS_FPRWIDTH; j++)
1674 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1676 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1677 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1682 case KVM_REG_PPC_TM_CR:
1683 vcpu->arch.cr_tm = set_reg_val(id, *val);
1685 case KVM_REG_PPC_TM_XER:
1686 vcpu->arch.xer_tm = set_reg_val(id, *val);
1688 case KVM_REG_PPC_TM_LR:
1689 vcpu->arch.lr_tm = set_reg_val(id, *val);
1691 case KVM_REG_PPC_TM_CTR:
1692 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1694 case KVM_REG_PPC_TM_FPSCR:
1695 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1697 case KVM_REG_PPC_TM_AMR:
1698 vcpu->arch.amr_tm = set_reg_val(id, *val);
1700 case KVM_REG_PPC_TM_PPR:
1701 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1703 case KVM_REG_PPC_TM_VRSAVE:
1704 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1706 case KVM_REG_PPC_TM_VSCR:
1707 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1708 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1712 case KVM_REG_PPC_TM_DSCR:
1713 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1715 case KVM_REG_PPC_TM_TAR:
1716 vcpu->arch.tar_tm = set_reg_val(id, *val);
1719 case KVM_REG_PPC_ARCH_COMPAT:
1720 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1731 * On POWER9, threads are independent and can be in different partitions.
1732 * Therefore we consider each thread to be a subcore.
1733 * There is a restriction that all threads have to be in the same
1734 * MMU mode (radix or HPT), unfortunately, but since we only support
1735 * HPT guests on a HPT host so far, that isn't an impediment yet.
1737 static int threads_per_vcore(void)
1739 if (cpu_has_feature(CPU_FTR_ARCH_300))
1741 return threads_per_subcore;
1744 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1746 struct kvmppc_vcore *vcore;
1748 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1753 spin_lock_init(&vcore->lock);
1754 spin_lock_init(&vcore->stoltb_lock);
1755 init_swait_queue_head(&vcore->wq);
1756 vcore->preempt_tb = TB_NIL;
1757 vcore->lpcr = kvm->arch.lpcr;
1758 vcore->first_vcpuid = core * kvm->arch.smt_mode;
1760 INIT_LIST_HEAD(&vcore->preempt_list);
1765 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1766 static struct debugfs_timings_element {
1770 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1771 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1772 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1773 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1774 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1777 #define N_TIMINGS (sizeof(timings) / sizeof(timings[0]))
1779 struct debugfs_timings_state {
1780 struct kvm_vcpu *vcpu;
1781 unsigned int buflen;
1782 char buf[N_TIMINGS * 100];
1785 static int debugfs_timings_open(struct inode *inode, struct file *file)
1787 struct kvm_vcpu *vcpu = inode->i_private;
1788 struct debugfs_timings_state *p;
1790 p = kzalloc(sizeof(*p), GFP_KERNEL);
1794 kvm_get_kvm(vcpu->kvm);
1796 file->private_data = p;
1798 return nonseekable_open(inode, file);
1801 static int debugfs_timings_release(struct inode *inode, struct file *file)
1803 struct debugfs_timings_state *p = file->private_data;
1805 kvm_put_kvm(p->vcpu->kvm);
1810 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1811 size_t len, loff_t *ppos)
1813 struct debugfs_timings_state *p = file->private_data;
1814 struct kvm_vcpu *vcpu = p->vcpu;
1816 struct kvmhv_tb_accumulator tb;
1825 buf_end = s + sizeof(p->buf);
1826 for (i = 0; i < N_TIMINGS; ++i) {
1827 struct kvmhv_tb_accumulator *acc;
1829 acc = (struct kvmhv_tb_accumulator *)
1830 ((unsigned long)vcpu + timings[i].offset);
1832 for (loops = 0; loops < 1000; ++loops) {
1833 count = acc->seqcount;
1838 if (count == acc->seqcount) {
1846 snprintf(s, buf_end - s, "%s: stuck\n",
1849 snprintf(s, buf_end - s,
1850 "%s: %llu %llu %llu %llu\n",
1851 timings[i].name, count / 2,
1852 tb_to_ns(tb.tb_total),
1853 tb_to_ns(tb.tb_min),
1854 tb_to_ns(tb.tb_max));
1857 p->buflen = s - p->buf;
1861 if (pos >= p->buflen)
1863 if (len > p->buflen - pos)
1864 len = p->buflen - pos;
1865 n = copy_to_user(buf, p->buf + pos, len);
1875 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1876 size_t len, loff_t *ppos)
1881 static const struct file_operations debugfs_timings_ops = {
1882 .owner = THIS_MODULE,
1883 .open = debugfs_timings_open,
1884 .release = debugfs_timings_release,
1885 .read = debugfs_timings_read,
1886 .write = debugfs_timings_write,
1887 .llseek = generic_file_llseek,
1890 /* Create a debugfs directory for the vcpu */
1891 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1894 struct kvm *kvm = vcpu->kvm;
1896 snprintf(buf, sizeof(buf), "vcpu%u", id);
1897 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1899 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1900 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1902 vcpu->arch.debugfs_timings =
1903 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1904 vcpu, &debugfs_timings_ops);
1907 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1908 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1911 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1913 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1916 struct kvm_vcpu *vcpu;
1919 struct kvmppc_vcore *vcore;
1922 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1926 err = kvm_vcpu_init(vcpu, kvm, id);
1930 vcpu->arch.shared = &vcpu->arch.shregs;
1931 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1933 * The shared struct is never shared on HV,
1934 * so we can always use host endianness
1936 #ifdef __BIG_ENDIAN__
1937 vcpu->arch.shared_big_endian = true;
1939 vcpu->arch.shared_big_endian = false;
1942 vcpu->arch.mmcr[0] = MMCR0_FC;
1943 vcpu->arch.ctrl = CTRL_RUNLATCH;
1944 /* default to host PVR, since we can't spoof it */
1945 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1946 spin_lock_init(&vcpu->arch.vpa_update_lock);
1947 spin_lock_init(&vcpu->arch.tbacct_lock);
1948 vcpu->arch.busy_preempt = TB_NIL;
1949 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1952 * Set the default HFSCR for the guest from the host value.
1953 * This value is only used on POWER9.
1954 * On POWER9 DD1, TM doesn't work, so we make sure to
1955 * prevent the guest from using it.
1956 * On POWER9, we want to virtualize the doorbell facility, so we
1957 * turn off the HFSCR bit, which causes those instructions to trap.
1959 vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1960 if (!cpu_has_feature(CPU_FTR_TM))
1961 vcpu->arch.hfscr &= ~HFSCR_TM;
1962 if (cpu_has_feature(CPU_FTR_ARCH_300))
1963 vcpu->arch.hfscr &= ~HFSCR_MSGP;
1965 kvmppc_mmu_book3s_hv_init(vcpu);
1967 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1969 init_waitqueue_head(&vcpu->arch.cpu_run);
1971 mutex_lock(&kvm->lock);
1974 core = id / kvm->arch.smt_mode;
1975 if (core < KVM_MAX_VCORES) {
1976 vcore = kvm->arch.vcores[core];
1979 vcore = kvmppc_vcore_create(kvm, core);
1980 kvm->arch.vcores[core] = vcore;
1981 kvm->arch.online_vcores++;
1984 mutex_unlock(&kvm->lock);
1989 spin_lock(&vcore->lock);
1990 ++vcore->num_threads;
1991 spin_unlock(&vcore->lock);
1992 vcpu->arch.vcore = vcore;
1993 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1994 vcpu->arch.thread_cpu = -1;
1995 vcpu->arch.prev_cpu = -1;
1997 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1998 kvmppc_sanity_check(vcpu);
2000 debugfs_vcpu_init(vcpu, id);
2005 kmem_cache_free(kvm_vcpu_cache, vcpu);
2007 return ERR_PTR(err);
2010 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2011 unsigned long flags)
2018 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2020 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2022 * On POWER8 (or POWER7), the threading mode is "strict",
2023 * so we pack smt_mode vcpus per vcore.
2025 if (smt_mode > threads_per_subcore)
2029 * On POWER9, the threading mode is "loose",
2030 * so each vcpu gets its own vcore.
2035 mutex_lock(&kvm->lock);
2037 if (!kvm->arch.online_vcores) {
2038 kvm->arch.smt_mode = smt_mode;
2039 kvm->arch.emul_smt_mode = esmt;
2042 mutex_unlock(&kvm->lock);
2047 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2049 if (vpa->pinned_addr)
2050 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2054 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2056 spin_lock(&vcpu->arch.vpa_update_lock);
2057 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2058 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2059 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2060 spin_unlock(&vcpu->arch.vpa_update_lock);
2061 kvm_vcpu_uninit(vcpu);
2062 kmem_cache_free(kvm_vcpu_cache, vcpu);
2065 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2067 /* Indicate we want to get back into the guest */
2071 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2073 unsigned long dec_nsec, now;
2076 if (now > vcpu->arch.dec_expires) {
2077 /* decrementer has already gone negative */
2078 kvmppc_core_queue_dec(vcpu);
2079 kvmppc_core_prepare_to_enter(vcpu);
2082 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2084 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2085 vcpu->arch.timer_running = 1;
2088 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2090 vcpu->arch.ceded = 0;
2091 if (vcpu->arch.timer_running) {
2092 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2093 vcpu->arch.timer_running = 0;
2097 extern int __kvmppc_vcore_entry(void);
2099 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2100 struct kvm_vcpu *vcpu)
2104 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2106 spin_lock_irq(&vcpu->arch.tbacct_lock);
2108 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2109 vcpu->arch.stolen_logged;
2110 vcpu->arch.busy_preempt = now;
2111 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2112 spin_unlock_irq(&vcpu->arch.tbacct_lock);
2114 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2117 static int kvmppc_grab_hwthread(int cpu)
2119 struct paca_struct *tpaca;
2120 long timeout = 10000;
2124 /* Ensure the thread won't go into the kernel if it wakes */
2125 tpaca->kvm_hstate.kvm_vcpu = NULL;
2126 tpaca->kvm_hstate.kvm_vcore = NULL;
2127 tpaca->kvm_hstate.napping = 0;
2129 tpaca->kvm_hstate.hwthread_req = 1;
2132 * If the thread is already executing in the kernel (e.g. handling
2133 * a stray interrupt), wait for it to get back to nap mode.
2134 * The smp_mb() is to ensure that our setting of hwthread_req
2135 * is visible before we look at hwthread_state, so if this
2136 * races with the code at system_reset_pSeries and the thread
2137 * misses our setting of hwthread_req, we are sure to see its
2138 * setting of hwthread_state, and vice versa.
2141 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2142 if (--timeout <= 0) {
2143 pr_err("KVM: couldn't grab cpu %d\n", cpu);
2151 static void kvmppc_release_hwthread(int cpu)
2153 struct paca_struct *tpaca;
2156 tpaca->kvm_hstate.hwthread_req = 0;
2157 tpaca->kvm_hstate.kvm_vcpu = NULL;
2158 tpaca->kvm_hstate.kvm_vcore = NULL;
2159 tpaca->kvm_hstate.kvm_split_mode = NULL;
2162 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2166 cpu = cpu_first_thread_sibling(cpu);
2167 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2169 * Make sure setting of bit in need_tlb_flush precedes
2170 * testing of cpu_in_guest bits. The matching barrier on
2171 * the other side is the first smp_mb() in kvmppc_run_core().
2174 for (i = 0; i < threads_per_core; ++i)
2175 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2176 smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2179 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2181 struct kvm *kvm = vcpu->kvm;
2184 * With radix, the guest can do TLB invalidations itself,
2185 * and it could choose to use the local form (tlbiel) if
2186 * it is invalidating a translation that has only ever been
2187 * used on one vcpu. However, that doesn't mean it has
2188 * only ever been used on one physical cpu, since vcpus
2189 * can move around between pcpus. To cope with this, when
2190 * a vcpu moves from one pcpu to another, we need to tell
2191 * any vcpus running on the same core as this vcpu previously
2192 * ran to flush the TLB. The TLB is shared between threads,
2193 * so we use a single bit in .need_tlb_flush for all 4 threads.
2195 if (vcpu->arch.prev_cpu != pcpu) {
2196 if (vcpu->arch.prev_cpu >= 0 &&
2197 cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2198 cpu_first_thread_sibling(pcpu))
2199 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2200 vcpu->arch.prev_cpu = pcpu;
2204 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2207 struct paca_struct *tpaca;
2208 struct kvm *kvm = vc->kvm;
2212 if (vcpu->arch.timer_running) {
2213 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2214 vcpu->arch.timer_running = 0;
2216 cpu += vcpu->arch.ptid;
2217 vcpu->cpu = vc->pcpu;
2218 vcpu->arch.thread_cpu = cpu;
2219 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2222 tpaca->kvm_hstate.kvm_vcpu = vcpu;
2223 tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2224 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2226 tpaca->kvm_hstate.kvm_vcore = vc;
2227 if (cpu != smp_processor_id())
2228 kvmppc_ipi_thread(cpu);
2231 static void kvmppc_wait_for_nap(void)
2233 int cpu = smp_processor_id();
2235 int n_threads = threads_per_vcore();
2239 for (loops = 0; loops < 1000000; ++loops) {
2241 * Check if all threads are finished.
2242 * We set the vcore pointer when starting a thread
2243 * and the thread clears it when finished, so we look
2244 * for any threads that still have a non-NULL vcore ptr.
2246 for (i = 1; i < n_threads; ++i)
2247 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2249 if (i == n_threads) {
2256 for (i = 1; i < n_threads; ++i)
2257 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2258 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2262 * Check that we are on thread 0 and that any other threads in
2263 * this core are off-line. Then grab the threads so they can't
2266 static int on_primary_thread(void)
2268 int cpu = smp_processor_id();
2271 /* Are we on a primary subcore? */
2272 if (cpu_thread_in_subcore(cpu))
2276 while (++thr < threads_per_subcore)
2277 if (cpu_online(cpu + thr))
2280 /* Grab all hw threads so they can't go into the kernel */
2281 for (thr = 1; thr < threads_per_subcore; ++thr) {
2282 if (kvmppc_grab_hwthread(cpu + thr)) {
2283 /* Couldn't grab one; let the others go */
2285 kvmppc_release_hwthread(cpu + thr);
2286 } while (--thr > 0);
2294 * A list of virtual cores for each physical CPU.
2295 * These are vcores that could run but their runner VCPU tasks are
2296 * (or may be) preempted.
2298 struct preempted_vcore_list {
2299 struct list_head list;
2303 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2305 static void init_vcore_lists(void)
2309 for_each_possible_cpu(cpu) {
2310 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2311 spin_lock_init(&lp->lock);
2312 INIT_LIST_HEAD(&lp->list);
2316 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2318 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2320 vc->vcore_state = VCORE_PREEMPT;
2321 vc->pcpu = smp_processor_id();
2322 if (vc->num_threads < threads_per_vcore()) {
2323 spin_lock(&lp->lock);
2324 list_add_tail(&vc->preempt_list, &lp->list);
2325 spin_unlock(&lp->lock);
2328 /* Start accumulating stolen time */
2329 kvmppc_core_start_stolen(vc);
2332 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2334 struct preempted_vcore_list *lp;
2336 kvmppc_core_end_stolen(vc);
2337 if (!list_empty(&vc->preempt_list)) {
2338 lp = &per_cpu(preempted_vcores, vc->pcpu);
2339 spin_lock(&lp->lock);
2340 list_del_init(&vc->preempt_list);
2341 spin_unlock(&lp->lock);
2343 vc->vcore_state = VCORE_INACTIVE;
2347 * This stores information about the virtual cores currently
2348 * assigned to a physical core.
2352 int max_subcore_threads;
2354 int subcore_threads[MAX_SUBCORES];
2355 struct kvmppc_vcore *vc[MAX_SUBCORES];
2359 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2360 * respectively in 2-way micro-threading (split-core) mode.
2362 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2364 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2366 memset(cip, 0, sizeof(*cip));
2367 cip->n_subcores = 1;
2368 cip->max_subcore_threads = vc->num_threads;
2369 cip->total_threads = vc->num_threads;
2370 cip->subcore_threads[0] = vc->num_threads;
2374 static bool subcore_config_ok(int n_subcores, int n_threads)
2376 /* Can only dynamically split if unsplit to begin with */
2377 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2379 if (n_subcores > MAX_SUBCORES)
2381 if (n_subcores > 1) {
2382 if (!(dynamic_mt_modes & 2))
2384 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2388 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2391 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2393 vc->entry_exit_map = 0;
2395 vc->napping_threads = 0;
2396 vc->conferring_threads = 0;
2399 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2401 int n_threads = vc->num_threads;
2404 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2407 if (n_threads < cip->max_subcore_threads)
2408 n_threads = cip->max_subcore_threads;
2409 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2411 cip->max_subcore_threads = n_threads;
2413 sub = cip->n_subcores;
2415 cip->total_threads += vc->num_threads;
2416 cip->subcore_threads[sub] = vc->num_threads;
2418 init_vcore_to_run(vc);
2419 list_del_init(&vc->preempt_list);
2425 * Work out whether it is possible to piggyback the execution of
2426 * vcore *pvc onto the execution of the other vcores described in *cip.
2428 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2431 if (cip->total_threads + pvc->num_threads > target_threads)
2434 return can_dynamic_split(pvc, cip);
2437 static void prepare_threads(struct kvmppc_vcore *vc)
2440 struct kvm_vcpu *vcpu;
2442 for_each_runnable_thread(i, vcpu, vc) {
2443 if (signal_pending(vcpu->arch.run_task))
2444 vcpu->arch.ret = -EINTR;
2445 else if (vcpu->arch.vpa.update_pending ||
2446 vcpu->arch.slb_shadow.update_pending ||
2447 vcpu->arch.dtl.update_pending)
2448 vcpu->arch.ret = RESUME_GUEST;
2451 kvmppc_remove_runnable(vc, vcpu);
2452 wake_up(&vcpu->arch.cpu_run);
2456 static void collect_piggybacks(struct core_info *cip, int target_threads)
2458 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2459 struct kvmppc_vcore *pvc, *vcnext;
2461 spin_lock(&lp->lock);
2462 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2463 if (!spin_trylock(&pvc->lock))
2465 prepare_threads(pvc);
2466 if (!pvc->n_runnable) {
2467 list_del_init(&pvc->preempt_list);
2468 if (pvc->runner == NULL) {
2469 pvc->vcore_state = VCORE_INACTIVE;
2470 kvmppc_core_end_stolen(pvc);
2472 spin_unlock(&pvc->lock);
2475 if (!can_piggyback(pvc, cip, target_threads)) {
2476 spin_unlock(&pvc->lock);
2479 kvmppc_core_end_stolen(pvc);
2480 pvc->vcore_state = VCORE_PIGGYBACK;
2481 if (cip->total_threads >= target_threads)
2484 spin_unlock(&lp->lock);
2487 static bool recheck_signals(struct core_info *cip)
2490 struct kvm_vcpu *vcpu;
2492 for (sub = 0; sub < cip->n_subcores; ++sub)
2493 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2494 if (signal_pending(vcpu->arch.run_task))
2499 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2501 int still_running = 0, i;
2504 struct kvm_vcpu *vcpu;
2506 spin_lock(&vc->lock);
2508 for_each_runnable_thread(i, vcpu, vc) {
2509 /* cancel pending dec exception if dec is positive */
2510 if (now < vcpu->arch.dec_expires &&
2511 kvmppc_core_pending_dec(vcpu))
2512 kvmppc_core_dequeue_dec(vcpu);
2514 trace_kvm_guest_exit(vcpu);
2517 if (vcpu->arch.trap)
2518 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2519 vcpu->arch.run_task);
2521 vcpu->arch.ret = ret;
2522 vcpu->arch.trap = 0;
2524 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2525 if (vcpu->arch.pending_exceptions)
2526 kvmppc_core_prepare_to_enter(vcpu);
2527 if (vcpu->arch.ceded)
2528 kvmppc_set_timer(vcpu);
2532 kvmppc_remove_runnable(vc, vcpu);
2533 wake_up(&vcpu->arch.cpu_run);
2537 if (still_running > 0) {
2538 kvmppc_vcore_preempt(vc);
2539 } else if (vc->runner) {
2540 vc->vcore_state = VCORE_PREEMPT;
2541 kvmppc_core_start_stolen(vc);
2543 vc->vcore_state = VCORE_INACTIVE;
2545 if (vc->n_runnable > 0 && vc->runner == NULL) {
2546 /* make sure there's a candidate runner awake */
2548 vcpu = next_runnable_thread(vc, &i);
2549 wake_up(&vcpu->arch.cpu_run);
2552 spin_unlock(&vc->lock);
2556 * Clear core from the list of active host cores as we are about to
2557 * enter the guest. Only do this if it is the primary thread of the
2558 * core (not if a subcore) that is entering the guest.
2560 static inline int kvmppc_clear_host_core(unsigned int cpu)
2564 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2567 * Memory barrier can be omitted here as we will do a smp_wmb()
2568 * later in kvmppc_start_thread and we need ensure that state is
2569 * visible to other CPUs only after we enter guest.
2571 core = cpu >> threads_shift;
2572 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2577 * Advertise this core as an active host core since we exited the guest
2578 * Only need to do this if it is the primary thread of the core that is
2581 static inline int kvmppc_set_host_core(unsigned int cpu)
2585 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2589 * Memory barrier can be omitted here because we do a spin_unlock
2590 * immediately after this which provides the memory barrier.
2592 core = cpu >> threads_shift;
2593 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2597 static void set_irq_happened(int trap)
2600 case BOOK3S_INTERRUPT_EXTERNAL:
2601 local_paca->irq_happened |= PACA_IRQ_EE;
2603 case BOOK3S_INTERRUPT_H_DOORBELL:
2604 local_paca->irq_happened |= PACA_IRQ_DBELL;
2606 case BOOK3S_INTERRUPT_HMI:
2607 local_paca->irq_happened |= PACA_IRQ_HMI;
2609 case BOOK3S_INTERRUPT_SYSTEM_RESET:
2610 replay_system_reset();
2616 * Run a set of guest threads on a physical core.
2617 * Called with vc->lock held.
2619 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2621 struct kvm_vcpu *vcpu;
2624 struct core_info core_info;
2625 struct kvmppc_vcore *pvc;
2626 struct kvm_split_mode split_info, *sip;
2627 int split, subcore_size, active;
2630 unsigned long cmd_bit, stat_bit;
2633 int controlled_threads;
2637 * Remove from the list any threads that have a signal pending
2638 * or need a VPA update done
2640 prepare_threads(vc);
2642 /* if the runner is no longer runnable, let the caller pick a new one */
2643 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2649 init_vcore_to_run(vc);
2650 vc->preempt_tb = TB_NIL;
2653 * Number of threads that we will be controlling: the same as
2654 * the number of threads per subcore, except on POWER9,
2655 * where it's 1 because the threads are (mostly) independent.
2657 controlled_threads = threads_per_vcore();
2660 * Make sure we are running on primary threads, and that secondary
2661 * threads are offline. Also check if the number of threads in this
2662 * guest are greater than the current system threads per guest.
2664 if ((controlled_threads > 1) &&
2665 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2666 for_each_runnable_thread(i, vcpu, vc) {
2667 vcpu->arch.ret = -EBUSY;
2668 kvmppc_remove_runnable(vc, vcpu);
2669 wake_up(&vcpu->arch.cpu_run);
2675 * See if we could run any other vcores on the physical core
2676 * along with this one.
2678 init_core_info(&core_info, vc);
2679 pcpu = smp_processor_id();
2680 target_threads = controlled_threads;
2681 if (target_smt_mode && target_smt_mode < target_threads)
2682 target_threads = target_smt_mode;
2683 if (vc->num_threads < target_threads)
2684 collect_piggybacks(&core_info, target_threads);
2687 * On radix, arrange for TLB flushing if necessary.
2688 * This has to be done before disabling interrupts since
2689 * it uses smp_call_function().
2691 pcpu = smp_processor_id();
2692 if (kvm_is_radix(vc->kvm)) {
2693 for (sub = 0; sub < core_info.n_subcores; ++sub)
2694 for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2695 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2699 * Hard-disable interrupts, and check resched flag and signals.
2700 * If we need to reschedule or deliver a signal, clean up
2701 * and return without going into the guest(s).
2702 * If the hpte_setup_done flag has been cleared, don't go into the
2703 * guest because that means a HPT resize operation is in progress.
2705 local_irq_disable();
2707 if (lazy_irq_pending() || need_resched() ||
2708 recheck_signals(&core_info) ||
2709 (!kvm_is_radix(vc->kvm) && !vc->kvm->arch.hpte_setup_done)) {
2711 vc->vcore_state = VCORE_INACTIVE;
2712 /* Unlock all except the primary vcore */
2713 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2714 pvc = core_info.vc[sub];
2715 /* Put back on to the preempted vcores list */
2716 kvmppc_vcore_preempt(pvc);
2717 spin_unlock(&pvc->lock);
2719 for (i = 0; i < controlled_threads; ++i)
2720 kvmppc_release_hwthread(pcpu + i);
2724 kvmppc_clear_host_core(pcpu);
2726 /* Decide on micro-threading (split-core) mode */
2727 subcore_size = threads_per_subcore;
2728 cmd_bit = stat_bit = 0;
2729 split = core_info.n_subcores;
2732 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2733 if (split == 2 && (dynamic_mt_modes & 2)) {
2734 cmd_bit = HID0_POWER8_1TO2LPAR;
2735 stat_bit = HID0_POWER8_2LPARMODE;
2738 cmd_bit = HID0_POWER8_1TO4LPAR;
2739 stat_bit = HID0_POWER8_4LPARMODE;
2741 subcore_size = MAX_SMT_THREADS / split;
2743 memset(&split_info, 0, sizeof(split_info));
2744 split_info.rpr = mfspr(SPRN_RPR);
2745 split_info.pmmar = mfspr(SPRN_PMMAR);
2746 split_info.ldbar = mfspr(SPRN_LDBAR);
2747 split_info.subcore_size = subcore_size;
2748 for (sub = 0; sub < core_info.n_subcores; ++sub)
2749 split_info.vc[sub] = core_info.vc[sub];
2750 /* order writes to split_info before kvm_split_mode pointer */
2753 for (thr = 0; thr < controlled_threads; ++thr)
2754 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2756 /* Initiate micro-threading (split-core) if required */
2758 unsigned long hid0 = mfspr(SPRN_HID0);
2760 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2762 mtspr(SPRN_HID0, hid0);
2765 hid0 = mfspr(SPRN_HID0);
2766 if (hid0 & stat_bit)
2772 /* Start all the threads */
2774 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2775 thr = subcore_thread_map[sub];
2778 pvc = core_info.vc[sub];
2779 pvc->pcpu = pcpu + thr;
2780 for_each_runnable_thread(i, vcpu, pvc) {
2781 kvmppc_start_thread(vcpu, pvc);
2782 kvmppc_create_dtl_entry(vcpu, pvc);
2783 trace_kvm_guest_enter(vcpu);
2784 if (!vcpu->arch.ptid)
2786 active |= 1 << (thr + vcpu->arch.ptid);
2789 * We need to start the first thread of each subcore
2790 * even if it doesn't have a vcpu.
2793 kvmppc_start_thread(NULL, pvc);
2794 thr += pvc->num_threads;
2798 * Ensure that split_info.do_nap is set after setting
2799 * the vcore pointer in the PACA of the secondaries.
2803 split_info.do_nap = 1; /* ask secondaries to nap when done */
2806 * When doing micro-threading, poke the inactive threads as well.
2807 * This gets them to the nap instruction after kvm_do_nap,
2808 * which reduces the time taken to unsplit later.
2811 for (thr = 1; thr < threads_per_subcore; ++thr)
2812 if (!(active & (1 << thr)))
2813 kvmppc_ipi_thread(pcpu + thr);
2815 vc->vcore_state = VCORE_RUNNING;
2818 trace_kvmppc_run_core(vc, 0);
2820 for (sub = 0; sub < core_info.n_subcores; ++sub)
2821 spin_unlock(&core_info.vc[sub]->lock);
2824 * Interrupts will be enabled once we get into the guest,
2825 * so tell lockdep that we're about to enable interrupts.
2827 trace_hardirqs_on();
2831 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2833 trap = __kvmppc_vcore_entry();
2835 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2839 trace_hardirqs_off();
2840 set_irq_happened(trap);
2842 spin_lock(&vc->lock);
2843 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2844 vc->vcore_state = VCORE_EXITING;
2846 /* wait for secondary threads to finish writing their state to memory */
2847 kvmppc_wait_for_nap();
2849 /* Return to whole-core mode if we split the core earlier */
2851 unsigned long hid0 = mfspr(SPRN_HID0);
2852 unsigned long loops = 0;
2854 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2855 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2857 mtspr(SPRN_HID0, hid0);
2860 hid0 = mfspr(SPRN_HID0);
2861 if (!(hid0 & stat_bit))
2866 split_info.do_nap = 0;
2869 kvmppc_set_host_core(pcpu);
2873 /* Let secondaries go back to the offline loop */
2874 for (i = 0; i < controlled_threads; ++i) {
2875 kvmppc_release_hwthread(pcpu + i);
2876 if (sip && sip->napped[i])
2877 kvmppc_ipi_thread(pcpu + i);
2878 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2881 spin_unlock(&vc->lock);
2883 /* make sure updates to secondary vcpu structs are visible now */
2886 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2887 pvc = core_info.vc[sub];
2888 post_guest_process(pvc, pvc == vc);
2891 spin_lock(&vc->lock);
2895 vc->vcore_state = VCORE_INACTIVE;
2896 trace_kvmppc_run_core(vc, 1);
2900 * Wait for some other vcpu thread to execute us, and
2901 * wake us up when we need to handle something in the host.
2903 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2904 struct kvm_vcpu *vcpu, int wait_state)
2908 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2909 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2910 spin_unlock(&vc->lock);
2912 spin_lock(&vc->lock);
2914 finish_wait(&vcpu->arch.cpu_run, &wait);
2917 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2920 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2921 vc->halt_poll_ns = 10000;
2923 vc->halt_poll_ns *= halt_poll_ns_grow;
2926 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2928 if (halt_poll_ns_shrink == 0)
2929 vc->halt_poll_ns = 0;
2931 vc->halt_poll_ns /= halt_poll_ns_shrink;
2934 #ifdef CONFIG_KVM_XICS
2935 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2937 if (!xive_enabled())
2939 return vcpu->arch.xive_saved_state.pipr <
2940 vcpu->arch.xive_saved_state.cppr;
2943 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2947 #endif /* CONFIG_KVM_XICS */
2949 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
2951 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
2952 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
2959 * Check to see if any of the runnable vcpus on the vcore have pending
2960 * exceptions or are no longer ceded
2962 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2964 struct kvm_vcpu *vcpu;
2967 for_each_runnable_thread(i, vcpu, vc) {
2968 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
2976 * All the vcpus in this vcore are idle, so wait for a decrementer
2977 * or external interrupt to one of the vcpus. vc->lock is held.
2979 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2981 ktime_t cur, start_poll, start_wait;
2984 DECLARE_SWAITQUEUE(wait);
2986 /* Poll for pending exceptions and ceded state */
2987 cur = start_poll = ktime_get();
2988 if (vc->halt_poll_ns) {
2989 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2990 ++vc->runner->stat.halt_attempted_poll;
2992 vc->vcore_state = VCORE_POLLING;
2993 spin_unlock(&vc->lock);
2996 if (kvmppc_vcore_check_block(vc)) {
3001 } while (single_task_running() && ktime_before(cur, stop));
3003 spin_lock(&vc->lock);
3004 vc->vcore_state = VCORE_INACTIVE;
3007 ++vc->runner->stat.halt_successful_poll;
3012 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3014 if (kvmppc_vcore_check_block(vc)) {
3015 finish_swait(&vc->wq, &wait);
3017 /* If we polled, count this as a successful poll */
3018 if (vc->halt_poll_ns)
3019 ++vc->runner->stat.halt_successful_poll;
3023 start_wait = ktime_get();
3025 vc->vcore_state = VCORE_SLEEPING;
3026 trace_kvmppc_vcore_blocked(vc, 0);
3027 spin_unlock(&vc->lock);
3029 finish_swait(&vc->wq, &wait);
3030 spin_lock(&vc->lock);
3031 vc->vcore_state = VCORE_INACTIVE;
3032 trace_kvmppc_vcore_blocked(vc, 1);
3033 ++vc->runner->stat.halt_successful_wait;
3038 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3040 /* Attribute wait time */
3042 vc->runner->stat.halt_wait_ns +=
3043 ktime_to_ns(cur) - ktime_to_ns(start_wait);
3044 /* Attribute failed poll time */
3045 if (vc->halt_poll_ns)
3046 vc->runner->stat.halt_poll_fail_ns +=
3047 ktime_to_ns(start_wait) -
3048 ktime_to_ns(start_poll);
3050 /* Attribute successful poll time */
3051 if (vc->halt_poll_ns)
3052 vc->runner->stat.halt_poll_success_ns +=
3054 ktime_to_ns(start_poll);
3057 /* Adjust poll time */
3059 if (block_ns <= vc->halt_poll_ns)
3061 /* We slept and blocked for longer than the max halt time */
3062 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3063 shrink_halt_poll_ns(vc);
3064 /* We slept and our poll time is too small */
3065 else if (vc->halt_poll_ns < halt_poll_ns &&
3066 block_ns < halt_poll_ns)
3067 grow_halt_poll_ns(vc);
3068 if (vc->halt_poll_ns > halt_poll_ns)
3069 vc->halt_poll_ns = halt_poll_ns;
3071 vc->halt_poll_ns = 0;
3073 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3076 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3079 struct kvmppc_vcore *vc;
3082 trace_kvmppc_run_vcpu_enter(vcpu);
3084 kvm_run->exit_reason = 0;
3085 vcpu->arch.ret = RESUME_GUEST;
3086 vcpu->arch.trap = 0;
3087 kvmppc_update_vpas(vcpu);
3090 * Synchronize with other threads in this virtual core
3092 vc = vcpu->arch.vcore;
3093 spin_lock(&vc->lock);
3094 vcpu->arch.ceded = 0;
3095 vcpu->arch.run_task = current;
3096 vcpu->arch.kvm_run = kvm_run;
3097 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3098 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3099 vcpu->arch.busy_preempt = TB_NIL;
3100 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3104 * This happens the first time this is called for a vcpu.
3105 * If the vcore is already running, we may be able to start
3106 * this thread straight away and have it join in.
3108 if (!signal_pending(current)) {
3109 if (vc->vcore_state == VCORE_PIGGYBACK) {
3110 if (spin_trylock(&vc->lock)) {
3111 if (vc->vcore_state == VCORE_RUNNING &&
3112 !VCORE_IS_EXITING(vc)) {
3113 kvmppc_create_dtl_entry(vcpu, vc);
3114 kvmppc_start_thread(vcpu, vc);
3115 trace_kvm_guest_enter(vcpu);
3117 spin_unlock(&vc->lock);
3119 } else if (vc->vcore_state == VCORE_RUNNING &&
3120 !VCORE_IS_EXITING(vc)) {
3121 kvmppc_create_dtl_entry(vcpu, vc);
3122 kvmppc_start_thread(vcpu, vc);
3123 trace_kvm_guest_enter(vcpu);
3124 } else if (vc->vcore_state == VCORE_SLEEPING) {
3130 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3131 !signal_pending(current)) {
3132 /* See if the HPT and VRMA are ready to go */
3133 if (!kvm_is_radix(vcpu->kvm) &&
3134 !vcpu->kvm->arch.hpte_setup_done) {
3135 spin_unlock(&vc->lock);
3136 r = kvmppc_hv_setup_htab_rma(vcpu);
3137 spin_lock(&vc->lock);
3139 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3140 kvm_run->fail_entry.hardware_entry_failure_reason = 0;
3146 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3147 kvmppc_vcore_end_preempt(vc);
3149 if (vc->vcore_state != VCORE_INACTIVE) {
3150 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3153 for_each_runnable_thread(i, v, vc) {
3154 kvmppc_core_prepare_to_enter(v);
3155 if (signal_pending(v->arch.run_task)) {
3156 kvmppc_remove_runnable(vc, v);
3157 v->stat.signal_exits++;
3158 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3159 v->arch.ret = -EINTR;
3160 wake_up(&v->arch.cpu_run);
3163 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3166 for_each_runnable_thread(i, v, vc) {
3167 if (!kvmppc_vcpu_woken(v))
3168 n_ceded += v->arch.ceded;
3173 if (n_ceded == vc->n_runnable) {
3174 kvmppc_vcore_blocked(vc);
3175 } else if (need_resched()) {
3176 kvmppc_vcore_preempt(vc);
3177 /* Let something else run */
3178 cond_resched_lock(&vc->lock);
3179 if (vc->vcore_state == VCORE_PREEMPT)
3180 kvmppc_vcore_end_preempt(vc);
3182 kvmppc_run_core(vc);
3187 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3188 (vc->vcore_state == VCORE_RUNNING ||
3189 vc->vcore_state == VCORE_EXITING ||
3190 vc->vcore_state == VCORE_PIGGYBACK))
3191 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3193 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3194 kvmppc_vcore_end_preempt(vc);
3196 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3197 kvmppc_remove_runnable(vc, vcpu);
3198 vcpu->stat.signal_exits++;
3199 kvm_run->exit_reason = KVM_EXIT_INTR;
3200 vcpu->arch.ret = -EINTR;
3203 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3204 /* Wake up some vcpu to run the core */
3206 v = next_runnable_thread(vc, &i);
3207 wake_up(&v->arch.cpu_run);
3210 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3211 spin_unlock(&vc->lock);
3212 return vcpu->arch.ret;
3215 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3219 unsigned long ebb_regs[3] = {}; /* shut up GCC */
3220 unsigned long user_tar = 0;
3221 unsigned int user_vrsave;
3223 if (!vcpu->arch.sane) {
3224 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3229 * Don't allow entry with a suspended transaction, because
3230 * the guest entry/exit code will lose it.
3231 * If the guest has TM enabled, save away their TM-related SPRs
3232 * (they will get restored by the TM unavailable interrupt).
3234 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3235 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3236 (current->thread.regs->msr & MSR_TM)) {
3237 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3238 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3239 run->fail_entry.hardware_entry_failure_reason = 0;
3242 /* Enable TM so we can read the TM SPRs */
3243 mtmsr(mfmsr() | MSR_TM);
3244 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3245 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3246 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3247 current->thread.regs->msr &= ~MSR_TM;
3251 kvmppc_core_prepare_to_enter(vcpu);
3253 /* No need to go into the guest when all we'll do is come back out */
3254 if (signal_pending(current)) {
3255 run->exit_reason = KVM_EXIT_INTR;
3259 atomic_inc(&vcpu->kvm->arch.vcpus_running);
3260 /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
3263 flush_all_to_thread(current);
3265 /* Save userspace EBB and other register values */
3266 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3267 ebb_regs[0] = mfspr(SPRN_EBBHR);
3268 ebb_regs[1] = mfspr(SPRN_EBBRR);
3269 ebb_regs[2] = mfspr(SPRN_BESCR);
3270 user_tar = mfspr(SPRN_TAR);
3272 user_vrsave = mfspr(SPRN_VRSAVE);
3274 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3275 vcpu->arch.pgdir = current->mm->pgd;
3276 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3279 r = kvmppc_run_vcpu(run, vcpu);
3281 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3282 !(vcpu->arch.shregs.msr & MSR_PR)) {
3283 trace_kvm_hcall_enter(vcpu);
3284 r = kvmppc_pseries_do_hcall(vcpu);
3285 trace_kvm_hcall_exit(vcpu, r);
3286 kvmppc_core_prepare_to_enter(vcpu);
3287 } else if (r == RESUME_PAGE_FAULT) {
3288 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3289 r = kvmppc_book3s_hv_page_fault(run, vcpu,
3290 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3291 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3292 } else if (r == RESUME_PASSTHROUGH) {
3293 if (WARN_ON(xive_enabled()))
3296 r = kvmppc_xics_rm_complete(vcpu, 0);
3298 } while (is_kvmppc_resume_guest(r));
3300 /* Restore userspace EBB and other register values */
3301 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3302 mtspr(SPRN_EBBHR, ebb_regs[0]);
3303 mtspr(SPRN_EBBRR, ebb_regs[1]);
3304 mtspr(SPRN_BESCR, ebb_regs[2]);
3305 mtspr(SPRN_TAR, user_tar);
3306 mtspr(SPRN_FSCR, current->thread.fscr);
3308 mtspr(SPRN_VRSAVE, user_vrsave);
3310 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3311 atomic_dec(&vcpu->kvm->arch.vcpus_running);
3315 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3318 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
3322 (*sps)->page_shift = def->shift;
3323 (*sps)->slb_enc = def->sllp;
3324 (*sps)->enc[0].page_shift = def->shift;
3325 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
3327 * Add 16MB MPSS support if host supports it
3329 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
3330 (*sps)->enc[1].page_shift = 24;
3331 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
3336 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3337 struct kvm_ppc_smmu_info *info)
3339 struct kvm_ppc_one_seg_page_size *sps;
3342 * Since we don't yet support HPT guests on a radix host,
3343 * return an error if the host uses radix.
3345 if (radix_enabled())
3349 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3350 * POWER7 doesn't support keys for instruction accesses,
3351 * POWER8 and POWER9 do.
3353 info->data_keys = 32;
3354 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3356 info->flags = KVM_PPC_PAGE_SIZES_REAL;
3357 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
3358 info->flags |= KVM_PPC_1T_SEGMENTS;
3359 info->slb_size = mmu_slb_size;
3361 /* We only support these sizes for now, and no muti-size segments */
3362 sps = &info->sps[0];
3363 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
3364 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
3365 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
3371 * Get (and clear) the dirty memory log for a memory slot.
3373 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3374 struct kvm_dirty_log *log)
3376 struct kvm_memslots *slots;
3377 struct kvm_memory_slot *memslot;
3381 struct kvm_vcpu *vcpu;
3383 mutex_lock(&kvm->slots_lock);
3386 if (log->slot >= KVM_USER_MEM_SLOTS)
3389 slots = kvm_memslots(kvm);
3390 memslot = id_to_memslot(slots, log->slot);
3392 if (!memslot->dirty_bitmap)
3396 * Use second half of bitmap area because radix accumulates
3397 * bits in the first half.
3399 n = kvm_dirty_bitmap_bytes(memslot);
3400 buf = memslot->dirty_bitmap + n / sizeof(long);
3403 if (kvm_is_radix(kvm))
3404 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3406 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3410 /* Harvest dirty bits from VPA and DTL updates */
3411 /* Note: we never modify the SLB shadow buffer areas */
3412 kvm_for_each_vcpu(i, vcpu, kvm) {
3413 spin_lock(&vcpu->arch.vpa_update_lock);
3414 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3415 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3416 spin_unlock(&vcpu->arch.vpa_update_lock);
3420 if (copy_to_user(log->dirty_bitmap, buf, n))
3425 mutex_unlock(&kvm->slots_lock);
3429 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3430 struct kvm_memory_slot *dont)
3432 if (!dont || free->arch.rmap != dont->arch.rmap) {
3433 vfree(free->arch.rmap);
3434 free->arch.rmap = NULL;
3438 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3439 unsigned long npages)
3442 * For now, if radix_enabled() then we only support radix guests,
3443 * and in that case we don't need the rmap array.
3445 if (radix_enabled()) {
3446 slot->arch.rmap = NULL;
3450 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3451 if (!slot->arch.rmap)
3457 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3458 struct kvm_memory_slot *memslot,
3459 const struct kvm_userspace_memory_region *mem)
3464 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3465 const struct kvm_userspace_memory_region *mem,
3466 const struct kvm_memory_slot *old,
3467 const struct kvm_memory_slot *new)
3469 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3470 struct kvm_memslots *slots;
3471 struct kvm_memory_slot *memslot;
3474 * If we are making a new memslot, it might make
3475 * some address that was previously cached as emulated
3476 * MMIO be no longer emulated MMIO, so invalidate
3477 * all the caches of emulated MMIO translations.
3480 atomic64_inc(&kvm->arch.mmio_update);
3482 if (npages && old->npages && !kvm_is_radix(kvm)) {
3484 * If modifying a memslot, reset all the rmap dirty bits.
3485 * If this is a new memslot, we don't need to do anything
3486 * since the rmap array starts out as all zeroes,
3487 * i.e. no pages are dirty.
3489 slots = kvm_memslots(kvm);
3490 memslot = id_to_memslot(slots, mem->slot);
3491 kvmppc_hv_get_dirty_log_hpt(kvm, memslot, NULL);
3496 * Update LPCR values in kvm->arch and in vcores.
3497 * Caller must hold kvm->lock.
3499 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3504 if ((kvm->arch.lpcr & mask) == lpcr)
3507 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3509 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3510 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3513 spin_lock(&vc->lock);
3514 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3515 spin_unlock(&vc->lock);
3516 if (++cores_done >= kvm->arch.online_vcores)
3521 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3526 static void kvmppc_setup_partition_table(struct kvm *kvm)
3528 unsigned long dw0, dw1;
3530 if (!kvm_is_radix(kvm)) {
3531 /* PS field - page size for VRMA */
3532 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3533 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3534 /* HTABSIZE and HTABORG fields */
3535 dw0 |= kvm->arch.sdr1;
3537 /* Second dword as set by userspace */
3538 dw1 = kvm->arch.process_table;
3540 dw0 = PATB_HR | radix__get_tree_size() |
3541 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3542 dw1 = PATB_GR | kvm->arch.process_table;
3545 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3548 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3551 struct kvm *kvm = vcpu->kvm;
3553 struct kvm_memory_slot *memslot;
3554 struct vm_area_struct *vma;
3555 unsigned long lpcr = 0, senc;
3556 unsigned long psize, porder;
3559 mutex_lock(&kvm->lock);
3560 if (kvm->arch.hpte_setup_done)
3561 goto out; /* another vcpu beat us to it */
3563 /* Allocate hashed page table (if not done already) and reset it */
3564 if (!kvm->arch.hpt.virt) {
3565 int order = KVM_DEFAULT_HPT_ORDER;
3566 struct kvm_hpt_info info;
3568 err = kvmppc_allocate_hpt(&info, order);
3569 /* If we get here, it means userspace didn't specify a
3570 * size explicitly. So, try successively smaller
3571 * sizes if the default failed. */
3572 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3573 err = kvmppc_allocate_hpt(&info, order);
3576 pr_err("KVM: Couldn't alloc HPT\n");
3580 kvmppc_set_hpt(kvm, &info);
3583 /* Look up the memslot for guest physical address 0 */
3584 srcu_idx = srcu_read_lock(&kvm->srcu);
3585 memslot = gfn_to_memslot(kvm, 0);
3587 /* We must have some memory at 0 by now */
3589 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3592 /* Look up the VMA for the start of this memory slot */
3593 hva = memslot->userspace_addr;
3594 down_read(¤t->mm->mmap_sem);
3595 vma = find_vma(current->mm, hva);
3596 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3599 psize = vma_kernel_pagesize(vma);
3600 porder = __ilog2(psize);
3602 up_read(¤t->mm->mmap_sem);
3604 /* We can handle 4k, 64k or 16M pages in the VRMA */
3606 if (!(psize == 0x1000 || psize == 0x10000 ||
3607 psize == 0x1000000))
3610 senc = slb_pgsize_encoding(psize);
3611 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3612 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3613 /* Create HPTEs in the hash page table for the VRMA */
3614 kvmppc_map_vrma(vcpu, memslot, porder);
3616 /* Update VRMASD field in the LPCR */
3617 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3618 /* the -4 is to account for senc values starting at 0x10 */
3619 lpcr = senc << (LPCR_VRMASD_SH - 4);
3620 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3622 kvmppc_setup_partition_table(kvm);
3625 /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3627 kvm->arch.hpte_setup_done = 1;
3630 srcu_read_unlock(&kvm->srcu, srcu_idx);
3632 mutex_unlock(&kvm->lock);
3636 up_read(¤t->mm->mmap_sem);
3640 #ifdef CONFIG_KVM_XICS
3642 * Allocate a per-core structure for managing state about which cores are
3643 * running in the host versus the guest and for exchanging data between
3644 * real mode KVM and CPU running in the host.
3645 * This is only done for the first VM.
3646 * The allocated structure stays even if all VMs have stopped.
3647 * It is only freed when the kvm-hv module is unloaded.
3648 * It's OK for this routine to fail, we just don't support host
3649 * core operations like redirecting H_IPI wakeups.
3651 void kvmppc_alloc_host_rm_ops(void)
3653 struct kvmppc_host_rm_ops *ops;
3654 unsigned long l_ops;
3658 /* Not the first time here ? */
3659 if (kvmppc_host_rm_ops_hv != NULL)
3662 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3666 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3667 ops->rm_core = kzalloc(size, GFP_KERNEL);
3669 if (!ops->rm_core) {
3676 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3677 if (!cpu_online(cpu))
3680 core = cpu >> threads_shift;
3681 ops->rm_core[core].rm_state.in_host = 1;
3684 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3687 * Make the contents of the kvmppc_host_rm_ops structure visible
3688 * to other CPUs before we assign it to the global variable.
3689 * Do an atomic assignment (no locks used here), but if someone
3690 * beats us to it, just free our copy and return.
3693 l_ops = (unsigned long) ops;
3695 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3697 kfree(ops->rm_core);
3702 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3703 "ppc/kvm_book3s:prepare",
3704 kvmppc_set_host_core,
3705 kvmppc_clear_host_core);
3709 void kvmppc_free_host_rm_ops(void)
3711 if (kvmppc_host_rm_ops_hv) {
3712 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3713 kfree(kvmppc_host_rm_ops_hv->rm_core);
3714 kfree(kvmppc_host_rm_ops_hv);
3715 kvmppc_host_rm_ops_hv = NULL;
3720 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3722 unsigned long lpcr, lpid;
3726 /* Allocate the guest's logical partition ID */
3728 lpid = kvmppc_alloc_lpid();
3731 kvm->arch.lpid = lpid;
3733 kvmppc_alloc_host_rm_ops();
3736 * Since we don't flush the TLB when tearing down a VM,
3737 * and this lpid might have previously been used,
3738 * make sure we flush on each core before running the new VM.
3739 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3740 * does this flush for us.
3742 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3743 cpumask_setall(&kvm->arch.need_tlb_flush);
3745 /* Start out with the default set of hcalls enabled */
3746 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3747 sizeof(kvm->arch.enabled_hcalls));
3749 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3750 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3752 /* Init LPCR for virtual RMA mode */
3753 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3754 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3755 lpcr &= LPCR_PECE | LPCR_LPES;
3756 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3757 LPCR_VPM0 | LPCR_VPM1;
3758 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3759 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3760 /* On POWER8 turn on online bit to enable PURR/SPURR */
3761 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3764 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3765 * Set HVICE bit to enable hypervisor virtualization interrupts.
3766 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3767 * be unnecessary but better safe than sorry in case we re-enable
3768 * EE in HV mode with this LPCR still set)
3770 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3772 lpcr |= LPCR_HVICE | LPCR_HEIC;
3775 * If xive is enabled, we route 0x500 interrupts directly
3783 * For now, if the host uses radix, the guest must be radix.
3785 if (radix_enabled()) {
3786 kvm->arch.radix = 1;
3788 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3789 ret = kvmppc_init_vm_radix(kvm);
3791 kvmppc_free_lpid(kvm->arch.lpid);
3794 kvmppc_setup_partition_table(kvm);
3797 kvm->arch.lpcr = lpcr;
3799 /* Initialization for future HPT resizes */
3800 kvm->arch.resize_hpt = NULL;
3803 * Work out how many sets the TLB has, for the use of
3804 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3806 if (kvm_is_radix(kvm))
3807 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */
3808 else if (cpu_has_feature(CPU_FTR_ARCH_300))
3809 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
3810 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3811 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
3813 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
3816 * Track that we now have a HV mode VM active. This blocks secondary
3817 * CPU threads from coming online.
3818 * On POWER9, we only need to do this for HPT guests on a radix
3819 * host, which is not yet supported.
3821 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3822 kvm_hv_vm_activated();
3825 * Initialize smt_mode depending on processor.
3826 * POWER8 and earlier have to use "strict" threading, where
3827 * all vCPUs in a vcore have to run on the same (sub)core,
3828 * whereas on POWER9 the threads can each run a different
3831 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3832 kvm->arch.smt_mode = threads_per_subcore;
3834 kvm->arch.smt_mode = 1;
3835 kvm->arch.emul_smt_mode = 1;
3838 * Create a debugfs directory for the VM
3840 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3841 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3842 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3843 kvmppc_mmu_debugfs_init(kvm);
3848 static void kvmppc_free_vcores(struct kvm *kvm)
3852 for (i = 0; i < KVM_MAX_VCORES; ++i)
3853 kfree(kvm->arch.vcores[i]);
3854 kvm->arch.online_vcores = 0;
3857 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3859 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3861 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3862 kvm_hv_vm_deactivated();
3864 kvmppc_free_vcores(kvm);
3866 kvmppc_free_lpid(kvm->arch.lpid);
3868 if (kvm_is_radix(kvm))
3869 kvmppc_free_radix(kvm);
3871 kvmppc_free_hpt(&kvm->arch.hpt);
3873 kvmppc_free_pimap(kvm);
3876 /* We don't need to emulate any privileged instructions or dcbz */
3877 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3878 unsigned int inst, int *advance)
3880 return EMULATE_FAIL;
3883 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3886 return EMULATE_FAIL;
3889 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3892 return EMULATE_FAIL;
3895 static int kvmppc_core_check_processor_compat_hv(void)
3897 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3898 !cpu_has_feature(CPU_FTR_ARCH_206))
3904 #ifdef CONFIG_KVM_XICS
3906 void kvmppc_free_pimap(struct kvm *kvm)
3908 kfree(kvm->arch.pimap);
3911 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3913 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3916 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3918 struct irq_desc *desc;
3919 struct kvmppc_irq_map *irq_map;
3920 struct kvmppc_passthru_irqmap *pimap;
3921 struct irq_chip *chip;
3924 if (!kvm_irq_bypass)
3927 desc = irq_to_desc(host_irq);
3931 mutex_lock(&kvm->lock);
3933 pimap = kvm->arch.pimap;
3934 if (pimap == NULL) {
3935 /* First call, allocate structure to hold IRQ map */
3936 pimap = kvmppc_alloc_pimap();
3937 if (pimap == NULL) {
3938 mutex_unlock(&kvm->lock);
3941 kvm->arch.pimap = pimap;
3945 * For now, we only support interrupts for which the EOI operation
3946 * is an OPAL call followed by a write to XIRR, since that's
3947 * what our real-mode EOI code does, or a XIVE interrupt
3949 chip = irq_data_get_irq_chip(&desc->irq_data);
3950 if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
3951 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3952 host_irq, guest_gsi);
3953 mutex_unlock(&kvm->lock);
3958 * See if we already have an entry for this guest IRQ number.
3959 * If it's mapped to a hardware IRQ number, that's an error,
3960 * otherwise re-use this entry.
3962 for (i = 0; i < pimap->n_mapped; i++) {
3963 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3964 if (pimap->mapped[i].r_hwirq) {
3965 mutex_unlock(&kvm->lock);
3972 if (i == KVMPPC_PIRQ_MAPPED) {
3973 mutex_unlock(&kvm->lock);
3974 return -EAGAIN; /* table is full */
3977 irq_map = &pimap->mapped[i];
3979 irq_map->v_hwirq = guest_gsi;
3980 irq_map->desc = desc;
3983 * Order the above two stores before the next to serialize with
3984 * the KVM real mode handler.
3987 irq_map->r_hwirq = desc->irq_data.hwirq;
3989 if (i == pimap->n_mapped)
3993 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
3995 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3997 irq_map->r_hwirq = 0;
3999 mutex_unlock(&kvm->lock);
4004 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4006 struct irq_desc *desc;
4007 struct kvmppc_passthru_irqmap *pimap;
4010 if (!kvm_irq_bypass)
4013 desc = irq_to_desc(host_irq);
4017 mutex_lock(&kvm->lock);
4018 if (!kvm->arch.pimap)
4021 pimap = kvm->arch.pimap;
4023 for (i = 0; i < pimap->n_mapped; i++) {
4024 if (guest_gsi == pimap->mapped[i].v_hwirq)
4028 if (i == pimap->n_mapped) {
4029 mutex_unlock(&kvm->lock);
4034 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4036 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4038 /* invalidate the entry (what do do on error from the above ?) */
4039 pimap->mapped[i].r_hwirq = 0;
4042 * We don't free this structure even when the count goes to
4043 * zero. The structure is freed when we destroy the VM.
4046 mutex_unlock(&kvm->lock);
4050 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4051 struct irq_bypass_producer *prod)
4054 struct kvm_kernel_irqfd *irqfd =
4055 container_of(cons, struct kvm_kernel_irqfd, consumer);
4057 irqfd->producer = prod;
4059 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4061 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4062 prod->irq, irqfd->gsi, ret);
4067 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4068 struct irq_bypass_producer *prod)
4071 struct kvm_kernel_irqfd *irqfd =
4072 container_of(cons, struct kvm_kernel_irqfd, consumer);
4074 irqfd->producer = NULL;
4077 * When producer of consumer is unregistered, we change back to
4078 * default external interrupt handling mode - KVM real mode
4079 * will switch back to host.
4081 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4083 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4084 prod->irq, irqfd->gsi, ret);
4088 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4089 unsigned int ioctl, unsigned long arg)
4091 struct kvm *kvm __maybe_unused = filp->private_data;
4092 void __user *argp = (void __user *)arg;
4097 case KVM_PPC_ALLOCATE_HTAB: {
4101 if (get_user(htab_order, (u32 __user *)argp))
4103 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4110 case KVM_PPC_GET_HTAB_FD: {
4111 struct kvm_get_htab_fd ghf;
4114 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4116 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4120 case KVM_PPC_RESIZE_HPT_PREPARE: {
4121 struct kvm_ppc_resize_hpt rhpt;
4124 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4127 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4131 case KVM_PPC_RESIZE_HPT_COMMIT: {
4132 struct kvm_ppc_resize_hpt rhpt;
4135 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4138 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4150 * List of hcall numbers to enable by default.
4151 * For compatibility with old userspace, we enable by default
4152 * all hcalls that were implemented before the hcall-enabling
4153 * facility was added. Note this list should not include H_RTAS.
4155 static unsigned int default_hcall_list[] = {
4169 #ifdef CONFIG_KVM_XICS
4180 static void init_default_hcalls(void)
4185 for (i = 0; default_hcall_list[i]; ++i) {
4186 hcall = default_hcall_list[i];
4187 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4188 __set_bit(hcall / 4, default_enabled_hcalls);
4192 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4197 /* If not on a POWER9, reject it */
4198 if (!cpu_has_feature(CPU_FTR_ARCH_300))
4201 /* If any unknown flags set, reject it */
4202 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4205 /* We can't change a guest to/from radix yet */
4206 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4207 if (radix != kvm_is_radix(kvm))
4210 /* GR (guest radix) bit in process_table field must match */
4211 if (!!(cfg->process_table & PATB_GR) != radix)
4214 /* Process table size field must be reasonable, i.e. <= 24 */
4215 if ((cfg->process_table & PRTS_MASK) > 24)
4218 mutex_lock(&kvm->lock);
4219 kvm->arch.process_table = cfg->process_table;
4220 kvmppc_setup_partition_table(kvm);
4222 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4223 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4224 mutex_unlock(&kvm->lock);
4229 static struct kvmppc_ops kvm_ops_hv = {
4230 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4231 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4232 .get_one_reg = kvmppc_get_one_reg_hv,
4233 .set_one_reg = kvmppc_set_one_reg_hv,
4234 .vcpu_load = kvmppc_core_vcpu_load_hv,
4235 .vcpu_put = kvmppc_core_vcpu_put_hv,
4236 .set_msr = kvmppc_set_msr_hv,
4237 .vcpu_run = kvmppc_vcpu_run_hv,
4238 .vcpu_create = kvmppc_core_vcpu_create_hv,
4239 .vcpu_free = kvmppc_core_vcpu_free_hv,
4240 .check_requests = kvmppc_core_check_requests_hv,
4241 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
4242 .flush_memslot = kvmppc_core_flush_memslot_hv,
4243 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4244 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
4245 .unmap_hva = kvm_unmap_hva_hv,
4246 .unmap_hva_range = kvm_unmap_hva_range_hv,
4247 .age_hva = kvm_age_hva_hv,
4248 .test_age_hva = kvm_test_age_hva_hv,
4249 .set_spte_hva = kvm_set_spte_hva_hv,
4250 .mmu_destroy = kvmppc_mmu_destroy_hv,
4251 .free_memslot = kvmppc_core_free_memslot_hv,
4252 .create_memslot = kvmppc_core_create_memslot_hv,
4253 .init_vm = kvmppc_core_init_vm_hv,
4254 .destroy_vm = kvmppc_core_destroy_vm_hv,
4255 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4256 .emulate_op = kvmppc_core_emulate_op_hv,
4257 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4258 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4259 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4260 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
4261 .hcall_implemented = kvmppc_hcall_impl_hv,
4262 #ifdef CONFIG_KVM_XICS
4263 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4264 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4266 .configure_mmu = kvmhv_configure_mmu,
4267 .get_rmmu_info = kvmhv_get_rmmu_info,
4268 .set_smt_mode = kvmhv_set_smt_mode,
4271 static int kvm_init_subcore_bitmap(void)
4274 int nr_cores = cpu_nr_cores();
4275 struct sibling_subcore_state *sibling_subcore_state;
4277 for (i = 0; i < nr_cores; i++) {
4278 int first_cpu = i * threads_per_core;
4279 int node = cpu_to_node(first_cpu);
4281 /* Ignore if it is already allocated. */
4282 if (paca[first_cpu].sibling_subcore_state)
4285 sibling_subcore_state =
4286 kmalloc_node(sizeof(struct sibling_subcore_state),
4288 if (!sibling_subcore_state)
4291 memset(sibling_subcore_state, 0,
4292 sizeof(struct sibling_subcore_state));
4294 for (j = 0; j < threads_per_core; j++) {
4295 int cpu = first_cpu + j;
4297 paca[cpu].sibling_subcore_state = sibling_subcore_state;
4303 static int kvmppc_radix_possible(void)
4305 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4308 static int kvmppc_book3s_init_hv(void)
4312 * FIXME!! Do we need to check on all cpus ?
4314 r = kvmppc_core_check_processor_compat_hv();
4318 r = kvm_init_subcore_bitmap();
4323 * We need a way of accessing the XICS interrupt controller,
4324 * either directly, via paca[cpu].kvm_hstate.xics_phys, or
4325 * indirectly, via OPAL.
4328 if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4329 struct device_node *np;
4331 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4333 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4339 kvm_ops_hv.owner = THIS_MODULE;
4340 kvmppc_hv_ops = &kvm_ops_hv;
4342 init_default_hcalls();
4346 r = kvmppc_mmu_hv_init();
4350 if (kvmppc_radix_possible())
4351 r = kvmppc_radix_init();
4355 static void kvmppc_book3s_exit_hv(void)
4357 kvmppc_free_host_rm_ops();
4358 if (kvmppc_radix_possible())
4359 kvmppc_radix_exit();
4360 kvmppc_hv_ops = NULL;
4363 module_init(kvmppc_book3s_init_hv);
4364 module_exit(kvmppc_book3s_exit_hv);
4365 MODULE_LICENSE("GPL");
4366 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4367 MODULE_ALIAS("devname:kvm");